Garry Tan
When everyone is building the same primitives it does speak to needs that will only intensify from here And we will eventually converge on the correct OS
中文: 当每个人都在构建相同的原语时，它确实会满足需求，而这种需求只会从这里开始加剧 我们最终会在正确的操作系统上达成一致
#729 🔗 原推
2026-10-05 00:15:43
Garry Tan
RT @blingdivinity: the post agi future is infinite undergrad for everyone forever. study, socialize, drink, party, hobbies, clubs, arts, sports, dating, fucking, status games college is already economically useless except self development and gaining experiences. thats what the future will be for
中文: RT @blingdivinity:未来是每个人永远无休无可读的本科。 学习、社交、饮酒、派对、爱好、俱乐部、艺术、体育、约会、他妈的、状态游戏 大学在经济上已经毫无用处，除了自我发展和积累经验之外。
#728 🔗 原推
2026-10-04 03:15:11
Garry Tan
RT @tszzl: this chart should be the death knell for the 2010s tech left whose theology compelled them to argue that Uber was some sort of labor law arbitrage that created no productivity benefits. the truth is that it created a far larger market than taxis ever had. the industrialists win
中文: RT @tszzl:这张图表本应成为2010年代科技行业的死钟，这些科技行业的神学认为优步是某种劳工法套利，不会带来任何生产力效益。事实上，它创造了一个比出租车规模大得多的市场。这些实业家赢得了胜利
#727 🔗 原推
2026-10-04 02:54:17
Garry Tan
RT @Twolfrecovery: "Chesa is one of the most principled, ethical, and approachable elected officials I've ever had the pleasure of working with." -Gary McCoy. Absolutely disqualifying. D8 can and must do better. #SanFrancisco https://twitter.com/Twolfrecovery/status/2106569459913867447/photo/1
中文: RT @Twolfrecovery:“Chesa是我有幸合作过的最有原则、最有道德且最平易近人的民选官员之一。”——加里·麦科伊 绝对取消。D8 能够而且必须做得更好。#旧金山
media 0 共 1 项
#726 🔗 原推
2026-10-04 02:29:41
Garry Tan
RT @robkhenderson: Interesting: https://rbaumeisterexistentialcontrarian.substack.com/p/freud-in-the-rearview-mirror https://twitter.com/robkhenderson/status/2106420213864395220/photo/1
中文: RT @robkhenderson:有趣:
media 0 共 1 项
#725 🔗 原推
2026-10-04 02:29:08
Garry Tan
RT @garryslist: California is full of people whose families gave up everything to build a better life here. Garry's List is where they come together to protect the California Dream. Every name tells a story. Add yours with a public donation to Garry's List Action: https://secure.actblue.com/donate/californiadream https://twitter.com/garryslist/status/2106429376866697325/video/1
中文: RT @garryslist:加利福尼亚州充满了家庭为在这里创造更美好生活而放弃一切的人。加里的清单是他们共同保护加州梦的地方。 每个名字都讲述着一个故事。将您的捐款与加里的清单行动分享至:
🎬
视频
#724 🔗 原推
2026-10-04 01:31:40
Garry Tan
RT @ycombinator: Congrats to @kiwicopple, @AntWilson, and @supabase on their $150 million round at a $10.65 billion valuation! More than 13 million developers use Supabase as the backend for their apps. They're adding more than 1 million users and over 4 million databases every month, and 70% of new databases are now created by agents or AI-driven tools. https://www.nytimes.com/2026/10/02/business/dealbook/data-center-ai-bonds.html
中文: RT @ycombinator:以10.65亿美元的估值向@kiwicopple、@AntWilson和@supabase表示祝贺! 超过1300万名开发者使用Supabase作为其应用程序的后端。他们每月增加超过100万用户和超过400万个数据库，而现在70%的新数据库是由代理或人工智能驱动的工具创建的。
media 0 共 1 项
#723 🔗 原推
2026-10-03 20:22:48
Garry Tan
RT @kane: SF supervisor candidate @garymccoysf was an exec at an NGO embezzling tax dollars who tried to sue people for retweeting a meme about him. He’s also a convicted felon who supports drug use on SF sidewalks. Absolutely do not vote for him. https://twitter.com/kane/status/2106427082054209629/photo/1
media 0 共 1 项
#722 🔗 原推
2026-10-03 19:07:20
Garry Tan
Just deleted about a thousand lines of markdown from gstack because frontier models are good now and the old tricks generally you don’t need anymore https://twitter.com/garrytan/status/2106298398441951497/photo/1
中文: 刚刚从 gstack 上删除了大约一千行标记，因为边框现在不错，而且通常不需要的旧花招
media 0 共 1 项
#721 🔗 原推
2026-10-03 08:20:45
Garry Tan
RT @ycombinator: This week’s Paper Club is all about alternative compute. Modern AI has been shaped by a tight coupling between transformers, backpropagation, and GPUs. But as compute efficiency starts to plateau, there may be very different ways to build the machines that come next. We explore what AI compute could look like beyond conventional digital hardware through optical systems that use light to perform computation, neuromorphic architectures inspired by the brain, and experiments teaching living brain cells to play Doom. 00:07 — @FrancoisChauba1: Why Alternative Compute? 03:38 — Beyond Backpropagation 08:08 — Zero-Order Optimization and SOMA 15:06 — Ilker Oguz: Computing With Light 19:27 — Why Optical Computing Isn’t Everywhere Yet 22:23 — Building a Diffusion Model With Light 30:03 — Optical Computing Q&A 39:00 — @AlokVasudev: What Is Neuromorphic Computing? 45:23 — What the Brain Can Teach Us About Chips 49:35 — Where Neuromorphic Computing Stands Today 54:24 — Neuromorphic Computing Q&A 1:04:24 — @vytalow: Teaching Brain Cells to Play Doom 1:07:48 — Training Biological Neurons With Reinforcement Learning 1:12:28 — How Brain Cells Learn From Feedback 1:15:40 — Biological Computing Q&A
🎬
视频
#720 🔗 原推
2026-10-03 08:18:07
Garry Tan
RT @robkhenderson: "Since American children...weren’t being taught what happened under actual socialist regimes, it was only a matter of time before simplistic slogans attacking private property, billionaires and 'profits before people,' would be successfully revived" https://www.wsj.com/opinion/the-scholar-who-saw-mamdani-coming-in-2003-socialism-politics-c38e41d8?reflink=desktopwebshare_permalink
中文: RT @robkhenderson:“由于美国儿童在实际社会主义政权下没有被教导，因此在攻击私有财产、亿万富翁和“人民之前就盈利”的简单化口号成功恢复之后，只是时间问题
#719 🔗 原推
2026-10-03 08:16:35
Garry Tan
RT @lina_colucci: It's mind blowing to me that we can give people real-time video at the same price as real-time voice 🤯
中文: RT @lina_colucci:让我大发热的是，我们能以与实时语音相同的价格为人们提供实时视频🤯
#718 🔗 原推
2026-10-03 08:16:23
Garry Tan
RT @TheCoppinger: 60% of the latest YC batch are using @supabase 🤯 also, woz hype intensifies 😂 https://twitter.com/TheCoppinger/status/2106069497019056209/photo/1
中文: RT @TheCoppinger:最新YC批次中有60%使用 @supabase 🤯 也，炒作加剧😂
media 0 共 2 项 media 1
#717 🔗 原推
2026-10-02 23:38:23
Garry Tan
RT @jakemintz: I misunderstood this when it launched. This is the first / best true agent<>human collaboration space I have seen. You are sharing a whole computer with AI and can collaborate on anything. When writing a doc the agent sees the comments, you can have a side text/voice chat, you can have it navigate or you navigate while you talk. It's really impressive and the largest step forward in AI UI that I have seen. It is not just adding a sidebar AI chat to productivity apps, it's much more seamless, comprehensive, and natural than that. I'm struggling to articulate because it looks similar but is very different. Thank you @LudvigSwanstrom. Everyone, check it out.
中文: RT @jakemintz:我在它发布时误解了这一点。这是我见过的第一个/最优秀的真实协作空间。你正在与人工智能共享一台完整的计算机，并且可以在任何事物上进行协作。编写文档时，代理会看到评论，可以进行侧边文字/语音聊天，或在通话时进行导航。 这确实令人印象深刻，也是我所见过的人工智能用户界面领域最大的进步。它不仅仅是在生产力应用中添加侧边栏的AI聊天，它比这要顺畅、全面和自然得多。我难以表达，因为它看起来相似，却截然不同。 谢谢@LudvigSwanstrom。大家，去看看。
#716 🔗 原推
2026-10-02 23:28:47
Garry Tan
RT @SusanDReynolds: Voters should never forget that Gary McCoy, a longtime executive at harm reduction grift machine HealthRight 360, is a proven liar and a bully: He fudged the numbers at the Linkage Center and lied about successes that never happened. It was nothing more than a drug den, which McCoy supports. Full story: https://susanreynolds.substack.com/p/the-grift-that-keeps-on-giving?utm_source=publication-search He wasted nearly 3 years tying up the courts trying to get the identity of a gay icon, ruining his life over a Twitter meme. McCoy also asked a judge to force me to give up my source, and tried to serve @EricaJSandberg in the lawsuit for simply sharing the meme. Full story: https://susanreynolds.substack.com/p/a-gay-icon-is-living-in-fear-because?utm_source=publication-search He was part of London Breed’s “troll patrol” — a group of bullies who called @AngelaAlioto1 supporters “Nazis” and said Alioto was “one white hood away from being in the KKK.” Full story: https://www.reynoldsrap.org/london-breeds-troll-patrol/ He was president of the Milk Club where the same group of people involved in the Manny accusations accused Shahid Buttar of sexual assault when he ran against McCoy’s boss Nancy Pelosi (claims that I later proved were untrue). Full story: https://www.marinatimes.com/smear-campaign-against-shahid-buttar-smacks-of-racism McCoy is a proven liar and a bully. He was still far behind Manny. Timing of the latest story is suspect — ballots are just going out — as is the author Joe Fitz, a known far left operative. Considering how McCoy lied about the Linkage Center, his involvement in this spectacle is entirely plausible. His reputation precedes him — as do the receipts found in my stories linked above. Keep all of this in mind when you go to the polls.
#715 🔗 原推
2026-10-02 19:22:31
Garry Tan
RT @trycua: 1/ We're reimagining what it means for your agents to work with all your computers. Today we're excited to share Cua Spaces, built on Cua Driver and our virtualization stack, rolling out on macOS today. Download it now, it's free and source-available: https://spaces.cua.ai/
中文: RT @trycua:1/ 我们正在重新构想您的代理人员使用您所有计算机的意义。 今天，我们很期待与基于 Cua Driver 和我们的虚拟化栈的 Cua Spaces 分享，今天将在 macOS 上推出。 立即下载，免费且可获取源代码:
🎬
视频
#714 🔗 原推
2026-10-02 19:22:07
Garry Tan
Capy is like a 24/7 standup for your agents
中文: Capy 就像为你的经纪人提供全天候的站立
#713 🔗 原推
2026-10-02 19:20:30
Garry Tan
Surprising cool @capydotai cross-session coordination I wasn't expecting: My GBrain collaborator Sina started a fix wave that is even steering around all of the things I'm already working on across multiple Capy threads. All model agnostic. https://twitter.com/garrytan/status/2106092289173213444/photo/1
中文: 令人意外的酷炫@capydotai 交叉会话协调，我原本没料到:我的 GBRAin 合作者 Sina 掀起了一股修复浪潮，甚至在多个 Capy 线程中围绕我已处理的所有内容进行引导。所有模型不可知。
media 0 共 1 项
#712 🔗 原推
2026-10-02 18:41:45
Garry Tan
RT @VirtualElena: when we think about ai in the physical world, we usually default to thinking about humanoids or self-driving cars or or models helping scientists discover new drugs and materials. but there’s another version of physical-world ai that i think deserves more attention: using intelligence to orchestrate all the benign-seeming things that have to happen for something to get built. which is especially important because the physical world is this wildly chaotic unpredictable place. what happens when the physical world refuses to cooperate. your shipment falls off a ship. or a strait closes. where, exactly, can an agent intervene? this is the problem we get into in my conversation with @askvladi, cofounder of lio, and @seema_amble. lio builds ai agents for procurement, in other words the work of figuring out what a company needs to buy, finding suppliers, negotiating terms, and making sure the right things arrive. when you’re building an aircraft or a data center, those decisions connect engineering, logistics, and finance in very consequential ways. vlad’s answer to the question of how to wrangle the physical world was about probability. suppliers have different records of reliability and routes have different risks. understanding those things in concert changes what you should buy / what you should pay: a cheap component that holds up an entire project can become a very expensive component. this all requires a much richer understanding of the world than the price and delivery date sitting in an enterprise database. i think this is a very compelling application of ai. a lot of our ability to build more ambitious things depends on making thousands of these decisions well. robots justifiably get a lot of attention; getting everything required to build them in the same place at the right time is just as important!
🎬
视频
#711 🔗 原推
2026-10-02 18:40:55
Garry Tan
My goal with GBrain: I want our agents to feel as native, expressive, and inevitable as the web eventually did—not as a chatbot feature, but as a personal Jiminy Cricket that knows you and helps you 24/7 always on, always thinking ahead, always improving.
中文: 我与GBrain合作的目标: 我希望我们的经纪人能像网络一样，拥有原生、富有表现力且不可避免的感觉——这并非作为聊天机器人的功能，而是作为一个个人吉米尼板球，他始终了解你，并全天候帮助你，始终在思考，不断进步。
#710 🔗 原推
2026-10-02 16:48:45
Garry Tan
RT @bradflora: YC gets founders into the room where it happens.
中文: RT @bradflora:YC 让创始人进入这个事情发生的房间。
#709 🔗 原推
2026-10-02 15:48:10
Garry Tan
True 5 years ago True today
中文: 真实5年前 今天是真的
#708 🔗 原推
2026-10-02 15:47:42
Garry Tan
RT @Twolfrecovery: This is likely your new D8 Supervisor San Francisco. He wants to decriminalize drugs and use your tax dollars to give free drugs to all the homeless drug users in the city. My God. @garrytan https://twitter.com/Twolfrecovery/status/2105813296318455857/photo/1
中文: RT @Twolfrecovery:这可能是您在旧金山的新D8主管。他希望将毒品非刑罪化，并用你的税款向全市所有无家可归的吸毒者免费提供毒品。我的天啊。@garrytan
media 0 共 1 项
#707 🔗 原推
2026-10-02 14:17:55
Garry Tan
RT @sdianahu: more benchmarking towards long term horizon tasks that feed into self recursive improving loops for sw and one of them is getting agents do successfully oncall for production code that they may be shipping themselves
中文: RT @sdianahu:针对长期前景任务进行更多基准测试，这些任务可生成自我递归改进的循环 其中一项是让代理商成功获取生产代码，而这些代码可能正在自行发货
#706 🔗 原推
2026-10-02 13:54:15
Garry Tan
RT @war24182236: Pay attention closely and hear me out Connie Chan is unfit to serve in ANY office. She would be a GENERATIONAL FAILURE in Congress. Look no further than how she treats residents in her own neighborhood. https://twitter.com/war24182236/status/2105834323215863949/video/1
中文: RT @war24182236:请仔细注意，并听我说康妮·陈不适合在任何办公室任职。她将成为国会中的一代失败者。再看看她对待自己社区居民的方式就不了。
🎬
视频
#705 🔗 原推
2026-10-02 01:38:39
Garry Tan
RT @ycombinator: Startup School is coming to Taipei on November 12! 🇹🇼 Hear from founders like @andyfang of DoorDash, @jeffchanginvest of Vest, @moseslo of Xendit, @mchiang0610 of Ollama, and more. And join builders and hackers from across Asia for a day of talks and sessions with YC partners. https://twitter.com/ycombinator/status/2105826448707932647/video/1
中文: RT @ycombinator:初创企业学校将于11月12日访问台北!🇹🇼 请先从DoorDash的@andyfang、@jeffchanginvest(来自Vest)、@moseslo(Xendit)、@mchiang0610(来自Ollama)等公司等公司听到。 与来自亚洲各地的建筑商和黑客一起，与YC合作伙伴进行一天的会谈和会谈。
🎬
视频
#704 🔗 原推
2026-10-02 01:10:32
Garry Tan
RT @ycombinator: Startup School is coming to Taipei on November 12! 🇹🇼 Hear from @andyfang of DoorDash, @jeffchanginvest of Vest, @moseslo of Xendit, @mchiang0610 of Ollama, and more. Join the best builders &amp; hackers from across Taiwan and Asia for a day of talks and sessions with YC partners. https://twitter.com/ycombinator/status/2105817542325580001/video/1
中文: RT @ycombinator:初创企业学校将于11月12日访问台北!🇹🇼 来自DoorDash的@andyfang、Vest的@jeffchanginvest、Xendit的@moseslo、Ollama的@mchiang0610等等。 加入来自台湾和亚洲地区最优秀的建筑商和黑客，与YC合作伙伴进行一天的会谈和会谈。
🎬
视频
#703 🔗 原推
2026-10-02 00:53:12
Garry Tan
When people say McCoy is not an extremist you can tell those people are gaslighting you
中文: 当人们说麦科伊不是极端分子时，你就能分辨出那些人在为你加油
#702 🔗 原推
2026-10-02 00:16:28
Garry Tan
RT @hanlonbt: You can just fix things. Prof. Elmendorf explains how @cayimby -sponsored bills (all of them!) work to remove CEQA as a barrier to urban housing.
中文: RT @hanlonbt:你可以解决问题。 教授埃尔门多夫解释了@cayimby 赞助的账单(全部全部!)努力消除CEQA作为城市住房的障碍。
#701 🔗 原推
2026-10-01 23:57:23
Garry Tan
This guy is now the frontrunner for D8 Supervisor in SF. You should probably be looking for alternatives. https://www.dailymail.com/news/article-10698085/amp/Taxpayer-funded-boss-San-Franciscos-open-air-drugs-market-accused-fudging-figures.html https://twitter.com/garrytan/status/2105809148365717985/photo/1
中文: 这家伙现在是旧金山D8主管的领跑者。 你可能应该寻找其他选择。
media 0 共 1 项
#700 🔗 原推
2026-10-01 23:56:39
Garry Tan
RT @zealcaiden: “make a techno optimist music video full of hype moments and insane motion graphics” https://twitter.com/zealcaiden/status/2105705296597524558/video/1
中文: RT @zealcaiden:“制作一段充满炒作瞬间和疯狂动态画面的科技乐观音乐视频”
🎬
视频
#699 🔗 原推
2026-10-01 23:37:28
Garry Tan
Unbelievable that the likely depressing result of this November election in SF is this guy will be my local supervisor
中文: 令人难以置信的是，今年11月在旧金山举行的选举可能令人沮丧的结果，就是这家伙会是我当地的主管
#698 🔗 原推
2026-10-01 23:35:34
Garry Tan
RT @kane: &gt; Progressive NGO &gt; “we’ll fix homelessness!” &gt; get $250M from @sfgov &gt; embezzle money &gt; commit crimes &gt; increase homelessness &gt; shut down
中文: RT @kane: &gt; 进步非政府组织 “我们来解决无家可归问题!” 从@sfgov获得2.5亿美元 加价;贪污钱财 犯罪 增加无家可归现象 关闭
#697 🔗 原推
2026-10-01 23:27:38
Garry Tan
RT @sdianahu: AI passing the Turing test 2022 was chat -&gt; 2023 was voice -&gt; 2026 is video with Tavus!!
中文: RT @sdianahu:人工智能通过了图灵测试 2022年是聊天 -&gt; 2023 是语音 -&gt; 2026 是与 Tavus 合作的视频!
#696 🔗 原推
2026-10-01 23:09:22
Garry Tan
Harm accelerationist Gary McCoy is going to ruin San Francisco
中文: 伤害加速者加里·麦考伊将毁掉旧金山
#695 🔗 原推
2026-10-01 23:05:06
Garry Tan
Pretty wild the woman in this video is a realtime generated agent
中文: 这个视频中的女人非常疯狂，是一个实时生成的代理
#694 🔗 原推
2026-10-01 19:04:15
Garry Tan
RT @vivianmshen: I’m joining @ycombinator as a General Partner. YC is a homecoming for every GP. All of us were previously YC founders, and lived the journey that we support founders through every day. Our alma mater taught us the lessons that we pass on. My founder path was infinitely harder than I thought it would be. We hit incredible highs and brutal lows. We came within a few months of zero runway. The YC mantras kept me going: make something people want, and don’t die. Over the course of my companies, we touched thousands of lives: employing people in almost every US state (Alaska still evades me!), and teaching STEM to students from every walk of life. Now, I’ve had the privilege of working with founders who are tackling some of the hardest problems today, including defense, space, energy, and manufacturing. We’re also entering an era of diffusing AI across businesses of all sizes, and founders are bringing these tools into their old workplaces and new customers to make work better. Finally, I’m particularly interested in our work at YC to partner closely with the government to make sure technology serves everyone. Our Startup Industrial Base conference yesterday was the start of this, with more to come. Founders are the backbone of American innovation and core to the future of our country and economy. Thank you to the YC partnership for believing in me, and the YC founders for trusting me. There’s never been a better time to make something people want.
中文: RT @vivianmshen:我将以普通合伙人的身份加入 @ycombinator。 YC 是每位全科医生的归宿。我们所有人都曾是YC的创始人，并一直过着我们每天支持创始人的旅程。我们的母校教会了我们传授的教训。 我的创始人之路比我想象的要艰难得多。我们达到了惊人的高点和惨淡的低谷。我们在零跑道的几个月内就来了。YC的口号让我坚持:做一件人们想要的东西，不要死。在我的公司过程中，我们影响了成千上万条生活:几乎在美国每个州雇佣员工(阿拉斯加仍然逃避我!)，并向各行各业的学生教授STEM。 现在，我有幸与那些正在应对当今最棘手问题(包括国防、航天、能源和制造业)的创始人合作。我们也正进入一个在各种规模企业中传播人工智能的时代，创始人们正将这些工具带入他们的旧工作场所和新客户，以改善工作。 最后，我特别关心我们在YC的工作，以便与政府紧密合作，确保技术服务于每个人。昨天我们的初创企业产业基地大会是开端，未来将有更多会议。创始人是美国创新的支柱，也是我国国家和经济未来的核心。 感谢YC合作伙伴对我的信任，以及YC创始人对我的信任。 创造人们想要的东西从未如此美好。
media 0 共 1 项
#693 🔗 原推
2026-10-01 19:02:03
Garry Tan
RT @garryslist: Three strangers on my red-eye spent the whole flight talking about California. Housing. Families who need room. Friends who left for more space and a yard. Eventually the conversation got heavier, a little more helpless. Then one said: "That's the thing, what are we gonna do?"
中文: RT @garryslist:我红眼的三个陌生人整班都在谈论加利福尼亚。住房。需要空间的家庭。为了腾出更多空间和一个院子的朋友。 最终，谈话变得更加沉重，更加无助。然后一个人说:“事情就是这样，我们该怎么办?”
#692 🔗 原推
2026-10-01 18:51:13
Garry Tan
Modus built an open-source test of whether AI can do actual junior-auditor work, not just answer accounting questions. Every job in the world will end up having eval benchmarks like this.
中文: Modus 构建了一个开源测试，测试人工智能是否能够完成真正的初级审计工作，而不仅仅是回答会计问题。 世界上每一份工作最终都会有这样的评估基准。
#691 🔗 原推
2026-10-01 18:41:17
Garry Tan
YC has the largest partnership in history now - 19 including me. We all went through YC and are alums and founders ourselves. I'm so proud to welcome Vivian and Raphael to the partnership.
中文: YC 拥有有史以来规模最大的合作伙伴关系——包括我在内的19个。我们都经历过YC，都是校友和创始人。 我非常自豪地欢迎薇薇安和拉斐尔加入这个搭档。
#690 🔗 原推
2026-10-01 18:37:53
Garry Tan
RT @raphaelschaad: Life update: dropping the "Visiting" — I'm joining @ycombinator as a General Partner 🤩 YC was my first choice as a founder in 2019. Seven years later, I'm choosing it again. This past year, I got to come full circle: working with founders through the W26 and S26 batches, helping them turn early ideas into products people love. I'm incredibly excited to keep finding founders and helping them build the iconic companies of this generation 🔥
中文: RT @raphaelschaad:生活更新:取消“访问”——我将加入 @ycombinator 担任普通合伙人 🤩 2019年，YC是我作为创始人的首选。七年后，我又选择了它。 过去一年，我开始全面合作:通过W26和S26系列产品与创始人合作，帮助他们将早期想法转化为人们喜爱的产品。 我非常兴奋能够继续寻找创始人，帮助他们打造这一代的标志性公司🔥
media 0 共 2 项 media 1
#689 🔗 原推
2026-10-01 18:36:53
Garry Tan
https://cfo.ai/ is the truth - if you want to know what the heck your company is spending on and if it is driving to ROI, you owe it to yourself to set this finance agent up for yourself The integrations out of box mean you're up and running almost instantly
中文: 是事实——如果你想了解公司所花费的资金，以及是否正驶向投资回报率(ROI)，你有责任为自己设立这家财务代理 集成的机身意味着你几乎立即启动并运行
#688 🔗 原推
2026-10-01 18:36:30
Garry Tan
GBrain now supports Jev. Retrieval is unchanged but remembering/dream cycle works better. https://twitter.com/garrytan/status/2105701930714685695/photo/1
中文: GBrain 现在支持 Jev。检索没有改变，但记忆/梦境循环效果更好。
media 0 共 1 项
#687 🔗 原推
2026-10-01 16:50:36
Garry Tan
RT @justinsunyt: Introducing Capy Desktop: the best way to orchestrate coding agents across all your devices Remote control and access local files/shells from any machine Close your laptop and Capy continues in the cloud Now available for MacOS, Windows, and Linux Comment for $100 in credits https://twitter.com/justinsunyt/status/2105692713085710388/video/1
中文: RT @justinsunyt:介绍Capy Desktop:在所有设备上编排编码代理的最佳方式 远程控制，并从任何机器访问本地文件/外壳 关闭笔记本电脑，继续在云端购买 现已可用于 MacOS、Windows 和 Linux 评论100美元积分
🎬
视频
#686 🔗 原推
2026-10-01 16:17:40
Garry Tan
RT @ycombinator: Halluminate is building reinforcement learning environments and benchmarks to help AI models do knowledge work beyond coding, starting with finance. With a team of fewer than 10 people, the company works with four of the top five closed-source US AI labs and recently raised a $30M Series A. In this fireside, co-founders @Jerr_Wu and @wgm752 sit down with YC's @xuster to share how testing browser agents led them to building the environments models now use to practice tasks like financial modeling. They explain why better verification, not just more training tasks, is critical to making models smarter, and how that work could eventually lead to simulated companies where teams of AI agents learn to work together. 00:00 — Halluminate’s Pivot From Evals to Training Environments 03:48 — Teaching AI to Do Financial Work 07:13 — Why Verification Matters More Than Volume 10:13 — From Finance to Simulated Companies 13:30 — AI Safety and Building the Team
中文: RT @ycombinator:Halluminate 正在构建强化学习环境和基准，以帮助人工智能模型在编程之外从事知识工作，从财务开始。 该公司拥有不到10人的团队，与美国五大闭源人工智能实验室中的四家合作，并近期筹集了3000万美元的A轮融资。 在这场火灾中，联合创始人@Jerr_Wu和@wgm752与YC的@xuster合作，分享测试浏览器代理如何促使他们构建现在用于实践财务建模等任务的模型环境。他们解释了为何更好的验证(而不仅仅是更多的训练任务)对于使模型变得更智能至关重要，以及这项工作最终如何导致模拟公司，让人工智能代理团队学会协同工作。 00:00 — 从椭圆形到训练环境的Halluminate's Pivot 03:48 — 教人工智能进行财务工作 07:13——为何验证比数量更重要 10:13 — 从金融到模拟公司 13:30 — 人工智能安全与团队建设
🎬
视频
#685 🔗 原推
2026-10-01 15:47:57
Garry Tan
RT @ishaan_jaff: Today we’re launching LiteLLM Lens. @LiteLLM is already the chokepoint for 100% of your enterprise’s AI traffic. We believe the next era of the gateway is using the data flowing through it to help your agents improve. We’re building for a future where developers run agentic swarms generating 200K+ traces. We’re focused on two things ⬇️ h/t @Mokhalil01 @yujonglee
中文: RT @ishaan_jaff:今天我们推出 LiteLLM 镜头。 @LiteLLM 已经是企业 100% 人工智能流量的瓶颈。我们相信，网关的下一个时代是利用流经它的数据来帮助您的代理改进。 我们正在为未来构建一个开发者能够运行产生20万多个痕迹的特化群体。 我们专注于两件事⬇️ @Mokhalil01 @yujonglee
#684 🔗 原推
2026-10-01 13:24:00
Garry Tan
RT @vivianmshen: Incredible day at the @ycombinator startup industrial base conference. It was an honor to speak to Senator @TimSheehyMT about his experience as a Navy SEAL officer, founder, and now senator - and how Congress can support founders. So many YC founders working on mission critical companies. Excited for the future here!
中文: RT @vivianmshen:在 @ycombinator 初创企业产业基地大会上，今天精彩一日。 很荣幸能与参议员@TimSheehyMT谈论他作为海军海豹突击队军官、创始人以及现任参议员的经历，以及国会如何支持创始人。 许多YC创始人从事关键任务公司的工作。对未来的到来感到兴奋!
media 0 共 1 项
#683 🔗 原推
2026-10-01 13:11:42
Garry Tan
RT @mernit: The most valuable advice we got during YC was to explain our company using verbs instead of nouns. Early on, I walked into a pitch and did the complete opposite: “We’re reinventing the cloud for AI” No one had any clue what that meant. I could (very painfully) see their eyes glazing over Then I started saying this instead: “Developers bring us their code and we run it on servers in the cloud so they don’t have to manage their own infra” This clicked immediately. People understand verbs - not labels. And no one buys or invests in things they don’t understand.
中文: RT @mernit:在YC期间我们得到的最有价值的建议是用动词而非名词来解释我们公司。 早期，我走进一个球场，做了完全相反的对立面: 我们正在为人工智能重新发明云计算 没人知道那意味着什么。我能(非常痛苦地)看到他们的眼睛瞪着 然后我开始说这个: 开发者会为我们带来他们的代码，我们在云端的服务器上运行，因此他们不必管理自己的基础设施。 立即点击了。 人们理解动词——而不是标签。 没有人购买或投资他们不理解的东西。
#682 🔗 原推
2026-10-01 01:58:29
Garry Tan
*chefs kiss*
中文: 厨师亲吻*
#681 🔗 原推
2026-10-01 01:38:01
Garry Tan
RT @MissionLoco: Every single person in the 4 left columns of endorsements has been thrown out of office by San Francisco voters. Jajajajajaja — It’s like getting an endorsement for your Ponzi scheme from Bernie Madoff, George Santos, and the board of Enron.
中文: RT @MissionLoco:旧金山选民已将四支左翼支持专栏中的每一位人士赶下台。贾贾贾贾贾贾——这就像从伯尼·麦道夫、乔治·桑托斯以及安然董事会那里获得你的庞氏骗局的代言。
#680 🔗 原推
2026-10-01 00:12:42
Garry Tan
Other code editors let you top down coordinate child agents I’ve never really seen this where agent swarms speak to one another as peers and self coordinate allowing you to organically explore the space instead of pre-arrange it, have you? Capy is special
中文: 其他代码编辑器可让您自上而下地协调子代理 我从未见过这种物质群作为同伴相互对话并自我协调，让你能够自然地探索空间，而不是预先安排它，你有吗? Capy 很特别
#679 🔗 原推
2026-10-01 00:08:49
Garry Tan
I asked Secretary Duffy what Americans should be able to do, if we get this decade right, that they can't do today. Sean Duffy is the Secretary of Transportation, and he sat down with me at our Startup Industrial Base event in DC today. He went straight to how our country's air traffic gets managed. At the command center, people sit in front of six screens with a paper flight chart above them, trying to make the calls on weather and routing. His example: four planes reach the same runway at the same time and only one can land. See it two hours out, speed up two, slow one down, and each one lands when the runway is open. "Like, not complicated stuff, but we don't do any of this." Paper and pencil and minimal human augmentation from software looks like this: Say an airport can take 15 planes in a 15 minute window. "Well, the airlines will schedule 32." He says he's handing the airlines that data and asking them to spread out. The tool is SMART, from Air Space Intelligence, a small company he says competed with the biggest companies and won. It predicts. His stage example: a 2.5 hour ground stop might only need 45 minutes. He's taken heat for supposedly handing air traffic control to AI. Duffy's answer? "it's dumb AI, not smart AI." Controllers still make every call. None of this needs a breakthrough. It needs the will of the people to know they need to use software and be ok with AI augmentation instead of insisting on a paper chart and copper wire.
🎬
视频
#678 🔗 原推
2026-09-30 20:34:47
Garry Tan
The Pentagon's 300-page RFP "was designed for a lawyer and a lobbyist and a prime contractor to know how to fill out." That's Emil Michael, Under Secretary of War for Research & Engineering, saying it out loud to a room full of founders at our Startup Industrial Base event in DC today. A 10-person startup reads 300 pages, gets one thing wrong, gets kicked out. Or survives and grinds through 3, 6, 9, 12 month rounds until it's two years in. Meanwhile the prime has a team whose whole job is filling out every page perfectly. Emil’s fix? Write a one pager not 300. "I need to fly a missile 500 miles in this kind of environment." If your payload won't work, you hear that fast. Target is 90 days. Drone Dominance runs the same way. Apply, top 20 go to a fly-off, objective metrics, cuts every round, and he says half a billion dollars of orders at the end. "I'd rather know if I lose and why." Miss round one and there are more rounds coming, and "it'll be 2 or 3" winners, not one. Cost per intercept is the game now. Shoot down a $50k drone with a Patriot missile and "the math doesn't math." He's funding directed energy and high power microwave that cost thousands, not millions. The tech was never what kept startups out. The 300 pages were. Two pages and a fly-off means the best product can actually win, and the prime has to compete like everyone else.
中文: 五角大楼300页的RFP“是为律师、游说人员以及主要承包商设计的，旨在了解如何填写。” 这是《战争研究与工程部副部长》的艾米尔·迈克尔，今天在华盛顿特区的初创企业产业基地活动上，向满是创始人的房间大声道。 一家10人的初创公司阅读了300页，弄错了一件事，被赶了出去。或者生存并磨练3、6、9、12个月，直到两年过。与此同时，首相拥有一支团队，其整个工作完全完整地填满了每一页。 埃米尔的修复?写一个单页，而不是300。我需要在这种环境下驾驶一枚500英里的导弹。如果你的有效载荷无法工作，你就会很快听到那声音。目标为90天。 无人机的统治方式也是如此。申请前20名将采用客观的客观指标，每轮都进行削减，他称到最后将完成5亿美元的订单。如果输了，我宁愿知道为什么。错过第一轮，还有更多的回合即将到来，“会是2到3个冠军，而不是一个。” 每次拦截的费用就是现在的游戏。用爱国者导弹击落一架5万美元的无人机，而“数学并不算数”。他资助的是能源和高功率微波，耗资数千，而不是数百万。 这项技术从未让初创企业脱颖而出。300页是。两页和一次飞逝意味着最优秀的产品能够真正获胜，而巅峰者必须像其他人一样竞争。
🎬
视频
#677 🔗 原推
2026-09-30 19:09:30
Garry Tan
RT @lucas0choa: "It's instinct but actually for work" for Instinct &amp; Muse, work and compliance was an afterthought; that's why we built Ace. Ace has its own computer, email, phone number, and logins. You onboard it like a new hire and it just gets work done. https://twitter.com/lucas0choa/status/2105342164658204705/video/1
中文: RT @lucas0choa:“这是本能，但实际上是为了工作” 对Instinct &amp; Muse而言，工作与合规是事后考虑;这就是我们打造Ace的原因。 Ace 拥有自己的电脑、电子邮件、电话号码和登录信息。你像新员工一样上岗，而且工作就能完成。
🎬
视频
#676 🔗 原推
2026-09-30 17:53:49
Garry Tan
https://twitter.com/garrytan/status/2105353588079759469/photo/1
#675 🔗 原推
2026-09-30 17:46:25
Garry Tan
RT @ycombinator: We're hosting an evening in SF with @oakhcft and π for early-stage builders in physical AI: robotics, autonomy, and intelligent hardware. Join us for a fireside chat with @physical_int co-founder @QuanVng on what it takes to build in the physical world. https://events.ycombinator.com/yc-oak-happyhour
中文: RT @ycombinator:我们将在旧金山举办一个晚上活动，与@oakhcft合作，为物理人工智能(如机器人、自主性和智能硬件)的早期构建者提供服务。 加入我们，与@physical_int联合创始人@QuanVng进行一场炉边聊天，讨论在现实世界中建立所需的内容。
#674 🔗 原推
2026-09-30 15:18:49
Garry Tan
@mtrainier2020 It wasn’t me
中文: @mtrainier2020 不是我
#673 🔗 原推
2026-09-30 15:16:25
Garry Tan
Full schedule
中文: 完整时间表
#672 🔗 原推
2026-09-30 14:01:53
Garry Tan
Follow along today. We are live streaming from DC on how startups are helping build the future of defense tech in America
中文: 今天就跟着走。我们正在从华盛顿特区进行直播，以实时直播初创企业如何助力美国国防技术的发展
#671 🔗 原推
2026-09-30 13:19:43
Garry Tan
It’s amazing you can just tell Capy coding agents to coordinate and reduce stomping on the same code paths and it just automatically injects agent to agent convos and it works without further orchestration https://twitter.com/garrytan/status/2105263275604689025/photo/1
中文: 真是太神奇了，你可以告诉Capy编码代理能够协调并减少对相同代码路径的踩踏，它会自动为代理convos注入代理，无需进一步编排即可
media 0 共 1 项
#670 🔗 原推
2026-09-30 11:47:33
Garry Tan
RT @mattyglesias: People have heard about the phonics vs triple-queuing fiasco but there is a similar issue in math where ed school ideologues have stigmatized the normal and necessary work of drilling novice students in the fundamentals. https://www.slowboring.com/p/math-needs-a-phonics-style-reckoning
#669 🔗 原推
2026-09-30 11:17:08
Garry Tan
RT @wallstengine: Nvidia CEO Jensen Huang on AI model distillation: “That’s called competition. You’re allowed to test somebody else’s products all you want.” https://twitter.com/wallstengine/status/2104604118937735553/video/1
中文: RT @wallstengine:英伟达首席执行官黄仁勋在AI模型蒸馏: 这叫做竞争。你可以完全测试别人的产品。”
🎬
视频
#668 🔗 原推
2026-09-30 05:17:16
Garry Tan
RT @stanine: The fastest way to achieve career growth is to join a company that's growing so quickly that promoting you into an impossible scope is a better option than hiring from the outside. Then, all you have to do is the impossible.
中文: RT @stanine:实现职业发展的最快方式是加入一家发展迅速的公司，而将你提升到不可能的领域，比从外部招聘更为好。然后，你要做的就是不可能。
#667 🔗 原推
2026-09-30 04:58:53
Garry Tan
RT @chrispisarski: every YC founder I'm chatting with is trying to figure out AEO and the amount of DMs I got from unicorn founders just off the last tweet is insane this is how I would do it: 1) figure out a) the exact prompts your ICP types into ChatGPT and Claude and b) the sources the AI pulls its answers from for a), you can build an AEO tracker yourself without spending $$$$$ on agencies and tools (they're still worth it if you'd rather save the time) if you want to self-host it, this is what I'd do: get Elmo (open source, github/elmohq/elmo) it runs your prompts through the actual chatgpt, claude, Gemini, Perplexity, AI Mode and AI Overviews interfaces via DataForSEO (don't do it via API) and shows you which prompts you win vs lose and how you rank you can run the same set across countries benefit of owning this is that you can be really creative with how you track the keywords / what to optimize for, since you have full control over everything we also store every answer and feed it back in as context b) we spent a ton of time reverse engineering our demo funnel whenever a demo came in referred by AI (Claude, ChatGPT), we offered free credits in exchange for an /export of the chat that sent them to crustdata then we analyzed every prompt / source in those chats 2) once you have the dashboard and the list of prompts + sources, it's just execution all of your copy across product, website, docs and socials should use these exact words and tell the same story over and over just do it all. we talked to a lot of AEO agencies and the amount of dashboards and tools they built that contradict each other (some say LinkedIn is the play, others YouTube, others review sites, others your product pages) tells you nobody has figured out how it *really* works what works is doing all of it, tracking the results and asking the people who signed up through AI, until you have enough of your own data to double down on what performs we did that for a while until it really *worked* and we knew which channels to optimize for the community part helped a ton everyone is trying to figure this out, so if you learn something, share it and grow together etc. we're having our next AEO meetup this Saturday, sign-up link in the comments!
#666 🔗 原推
2026-09-30 04:58:41
Garry Tan
RT @mattdorsey: Crime-fighting technologies like ALPRs are too important to risk abuses that threaten civil liberties or erode public confidence. I strongly support approaches that prevent and punish misuse — without abandoning their use. Thank you, @DanielLurie, @BrookeJenkinsSF, @MattHaneySF, and @SFPDChiefLew.
中文: RT @mattdorsey:像ALPR这样的打击犯罪技术过于重要，不得冒威胁公民自由或削弱公众信心的滥用风险。我强烈支持那些能够防止和惩治滥用行为的方法——但不会放弃滥用。 谢谢，@DanielLurie，@BrookeJenkinsSF，@MattHaneySF，@SFPDChiefLew。
#665 🔗 原推
2026-09-30 04:58:19
Garry Tan
RT @koomen: We're co-hosting YC Cyber Night with @cotoolai on 10/15, with some great panelists from the security sector. RSVP soon, space is limited. https://events.ycombinator.com/yc-cyber-night
#664 🔗 原推
2026-09-30 04:58:15
Garry Tan
Just merged a version of OpenClaw’s test-audit skill into GStack. Test bloat is a real problem. Luckily intelligence is on tap these days! https://twitter.com/garrytan/status/2105049005147525231/photo/1
中文: 刚刚将 OpenClaw 的测试审计技能合并到 GStack 中。测试膨胀是一个真正的问题。 幸运的是，如今情报已经开启!
media 0 共 1 项
#663 🔗 原推
2026-09-29 21:36:07
Garry Tan
RT @Suhail: I think a lot of the fear that AI has on the world can be replaced with thinking more ambitiously. Instead of being a software engineer, you can think of yourself as a Creator of Open Worlds. It will still be advantageous to be technical but also creative simultaneously.
中文: RT @Suhail:我认为，人工智能对世界的许多恐惧，却可以被更具雄心的思维所取代。与其做软件工程师，不如将自己视为开放世界的创造者。同时具有技术性，但同时也具有创造性，仍然具有优势。
#662 🔗 原推
2026-09-29 21:32:18
Garry Tan
RT @browser_use: Browser Use Ultrafast available 🔓 &gt; 10x faster and cheaper &gt; compare flights for $0.004 &gt; in &lt;20s Web agents are now fast, accurate, and cheap. DM to skip the waitlist or comment why you want access 👀
中文: RT @browser_use:浏览器使用 Ultrafast 可用 🔓 速度快10倍，更便宜 价格:&gt;比较机票价格为0.004美元 在 &lt;20s 网络代理现在速度快、准确且成本低廉。 DM 跳过等待列表或说明为何要访问 👀
#661 🔗 原推
2026-09-29 21:23:25
Garry Tan
RT @MHurabiell: This is the appropriate approach. Let's not throw the baby out with the bathwater. Well done @DanielLurie https://www.sfchronicle.com/sf/article/flock-cameras-lurie-haney-22453554.php
中文: RT @MHurabiell:这是合适的方法。 我们不要把婴儿和洗澡水一起扔出去。 干得好 @DanielLurie
#660 🔗 原推
2026-09-29 21:12:20
Garry Tan
RT @gbrainio: Just rolled out an upgrade to the hosted GBrain fleet. OpenClaw, GBrain, and Node were bumped up to the latest versions. ✌️ https://gbrain.io/notes/openclaw-2026-9-6-and-gbrain-0-59-3
中文: RT @gbrainio:刚刚推出了对托管的GBrain机队的升级。OpenClaw、GBrain 和 Node 被升级到了最新版本。 ✌️
#659 🔗 原推
2026-09-29 21:01:11
Garry Tan
RT @gbrainio: We got to walk and talk with GBrain's open-source maintainer, @fromsinaimportx, during the @ycombinator Own Your Intelligence hackathon. Watch if you want to know what's changed in GBrain since July and what's next for the project. https://twitter.com/gbrainio/status/2105009466479632398/video/1
中文: RT @gbrainio:在 @ycombinator Own Your Intelligence 黑客马拉松期间，我们与 GBrain 的开源维护者 @fromsinaimportx 进行了交谈。 请注意自7月以来GBRain的变化以及项目接下来会发生什么。
🎬
视频
#658 🔗 原推
2026-09-29 19:53:32
Garry Tan
RT @simonw: I'm at OpenAI's DevDay event in San Francisco today - as I have for the past three DevDay events, I'm running a live blog where I'll be posting updates during the keynote, which starts in five minutes https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/
中文: RT @simonw:今天我参加了位于旧金山的OpenAI的DevDay活动——就像过去三场DevDay活动一样，我正在运营一个实时博客，在主题演讲中发布最新动态，主题演讲将从五分钟开始
#657 🔗 原推
2026-09-29 17:49:08
Garry Tan
RT @slash1sol: Garry Tan, the CEO of Y Combinator, open-sourced 55 ready-made agents for Claude Code Each agent is a slash command in plain Markdown with its own role and rules. 134k stars, MIT. What is inside: > 7 for browser work and scraping > 6 for product planning > 6 for design > 6 for memory and retros > 5 for code review and debugging > 5 for security and guardrails > 5 for QA and performance > 5 for shipping and deploy > 5 for iOS apps > 3 for docs They hand work to each other. /office-hours writes a design doc, the plan reviewers read it, /qa and /ship check against it. Clone it into ~/.claude/skills/gstack, run ./setup, then try /review on any branch.
中文: RT @slash1sol:Y Combinator 的首席执行官加里·谭为 Claude Code 开源了 55 个现成的代理 每个代理都是普通Markdown中的斜线指令，具有其自身的角色和规则。134k颗恒星，麻省理工学院。 里面有什么: 7 用于浏览器操作和抓取 产品规划6 设计6 记忆与复古的6 代码审查和调试5 安全护栏和护栏5 5 用于质量保证和性能 5 用于运输和部署 5 个适用于 iOS 应用 文档3 他们互相处理工作。/office-houris 会编写一个设计文档，计划审核人员会阅读，并 /qa 进行检查。 将其克隆成 ~/.claude/skills/gstack，运行 ./setup，然后尝试在任何分支上进行 /review。
media 0 共 1 项
#656 🔗 原推
2026-09-29 17:48:07
Garry Tan
RT @gregpr07: Introducing: Browser Use Ultrafast, now on our cloud ⚡ 10x faster and 10x cheaper. > cheapest af > extremely fast > compare flights for $0.0041 Need more brain power? Also introducing BU Fast (smarter). Compute is a bit limited, so we're letting people in batch by batch. DM me to jump the line 🔓 (1x speed btw)
中文: RT @gregpr07:介绍:浏览器使用 Ultrafast，现在已上架云 ⚡ 10倍更快，便宜10倍。 > 最便宜的 速度极快 比价0.0041美元 需要更多的脑力吗?还引入了BU Fast(更智能)。 计算有点有限，所以我们允许人们批量进行。让我跳线 🔓 (速度快)
🎬
视频
#655 🔗 原推
2026-09-29 17:26:34
Garry Tan
RT @benspringwater: I can't tell you how excited and happy this makes me. Confusing, hard-to-use government software imposes a massive time tax, especially on those who can least afford it (see @AnnieLowrey's work). What if government services were as clear, well-designed, and easy to navigate as Airbnb or Stripe? Hats off to @jgebbia and his team for this exceptionally important work. https://www.theatlantic.com/podcasts/2026/09/ra-the-time-tax/688503/
中文: RT @benspringwater:我无法告诉你这让我多么兴奋和开心。令人困惑且难以使用的政府软件会征收巨额的时间税，尤其是那些负担不起的人(参见@AnnieLowrey 的工作)。如果政府服务像Airbnb或Stripe一样清晰、设计精良且易于导航，会怎样?向@jgebbia及其团队致敬，感谢他们从事这项极其重要的工作。
media 0 共 2 项 media 1
#654 🔗 原推
2026-09-29 17:05:28
Garry Tan
The people who created the SF doom spiral are back for doom spiral 2
中文: 制造科幻“末形”螺旋的人，重新回到了末旋。
#653 🔗 原推
2026-09-29 17:04:07
Garry Tan
This is crazy to see agents just do clean handoff with no additional work on my part https://twitter.com/garrytan/status/2104964759724736624/photo/1
中文: 看到代理人员在没有额外工作的情况下进行彻底的交给，这简直太疯狂了
media 0 共 1 项
#652 🔗 原推
2026-09-29 16:01:21
Garry Tan
Imagine saying this in 2026 https://twitter.com/garrytan/status/2104962980286996588/photo/1
中文: 想象一下2026年说这句话
#651 🔗 原推
2026-09-29 15:54:17
Garry Tan
RT @dhh: "Slop" has become a comfort blanket for a cohort of programmers stuck somewhere between anger and bargaining on their way to acceptance. There might be some kicking and screaming, maybe a little crying, but eventually you have to let the blanket go and update your priors.
#650 🔗 原推
2026-09-29 15:53:06
Garry Tan
Jev and Respan are showing us small tool models can be a big unlock alongside the frontier intelligent models
中文: 杰夫和雷斯潘向我们展示了小型工具模型，与前沿智能模型并驾齐驱
#649 🔗 原推
2026-09-29 15:50:06
Garry Tan
Smart browsers like @AsideAI that make sure agents can do things any human would be able to do in a browser are a real and growing category Aside’s password management is top notch to support exactly this
中文: 像@AsideAI这样的智能浏览器，确保代理能够完成任何人类在浏览器中能够完成的操作，这一类别确实在不断增长 Aside 的密码管理功能完全支持此功能
#648 🔗 原推
2026-09-29 15:44:48
Garry Tan
RT @hanghuang_: We just raised an $8M seed round to kill AWS, GCP, and Azure. Introducing https://t.co/E3LHVE410x, the agent-native serverless cloud. Your team is shipping code like never before. But you're getting caught up in manual, tedious DevOps work trying to deploy it. InstaCloud provides the serverless compute that lets your services autoscale, with all the infrastructure managed for you. Agents branch into complete replica environments when working, keeping prod safe and iteration speed high. And of course, it all works seamlessly with agents through MCP/CLI. Get off the traditional, legacy cloud. Start deploying your services on InstaCloud today.
🎬
视频
#647 🔗 原推
2026-09-29 15:41:49
Garry Tan
Palo Alto is amazingly ground zero for educrats ruining honors classes and banning personalized education in the name of equity All the parents need to do is show up, vote out the education bureaucrat virtue signalers and support merit and advanced placement again
中文: 帕洛阿尔托是教育阶层以公平为名破坏荣誉课程并禁止个性化教育的绝佳起点 父母只需要上场，投票选出教育官僚的美德信号，支持优点，并再次获得高级职位
#646 🔗 原推
2026-09-29 15:39:32
Garry Tan
Personal AI is in its Homebrew computer club era but it won’t stay there Truly personal AI will mean you need to own your own skills and memory
中文: 个人人工智能正处于其Homebrew电脑俱乐部时代，但不会停留在那里 真正个人化的人工智能意味着你需要拥有自己的技能和记忆
#645 🔗 原推
2026-09-29 15:37:53
Garry Tan
RT @eladgil: hmm https://twitter.com/eladgil/status/2104811178434392380/photo/1
media 0 共 1 项
#644 🔗 原推
2026-09-29 15:17:37
Garry Tan
One of the coolest things @capydotai does is coordination across multiple PRs and threads by default. You can land features and fixes that are adjacent and have each agent be aware and coordinate cross-process And you can see the agents speak across threads at your direction https://twitter.com/garrytan/status/2104953331533901863/photo/1
中文: @capydotai 最酷的事情之一就是默认地协调多个 PR 和线程。您可以找到相邻的功能和修复，并让每个代理都保持意识并协调交叉过程 你可以看到代理人员会朝你的方向通过相反的思路进行对话:
media 0 共 1 项
#643 🔗 原推
2026-09-29 15:15:57
Garry Tan
Incredible work @jgebbia and team Amazing design and agent experience in a US govt product is mind blowing 🤯
中文: 令人难以置信的工作 @jgebbia 和团队 在美国政府产品中，令人惊叹的设计和代理体验令人惊叹🤯
#642 🔗 原推
2026-09-29 15:12:20
Garry Tan
RT @jgebbia: Thanks @BretBaier for having me 🇺🇸 Airbnb co-founder leads Trump admin in creating 'concierge' to navigate government https://www.foxnews.com/video/6405770809112 #FoxNews
中文: RT @jgebbia:谢谢@BretBaier 给我的 Airbnb联合创始人带领特朗普政府制定“合作”以应对政府 #福克斯新闻
#641 🔗 原推
2026-09-29 15:11:02
Garry Tan
RT @tracewoodgrains: Writing for the New Yorker, Kang points out the realities of school funding in NYC, endorses ability grouping and standardized tests, and calls out the whims of graduate schools of education. Encouraging! https://twitter.com/tracewoodgrains/status/2104940912615072050/photo/1
中文: RT @tracewoodgrains:为《纽约客》撰文，康指出了纽约市学校资金支持的现实情况，支持能力分组和标准化考试，并指出了教育研究生院的奇思妙想。 鼓励!
media 0 共 4 项 media 1 media 2 media 3
#640 🔗 原推
2026-09-29 14:49:24
Garry Tan
Just fixed 28 bugs in gbrain in one fix wave PR with @capydotai and it feels so good https://twitter.com/garrytan/status/2104945927668850848/photo/1
中文: 只需在 @capydotai 的 fix Wave PR 中修复 gbrain 中的 28 个漏洞，感觉就太好了
media 0 共 1 项
#639 🔗 原推
2026-09-29 14:46:31
Garry Tan
This is what peak performance looks like
中文: 这就是巅峰表现的表现
#638 🔗 原推
2026-09-29 14:15:29
Garry Tan
RT @danprimack: Is there a point in doing next year’s Midas List? Just give Martin a trophy and move on.
中文: RT @danprimack:明年的《迈达斯榜单》有意义吗?给马丁一个奖杯，然后继续前进。
#637 🔗 原推
2026-09-29 14:03:23
Garry Tan
RT @justinsunyt: we built persistent agents for capy called captains our team used to manually forward agent messages through slack but now our captains just talk to each other in caveman speak behind our backs https://twitter.com/justinsunyt/status/2104811355689873446/photo/1
中文: RT @justinsunyt:我们为名为“船长”的公司建立了持久的代理 我们团队过去常常通过闲置来手动转发代理信息，但现在我们的队长只是用空洞的语调互相交谈，背后是
media 0 共 1 项
#636 🔗 原推
2026-09-29 14:03:11
Garry Tan
RT @raphaelschaad: I think fundamentally we've ~now arrived at the correct form factor for mass consumer AI: It was never going to be Google's search box but with AI, and a left-hand list of Chats was a 18mo experiment that has reached 100M users but won't be the thing that will scale to 1B users for next 10yrs — it will be messaging with a single bot, occasionally split to specific bots for special areas of work/life. Still ton of innovation to be done, e.g., in the bots–bots collab mechanisms, credentials, computer use, and generally optimizing the heck out of everything … … but the form factor has settled; the iPhone Classic has landed; the News Feed has been established.
#635 🔗 原推
2026-09-29 14:02:50
Garry Tan
RT @Voxyz_ai: Opus 5.5 can already run all night on its own. Give it 𝘁𝗵𝗲𝘀𝗲 𝟴 𝗚𝗶𝘁𝗛𝘂𝗯 𝗿𝗲𝗽𝗼𝘀 as a second brain and it feels like cheating: a code graph that indexes the Linux kernel in 3 minutes, a 95K-star memory plugin for Claude Code, and the agent brain where Garry Tan keeps 155K pages. Reading code (code graph) → codebase-memory: code only. It builds the graph locally without calling a model, and it's what we use for our own code graph → Graphify: graphs your docs, PDFs, images and videos along with the code. 122K stars Managing notes (knowledge base) → gbrain: every fact keeps its source, can be corrected or withdrawn, and is shared across your agents. We've hooked Gmail and Calendar up to it → claude-obsidian: turns your sources into linked, cited Obsidian pages, stored locally → qmd: local search written by Shopify CEO Tobi Lütke. It searches your notes, meeting transcripts and docs on your own machine, with no API key Remembering chats (session memory) → claude-mem: installs with one command, compresses every session into a summary, and picks up from it next time → mem0: a browser extension that shares one memory across ChatGPT, Perplexity and Claude on the web Tracking long tasks (task progress) → beads: replaces scattered plan docs with one task dependency graph, so long tasks don't lose context Send this to Claude Code and have 𝟯 𝘀𝘂𝗯𝗮𝗴𝗲𝗻𝘁𝘀 set it up for you 👇 "Help me build a second brain from these 8 GitHub repos: DeusData/codebase-memory-mcp, Graphify-Labs/graphify, garrytan/gbrain, AgriciDaniel/claude-obsidian, tobi/qmd, thedotmack/claude-mem, mem0ai/mem0, gastownhall/beads. Pick by four categories: code graph, knowledge base, session memory, task progress, at most one per category. First check what CLAUDE.md and Claude Code's built-in memory already record, and only add what they can't. Send out three subagents: - scout (use Haiku): read each README and list the install command, where the data is stored, whether it needs an API key, whether it costs extra tokens, whether it runs a background process, and the license. - architect: look at my project and CLAUDE.md, pick one per category, and explain why you picked it over the other one in that category. - verifier: after I restart Claude Code, open a brand-new session with claude -p and ask about a decision we made earlier. For the code graph, ask who calls a specific function. Paste me the raw output. Show me the commands before installing anything, and install only after I confirm. If you can't read a README, stop and ask me to paste it in. Don't fill anything in from memory."
#634 🔗 原推
2026-09-29 12:08:36
Garry Tan
RT @hiddnest: Good morning and now @AsideAI uses 14x less tokens for compaction and dreaming. (Fable 5.1) We heard you and tweaked some settings. Now the cache hit rate is 2x improved, 7x less token usage for dreaming and compaction in average, while increasing the accuracy. Also, we're planning to ship new harness updates which will improve token efficiency and speed. Enjoy.
中文: RT @hiddnest:早上好，现在@AsideAI使用的代币减少了14倍，用于压缩和做梦。(寓言 5.1) 我们听到了你的声音，并调整了一些设置。现在缓存命中率提高了2倍，用于做梦和压实的代币使用量减少了7倍，同时提高了准确度。 此外，我们计划发布新的线束更新，以提高代币的效率和速度。享受。
#633 🔗 原推
2026-09-29 12:07:31
Garry Tan
Codegen meets WhatsApp actually makes a ton of sense
中文: Codegen 与 WhatsApp 的交汇实际上意义有道理
#632 🔗 原推
2026-09-29 03:12:49
Garry Tan
RT @GJarrosson: Paul Graham's new essay has a line that explains a lot of what I see win in YC Sell to companies when they're tiny, and you grow at their rate, not yours. His e.g. is Stripe. Sign companies up at the first possible moment, charge on usage, and because payments never churn once they work, you compound at the growth rate of every startup you ever onboarded. Why selling to tiny startups beats chasing enterprise: 1. They decide in days, not quarters. Founders are sophisticated and move fast, so the best product wins. Enterprise sales is mud, and mud is not where quality wins. 2. They don't churn. Get in early on usage-based pricing and you're strapped to their growth. 3. Everyone underrates them. "They have no money" is backwards. They have VC money and a 10x-a-year goal growth rate. It is often a criticism of YC companies to sell to their own batchmates on day one, but do the ones who brag "we sell to enterprises" cap their growth at how fast a procurement team moves? Imagine being in a YC batch and solving problems for companies like WhatNot, After Query or Starcloud.... you are attaching yourself to a rocketship. Get them early. Charge on usage. Grow at their speed and hold on tight!
#631 🔗 原推
2026-09-29 02:31:32
Garry Tan
RT @garryslist: "You cannot build in a place where the state can reach in and take what you've already earned." In California, it stays yours. Don’t take that for granted. Chip in to defend the California Dream 👇 https://secure.actblue.com/donate/californiadream https://twitter.com/garryslist/status/2104738935855628784/video/1
中文: RT @garryslist:“你无法在国家能够触达并获取你已经获得收入的地方进行建设。” 在加利福尼亚，它一直留着你。不要把那视为理所当然。加入《加利福尼亚梦》👇
🎬
视频
#630 🔗 原推
2026-09-29 01:09:25
Garry Tan
RT @RespanAI: Introducing Span-01, the first hyper-parallel reasoning classifier built for unseen challenges (RLAIF). 2x cheaper, 18% better than Jev. 700x cheaper, 4% better than GPT-6 Luna. Frontier reasoning for every behavior, at classifier speed. • Span-01: #1 on Behavior Benchmark • Span-01 Lite: Better than Jev and completely free!
media 0 共 1 项
#629 🔗 原推
2026-09-29 01:06:34
Garry Tan
RT @charliee_: Won 1st place solo at @ycombinator's Own Your Intelligence hackathon! 🏆 Built QM Motion to give coding agents a sense of time, so they can catch UI bugs that only show up in motion. Here’s the demo 👇
中文: RT @charliee_:在 @ycombinator 的“你自己的智力”黑客马拉松上独得第一名!🏆 构建QM Motion，让编码代理获得时间感，以便他们能够捕捉到仅在运动中出现的UI漏洞。 演示👇
#628 🔗 原推
2026-09-29 01:03:56
Garry Tan
RT @anabology: Made a public google drive with a 'how it was made,' prompts + generated image assets, high quality audio exports, and the master mp4. Enjoy: https://drive.google.com/drive/folders/1OKYaR5Kv5lOeIROUtigpPZtCWauUJ01y?usp=sharing
中文: RT @anabology:通过“如何制作”进行公开谷歌播放，提示 + 生成图像资源，提供高质量的音频导出以及主 mp4。享受:
#627 🔗 原推
2026-09-28 20:32:39
Garry Tan
RT @gbrainio: The best in the world gathered @ycombinator for the Bring Your Own Intelligence hackathon. https://twitter.com/gbrainio/status/2104307776264798213/photo/1
中文: RT @gbrainio:全球最优秀的人齐聚@ycombinator，共同参加“自带智能”黑客马拉松。
media 0 共 1 项
#626 🔗 原推
2026-09-28 15:13:30
Garry Tan
RT @Karmedge: Meet Stormy. You hire it like anyone else: in a meeting. - Walk it through your business, live call - Runs your front desk and back office ops - Answers calls and texts - Fully generative software which builds itself as u use it Mini demo: Stormy on a rental shop's front desk:
中文: RT @Karmarge:见到Stormy。你像其他人一样雇佣它:在会议中。 - 通过您的业务、实时通话来行 - 运行您的前台和后台操作 - 回答电话和短信 - 完全生成的软件，可随您自身使用而自行构建 迷你演示:在租赁店的前台上大放不见:
🎬
视频
#625 🔗 原推
2026-09-28 14:49:39
Garry Tan
Karma isn’t a thing that happens, it’s a choice you can make now https://x.com/Padmaja_Yatra/status/2104217419917189226/video/1
中文: 因果可能并非偶然发生，而是您现在可以做出的选择
🎬
视频
#624 🔗 原推
2026-09-28 13:11:03
Garry Tan
RT @chr1sa: The bioethics circle jerks during the Human Genome Project, which were allocated 5% of the project's funding, were a total waste of time. They had countless conferences and workshops imagining bad things that could happen like punitive health insurance rates based on genetic risk, none of which actually came to pass. Meanwhile all the real issues that arose from ubiquitous DNA data, like accidentally doxing your criminal relatives with your own 23andMe data, were never imagined. It's a condemnation of the precautionary principle. Philosophers can make up all sorts of hypothetical scenarios in a vacuum but it's nothing like actually trying things and seeing all the weird second and third-order consequences that emerge and *then* dealing with the unwelcome ones.
中文: RT @chr1sa:人类基因组计划期间的生物伦理学圈抽搐，该项目获得了该项目5%的资金，这完全是浪费时间。 他们举办了无数次会议和研讨会，想象可能发生的坏事，比如基于遗传风险的惩罚性医疗保险费率，而这些风险实际上都没有通过。与此同时，所有因无处不在的DNA数据而产生的真实问题，例如不小心用你自己的23andMe数据对你的犯罪亲属进行掺杂，都从未被想象过。 这是对预防原则的谴责。哲学家可以在真空中构成各种假设情景，但这完全不像是真正尝试并看到那些奇怪的二阶和三阶后果，这些后果会随着那些不受欢迎的情况而产生。
#623 🔗 原推
2026-09-28 13:05:56
Garry Tan
RT @mikepat711: How is this real life? Seriously wtf man. https://twitter.com/mikepat711/status/2104256370757521502/video/1
中文: RT @mikepat711:这真实生活怎么样?真的，嗯。
🎬
视频
#622 🔗 原推
2026-09-28 04:49:11
Garry Tan
Anti-bot is for losers tbh
中文: 反机器人是失败者
#621 🔗 原推
2026-09-28 02:48:06
Garry Tan
RT @silennai: Last week, the S26 batch ended here are the lessons that we used to go 0 -&gt; 7 figures in 2 weeks ⬇️ https://twitter.com/silennai/status/2104256047636783332/photo/1
中文: RT @silennai:上周，S26批次结束 以下是我们过去在两周内 0 -&gt; 得出的7个数字⬇️
media 0 共 1 项
#620 🔗 原推
2026-09-28 01:59:36
Garry Tan
RT @pitdesi: friend of ours had their car stolen in the Mission While the police were taking my friends report, Flock ALPR’s spotted the car downtown and they were able to retrieve it. I wonder what would have happened without Flock! At the very least it would have taken much longer
中文: RT @pitdesi:我们的朋友在任务中偷走了他们的汽车 警察正在接我朋友的报告时，弗洛克·阿尔普发现了车子在市中心，他们得以找回。 不知道没有Flock会发生什么!至少需要更长的时间
#619 🔗 原推
2026-09-28 01:50:05
Garry Tan
"Oh I can do this with CDP" Haha you have not actually hit any real anti-bot then
中文: 哦，我可以用CDP做到这一点。哈哈，你其实还没有真正打击过任何反机器人
#618 🔗 原推
2026-09-27 19:13:32
Garry Tan
RT @stretchcloud: AgentMail just shipped AgentID, and it's a small feature pointed at a problem that's already out of control. The idea: an agent gets its own OpenID Connect identity, backed by a verified email and a signing key stored locally. No SDK, just two config values, and it slots straight into Clerk, Supabase, Auth0, and Better Auth. The demo shows an agent signing into Firecrawl and Turso with its own credential instead of a copy of yours. Here's why that matters right now. GitGuardian's 2026 report found 29 million leaked secrets last year, and flagged AI agent credentials specifically as a category that's spiraling, because the easiest way to give an agent access to anything is still to hand it your API key and hope. The bigger players agree this is the next infrastructure layer, not a nice to have. Okta launched a Universal Agent Identity Provider this year, in early access, plus a Cross App Access protocol built for agent to app handoffs. Auth0 shipped its own agent identity layer in May. WorkOS added agent registration as a credential type distinct from human users. Twilio went further and acquired Stytch this year specifically for its agent identity and consent tooling, which is about as clear a signal as an acquisition gets. I keep thinking about how OAuth itself only existed because password sharing between apps became unworkable around 2007. Same shape of problem, new class of actor. We handed agents write access to our inboxes, our databases, our deploy pipelines, and gave almost no thought to how they prove who they are or how you revoke them. AgentMail is a small startup solving a sliver of this, agent sign in specifically. But the fact that Okta, Auth0, WorkOS, and Twilio are all building some version of the same primitive in the same year tells me identity is about to become as core to agent infra as memory and tool calling already are. https://x.com/utpalnadiger/status/2104098648980439132
#617 🔗 原推
2026-09-27 19:11:48
Garry Tan
RT @ycombinator: “Tradeoffs that only pay off in 10 years will usually be underpriced.” Play the long game. When you think in decades rather than quarters, you’ll see that your potential customers have trajectories, not fixed sizes. Paul Graham on making startups more powerful: https://paulgraham.com/powerful.html
中文: RT @ycombinator:“10年内仅能取得回报的优惠通常会被低估。” 玩长游戏。当你在几十年而非季度时，就会看到潜在客户拥有轨迹，而非固定尺寸。 保罗·格雷厄姆:让初创企业更具影响力:
media 0 共 1 项
#616 🔗 原推
2026-09-27 17:12:13
Garry Tan
RT @ledwards: You may recall Larry Krasner, Philly's own Chesa Boudin. The Philadelphia DA used false claims in court to overturn a murder conviction and then tried to hide his lie/mistake from a judge when caught. You do not hate this movement enough. https://www.inquirer.com/crime/inq2/larry-krasner-post-conviction-case-controversy-timeline-20260925.html
中文: RT @ledwards:你可能还记得拉里·克拉斯纳，他是费城人队的切萨·布丁。 费城市立法院通过虚假陈述推翻谋杀罪，随后在被抓获后试图向法官隐瞒其谎言或错误。 你对这场运动的憎恨还不够。
#615 🔗 原推
2026-09-27 15:47:05
Garry Tan
Do not nerf Made in San Francisco
中文: 不要紧张 旧金山制造
#614 🔗 原推
2026-09-27 15:25:59
Garry Tan
Opus 5.5 with Openclaw is strangely smarter and better at completing tasks than with GPT-6 Astra Surprising
中文: 使用Openclaw的Opus 5.5比使用GPT-6 Astra更智能且更擅长完成任务 令人惊讶
#613 🔗 原推
2026-09-27 14:59:32
Garry Tan
If you like Muse and Grok Bot but still run into crazy antibot annoyances try @AsideAI browser with MCP with your agent on a spare laptop or computer you keep plugged in somewhere. It lets your agents your real credentials as yourself from a real Chromium. Gamechanger.
中文: 如果你喜欢Muse和Grok Bot，但仍然遇到疯狂的反机器人烦恼，可以尝试使用MCP的@AsideAI浏览器，使用备用笔记本电脑或电脑，让你一直用电脑插上电脑。 它让你的经纪人能够从真实的 Chromium 中实现真实的资质。游戏变革者
#612 🔗 原推
2026-09-27 14:47:02
Garry Tan
RT @giohuh_: What can Astra do when given a humanoid embodiment? We built HomeBody to find out. Controlled by GPT Astra, it carries out long-horizon tasks in a previously unseen kitchen—from tidying up across the room to retrieving remembered objects from ambiguous requests—without environment-specific training data or additional policy learning. Here's how we did it 👀: https://tml.stanford.edu/homebody/
中文: RT @giohuh_:当Astra被赋予人形实施例时，可以做些什么? 我们为了弄清楚，我们建立了HomeBody。由GPT Astra控制，它在一个前所未见的厨房中执行长距离任务——从隔场到从模糊请求中取回记忆中的物体——无需环境特定的训练数据或额外的政策学习。 我们是如何做到的👀:
🎬
视频
#611 🔗 原推
2026-09-27 04:50:27
Garry Tan
Most people get mad on social media but don’t know what to do to actually effect change We think you should come to a Garry’s List event and meet other people and organize to make it happen Mad about our kids educations? Come to a house party and organize for merit.
中文: 大多数人在社交媒体上会生气，却不知道该怎么做才能真正影响变化 我们认为你应该参加加里的名单活动，并与其他人见面并组织起来，以实现这一目标 疯狂地谈论我们的孩子教育?参加家庭聚会并组织起来以达到应有的价值。
#610 🔗 原推
2026-09-27 04:40:13
Garry Tan
This is my favorite way to fix bugs now @capydotai with GStack /autoplan on a production issue GPT-6 medium reasoning https://twitter.com/garrytan/status/2103989902476259702/photo/1
中文: 这是我现在最喜欢修复错误的方法 @capydotai 与 GStact 合作 /autoplan 处理生产问题 GPT-6 中等推理
media 0 共 2 项 media 1
#609 🔗 原推
2026-09-26 23:27:37
Garry Tan
RT @Scott_Wiener: We asked Richmond District residents for their thoughts on Connie Chan and Scott Wiener. They were clear about who they want to replace Nancy Pelosi in Congress. https://twitter.com/Scott_Wiener/status/2103934461712945280/video/1
中文: RT @Scott_Wiener:我们向里士满地区居民征求他们对康妮·陈和斯科特·维纳的看法。他们明确了自己想在国会中接替南希·佩洛西的是谁。
🎬
视频
#608 🔗 原推
2026-09-26 23:25:03
Garry Tan
Be chalant Don’t LARP
中文: 保持冷漠 别拉普
#607 🔗 原推
2026-09-26 19:39:54
Garry Tan
RT @poteto: https://x.com/i/article/2103566873501835264
#606 🔗 原推
2026-09-26 19:35:13
Garry Tan
RT @kiwicopple: we're running a hackathon after our conference - happening on Oct 3 anyone attending Select gets automatic admission come hack with the Supabase team and get a bunch of credits from @supabase, @claudeai , @stripe, and @vercel to hack sign up https://select.supabase.com/ https://twitter.com/kiwicopple/status/2103307290745172414/video/1
中文: RT @kiwicopple:我们正在举办一场关于10月3日的新闻发布会 参加Select的任何人都会自动入场 与Supabase团队合作，获取来自@supabase、@claudeai、@stripe和@vercel的大量信用信托 注册
🎬
视频
#605 🔗 原推
2026-09-26 19:01:25
Garry Tan
RT @ArmandDoma: “Endorsement: Who should replace Nancy Pelosi in Congress? It’s not even a choice” 🔥🔥🔥 https://twitter.com/ArmandDoma/status/2103907960544186393/photo/1
中文: RT @ArmandDoma:“支持:谁应该取代南希·佩洛西在国会?这甚至不是一种选择 🔥🔥🔥
media 0 共 1 项
#604 🔗 原推
2026-09-26 19:00:34
Garry Tan
RT @gradypb: The @BostonCollege Investment Committee (an LP and my beloved alma mater) asked for a few thoughts on what's happening in AI. I recorded a test run yesterday morning and then shared it with my partners, who encouraged me to share it more broadly... so here you go! This is not a sales pitch, it's just a reflection on what we're seeing. And it wasn't intended to be shared, so please pardon the rough edges. https://www.loom.com/share/c016702964a04dfba40e1777cf5f053a
中文: RT @gradypb:波士顿大学投资委员会(一位合伙人，也是我心爱的母校)就人工智能领域的情况提出了一些看法。昨天早上我记录了一次试运行，然后和我的伴侣分享，他们鼓励我更广泛地分享......所以你就来吧! 这并不是一个推销，只是我们所看到的一个反映。而且它并非有意分享，因此请原谅那些粗糙的边缘。
#603 🔗 原推
2026-09-26 15:59:18
Garry Tan
RT @bosmeny: Moritz is already the go-to law firm for so many companies I work with They pair top quality lawyers with AI to make everything faster and cheaper than seems reasonable
#602 🔗 原推
2026-09-26 15:50:33
Garry Tan
RT @dhh: Computer use on Omarchy is going to be magic with Cua! 🤘
中文: RT @dh:在《Cua》中，使用计算机对Omarchy来说将是神奇的!🤘
#601 🔗 原推
2026-09-26 15:50:15
Garry Tan
RT @dhh: It'll soon be seen as deeply irresponsible to have critical code reviewed by human eyes alone.
中文: RT @dh:仅凭人眼就审查关键代码，很快就会被视为极不负责任。
#600 🔗 原推
2026-09-26 15:39:31
Garry Tan
I think she should be a private citizen. She has done incalculable harm to our community and last year I asked for her resignation.
#599 🔗 原推
2026-09-26 15:38:26
Garry Tan
https://www.sfchronicle.com/opinion/editorials/article/scott-wiener-connie-chan-pelosi-22448323.php
#598 🔗 原推
2026-09-26 15:37:20
Garry Tan
The SF Chronicle agrees with me: Connie Chan is an awful candidate to represent San Francisco and her track record on helping the people SF on housing, public safety and any kind of accountability is a disgrace and abysmal https://twitter.com/garrytan/status/2103871405511311501/photo/1
中文: 《旧金山纪事报》同意我的观点:康妮·陈是代表旧金山的糟糕候选人，她在住房、公共安全以及任何问责方面帮助旧金山人民的记录都是一种耻辱，也非常糟糕
media 0 共 1 项
#597 🔗 原推
2026-09-26 15:36:45
Garry Tan
RT @dhh: This has never been more true. The opportunity cost of endless planning and contemplation just went way up. Agents allow you to just try vastly more stuff, so let them, and you'll learn way more, way faster.
中文: RT @dh:这一点从未如此真实。无休止的规划和思考的机遇成本刚刚上涨。代理允许你尝试更多内容，所以可以让他们学习，而且学习速度更快。
#596 🔗 原推
2026-09-26 15:33:55
Garry Tan
RT @SusanDReynolds: These pro-crime people are ridiculous. Just this week @flocksafety led @SanJosePD to an Uber driver who molested a teen boy in his car. For the 1% of rogue bad actors, 99% of the usage has been indispensable in catching criminals.
中文: RT @SusanDreynolds:这些支持犯罪的人简直荒谬至笑发。就在本周，@flocksafety 将@SanJosePD 引向了一名优步司机，该司机在其车内对一名未成年男孩进行了性侵。对于1%的流氓恶意行为者而言，99%的用法在抓捕罪犯方面一直不可或缺。
#595 🔗 原推
2026-09-26 15:19:01
Garry Tan
RT @Twolfrecovery: The Mayor of San Jose and hopefully a future governor has more spine than almost everyone in a leadership position in the Democratic party in California. He knows criminal justice reform and "harm reduction" have gone too far.
中文: RT @Twolfrecovery:圣何塞市长以及未来的州长希望比加利福尼亚州民主党中几乎所有处于领导地位的人都更有希望。他知道刑事司法改革和“减少伤害”已经走得太远了。
#594 🔗 原推
2026-09-26 15:18:27
Garry Tan
RT @levelsio: It's really crazy how strongly wealth inverse correlates with regulatory burden If you make it easy for people to start and run businesses and don't get in their way too much, everyone gets wealthier If you make it hard for people to start and run businesses and get in their way constantly, everyone gets poorer You can see the effect clearly within the EU: Countries with low regulatory burden like Switzerland and Ireland have much higher incomes than places like Spain or Hungary where its regulatory burden is high https://nomads.com/europe/chart?chart_x=regulatory_quality&chart_y=gdp_per_capita_in_usd&show_labels=true&show_population_as_bubble_size=true
中文: RT @levelsio:财富逆向与监管负担的关联程度非常疯狂 如果你让人们轻松创办和经营企业，而不会过多地妨碍他们，那么每个人都会变得更富有 如果你让人们难以创办和经营企业，并不断妨碍他们，那么每个人都会变得更穷 你可以清楚地看到欧盟内部的影响: 像瑞士和爱尔兰这样监管负担较低的国家，其收入远高于西班牙或匈牙利等监管负担较高的地区
media 0 共 1 项
#593 🔗 原推
2026-09-26 15:18:18
Garry Tan
RT @ycombinator: One of the biggest surprises in AI over the last few years has been how well coding agents generalize beyond software. In a recent essay, MIT professor Philip Isola argued that we may be entering the era of robot-use agents: general-purpose models that can control different robots, write policies, and learn new physical tasks with little or no robot-specific training. In this episode of Decoded, we're joined by the founders of @theWaddleLabs and @Robocurve, two of the startups whose work helped drive this realization. They're working at the frontier of using general-purpose models to control robots, and their recent demos helped inspire the growing conversation around robot-use agents. Together, we dig into the research behind that idea, from code-as-policies and vision-language-action models to the harnesses and evals needed to make these systems work in the real world. 00:00 — What Are Robot-Use Agents? 02:11 — From Vision-Language-Action Models to General LLMs 04:07 — The Bitter Lesson for Robotics 07:04 — How Coding Agents Learned to Control Robots 10:41 — How Robots Learn From Experience 14:22 — The Harness as a Form of Robot Intelligence 16:14 — Watching Astra Control a Robot 20:55 — Why General Models May Win in Robotics 26:05 — How Close Are General-Purpose Robots?
中文: RT @ycombinator:过去几年人工智能领域最大的惊喜之一，是编码代理在软件之外的推广程度。 在最近的一篇文章中，麻省理工学院教授菲利普·伊斯拉指出，我们可能正在进入机器人使用代理的时代:通用模型，能够控制不同的机器人，编写策略，并在几乎没有或没有针对特定机器人的训练的情况下学习新的物理任务。 在《解码》的这一集中，我们与@theWaddleLabs和@Robocurve的创始人们共同共同关注，这两家初创企业的工作推动了这一认识。他们正在前沿使用通用模型来控制机器人，而他们最近的演示也推动了围绕机器人使用代理的日益发展的讨论。 我们共同深入探讨这一理念背后的研究，从代码即政策、视觉语言行动模型，到使这些系统在现实世界中发挥作用所需的利带和椭圆形。 00:00 — 什么是机器人使用代理? 02:11 — 从视觉语言-行动模型到通用LLM 04:07 — 机器人的苦涩教训 07:04 — 编码代理如何学会控制机器人 10:41 — 机器人如何从经验中学习 14:22 — 机器人智慧的一种方式 16:14 — 观看Astra Control机器人 20:55 — 为什么通用模型可能在机器人领域获胜 26:05 — 通用机器人有多近?
🎬
视频
#592 🔗 原推
2026-09-26 15:17:01
Garry Tan
RT @jochenmadler: We trained a model to predict AI-written blog posts from structural features - and discovered the shape of slop. On posts it had never seen, it told AI and human apart with 98% accuracy, getting only 19 of 1,740 wrong. Paper and code: https://arxiv.org/abs/2609.15369 https://twitter.com/jochenmadler/status/2103530162318901673/photo/1
中文: RT @jochenmadler:我们训练了一个模型，从结构特征中预测了AI编写的博客文章，并发现了斜体的形状。 在它从未见过的帖子中，它以98%的准确率告诉了人工智能和人类，在1740个错误中只有19个。 纸质代码:
media 0 共 1 项
#591 🔗 原推
2026-09-26 15:12:00
Garry Tan
RT @scottstts: My god this is such a good speech that every SWE needs to hear. You know what? Every person should hear it Keep the happy memories, eyes on the reality, be excited about the future. That’s the best that anyone can do https://twitter.com/scottstts/status/2103401938058723765/video/1
中文: RT @scottstts:天哪，这是一篇非常好的演讲，每位SWE都需要听。你知道吗?每个人都应该听到它 保持快乐的回忆，关注现实，对未来充满期待。这是任何人都可以做的最好的事情
🎬
视频
#590 🔗 原推
2026-09-26 14:11:19
Garry Tan
RT @jeffzwang: We built Codex/Claude ultra mode, but for web research! Uses heavy agent parallelization, code execution, DB tool, and some other goodies, all orchestrated with frontier models The squad kept iterating until it was clearly SOTA. Lemme know what you think 🫡
中文: RT @jeffzwang:我们为 Codex/Claude 超模式构建了代码，但用于网络研究! 采用重代理并行化、代码执行、数据库工具以及一些其他好工具，这些工具都采用前沿模型编排 小队一直在进行分队，直到明显是索塔。莱姆知道你的想法🫡
#589 🔗 原推
2026-09-26 00:57:05
Garry Tan
Astra is very impressive
中文: 阿斯特拉非常令人印象深刻
#588 🔗 原推
2026-09-26 00:56:55
Garry Tan
RT @Twolfrecovery: Gary McCoy would be a disaster for San Francisco. Gary is thin skinned, vindictive and oversaw the Tenderloin Linkage Center in 2021 where $26 million was wasted ($10m went to his org HealthRIGHT360) and only 1.7% of the homeless received a treatment referral. Hell NO! https://twitter.com/Twolfrecovery/status/2103599308217917549/photo/1
media 0 共 1 项
#587 🔗 原推
2026-09-26 00:55:59
Garry Tan
RT @mathewpregasen: Given that I was part of YC W20 and now YC F26, I'm often asked about the differences between the batches. This either comes from actual curiosity or some hidden expectation for me to say it's "gone downhill". Let me answer the former and dispel the latter. There are huge differences, but probably not in the way you'd expect. There's a lot of public chatter about YC backing younger founders, inexperienced founders, and losing its touch. This isn't true. Remotely. There were incredibly talented founders in YC W20, and the same is true for this batch as well. Last batch, we had PostHog and WhatNot. This time, we'll have similar outcomes. The other day, someone tweeted "YC doesn't back visionary founders". But the point of doing YC isn't to advertise your vision; it's to quickly test your hypothesis in a structured environment. The biggest difference between my batch experiences is the location. Previously, YC was in Mountain View, and while YC recommended founders to live in Mountain View, many (if not most) lived in San Francisco. This meant a fair part of the YC experience was commuting down on the Caltrain, attending events, and then going your separate ways. Now, YC is in Dogpatch, and literally everyone lives in the same four apartment buildings. That makes a massive difference. You run into founders in the elevator. You run into founders when getting packages, You run into founders on the street. Everyone is building something, everything is talking about agents & harnesses, everyone is talking about the YC calendar. This sense of location builds community. I think YC's done a great job of not going full-send on campus. It's good to have independence. You choose where you live, what you eat, how you work, what you do. But the co-location in Dogpatch creates a lot of spontaneous encounters that lead to learnings and sometimes even new customers. Another difference is the schedule and partitions. One issue with YC W20 is it had a rigid schedule, and too many founders in one room. It's hard to build relationships if you feel the constant pressure to meet more and more people. The new structure keeps everyone in consistent groups so you build actual connections; it also has opportunities to meet everyone else without exhausting everyone's social battery. I can somewhat understand the YC haters because YC can look like a bunch of copycats from the outside. I remember a few batches after YC W20, I was looking at YC announcements from W23 or S24 and thinking the same. But then you return to YC and realize that's all a bunch of nonsense: social media distorts reality, and the same brilliant founders that YC backs have been in YC all along. All in all, I think @ycombinator's evolution has been positive; the move to Dogpatch was brilliant; and the orange remains really f'ing orange.
#586 🔗 原推
2026-09-26 00:55:46
Garry Tan
RT @vivianmshen: they signed $1.37B in LOIs during the YC summer batch! incredibly ambitious
中文: RT @vivianmshen:他们在YC夏季批次期间签署了13.7亿美元的LOI!极具雄心
#585 🔗 原推
2026-09-26 00:45:59
Garry Tan
RT @benspringwater: I know so many people who are curious about AI but don't use X because of negative perceptions or bc they don't know where to start. Yesterday my wife asked me to get her set up on X. So, I took the opportunity to make https://xstarterpack.com/ - it's a simple prompt anyone can use to set up their X to follow AI news & discourse (and avoid rage bait). It's ~200 accounts - admittedly incomplete and reflecting my biases - but it covers many important voices across research, policy, economics, business, finance, applications, and practical use, with a range of political and ideological orientations. At the very least, it’s a good 80/20 (done is better than perfect). Hope it helps more people tune in to the greatest show on earth!
#584 🔗 原推
2026-09-26 00:45:47
Garry Tan
RT @garryslist: Declinism is a status game. It lets you sound worldly while betting against your own state. California isn't dying. It's being rebuilt by people too busy building to post about the apocalypse. If you came here, built something, and made it yours, you're the rebuttal. DM us or email [email protected] and share your #CaliforniaDream.
🎬
视频
#583 🔗 原推
2026-09-26 00:35:35
Garry Tan
RT @Rafa_Mangual: A woman visiting family in Richmond, VA had her car jacked with her two babies still inside. Police used their Flock camera network to narrow the search area and found the vehicle, and children within 45 minutes. This is good. We should not stop the police from doing this. https://www.wtvr.com/news/local-news/richmond/mom-wants-more-flock-sept-24-2026
中文: RT @Rafa_Mangual:一名女子在弗吉尼亚州里士满探亲时，车内还挂着她的两个孩子。警方利用他们的Flock摄像头网络缩小了搜查范围，并在45分钟内找到了车辆和儿童。 这很好。我们不应阻止警察这样做。
#582 🔗 原推
2026-09-25 21:55:34
Garry Tan
RT @ycinnewyork: Nikole Hannaj-Jones criticized the teachers and principal in her daughter's school for low-expectations, for not grading assignments, for not addressing unruly cafeterias but she continues to blame racism and resources. She's too smart to not name the true systemic culprit. "Hannah-Jones blames this dysfunction on persistent inequality—wrongly. But she misses the true culprit driving the problems: the teachers’ union constraints on schools and the concurrent lack of accountability for low-performing schools." ~@dany_egorov https://cityjournal.substack.com/p/the-unmentioned-villain-in-nikole
中文: RT @ycinnewyork:尼科尔·汉纳吉-琼斯批评她女儿学校的老师和校长因未对作业进行评分而感到低调，却指责她未处理不守规矩的餐厅，但她仍继续将责任归咎于种族主义和资源。她太聪明了，不能说出真正的系统性罪魁祸首。 汉娜-琼斯错误地将这种功能失调归咎于长期存在的不平等。但她忽略了导致问题的真正罪魁祸首:教师工会对学校的限制，以及同时对表现不佳的学校缺乏问责。~@dany_egov
media 0 共 1 项
#581 🔗 原推
2026-09-25 21:54:29
Garry Tan
Legalize personalized education
中文: 将个性化教育合法化
#580 🔗 原推
2026-09-25 13:03:58
Garry Tan
RT @evanwch: Progressives behind the scenes are reading and debating Nothing Left because they know everything in it is true. They just can’t admit so publicly https://nymag.com/intelligencer/article/evan-barker-nothing-left-interview-democrat-midterm-burn-book.html
中文: RT @evanwch:幕后的进步人士正在阅读并争论“Nothing Left”，因为他们知道其中的一切都是真实的。他们就是不能公开承认
#579 🔗 原推
2026-09-25 13:02:41
Garry Tan
RT @sarthakgh: “Just because it came from a scientist, it doesn’t make it scientific….Don’t think that just because you are an alarmist, you are doing any social good.”
中文: RT @sarthakgh:“仅仅因为它来自一位科学家，它并不能让它变得科学。”不要以为仅仅因为你是个危言耸听者，你就在做任何社会好事。
#578 🔗 原推
2026-09-25 12:48:50
Garry Tan
RT @jaltma: What an insane time. We are so lucky to be working in tech right now.
中文: RT @jaltma:真是疯狂的时刻。我们很幸运，现在能从事科技行业工作。
#577 🔗 原推
2026-09-24 19:00:34
Garry Tan
RT @paulg: Paweł Huryn tested models' ability to find bugs planted in code. Cost increases exponentially with performance (note the log scale on the x axis). I had ChatGPT superimpose a trend line. Bargain = above the line. https://twitter.com/paulg/status/2102732689245958313/photo/1
中文: RT @paulg:Paweł Huryn 测试了模型在代码中发现错误的能力。成本随着性能的提升呈指数级增长(注意x轴上的日志量表)。我让ChatGPT叠加了一条潮流线。交易 = 高于行。
media 0 共 1 项
#576 🔗 原推
2026-09-24 04:19:12
Garry Tan
Muse is very impressive
中文: 缪斯非常令人印象深刻
#575 🔗 原推
2026-09-24 04:11:27
Garry Tan
RT @himanshustwts: TLDR - you have to be extremely high agency with a lot of ownership so you can work around across the stack. https://twitter.com/himanshustwts/status/2102824319831400692/photo/1
中文: RT @himanshustwts:TLDR——你必须拥有极高的代理权，以便能够跨层工作。
media 0 共 1 项
#574 🔗 原推
2026-09-24 04:06:58
Garry Tan
This is not priced in yet
中文: 这尚未定价
#573 🔗 原推
2026-09-24 04:05:53
Garry Tan
RT @HariAyapps: Archgen Labs (@archgen_) is now backed by Y Combinator (@ycombinator) We are a research lab for silicon design, with one ambition: make chip design 1,000× faster. For decades, computing advanced by shrinking transistors. The next chapter for compute workloads will increasingly depend on building vertically - stacking active logic layers to create 3D chips. This makes designing those chips much harder but also opens up new possibilities for performance and efficiency. At @archgen_ , we’re bringing together foundational models, EDA tools, and physics-informed surrogate models to explore more designs, predict their behavior faster, and optimize how they’re built. Grateful to Kulveer Taggar (@kul), Garry Tan (@garrytan) and the @ycombinator team for backing us, and to the customers, mentors, and collaborators who have helped us get here. If you want to get to tapeout faster, we’d love to talk. DM us. #Semiconductors #OpenROAD #OpenLane #OpenSourceEDA #ChipDesign #VLSI #PhysicalDesign #AIForHardware #3DIC @naveen_venk
中文: RT @HariAyapps:Archgen Labs(@archgen_)现在由Y Combinator(@ycombinator)支持 我们是硅设计的研究实验室，只有一个目标:让芯片设计速度更快1000倍。 几十年来，计算技术一直通过缩小晶体管来推进。计算工作负载的下一章将越来越依赖于垂直构建——将活动逻辑层叠叠以创建3D芯片。 这使得设计这些芯片变得更加困难，同时也为性能和效率开辟了新的可能。 在 @archgen_ ，我们将汇集基础模型、EDA 工具和物理知识的代孕模型，以探索更多设计，更快地预测其行为，并优化其构建方式。 感谢Kulveer Taggar(@kul)、Garry Tan(@garrytan)以及@ycombinator团队，感谢他们支持我们，感谢那些帮助我们来到这里的客户、导师和合作者。 如果你想更快地进行胶带，我们很乐意聊聊。DM我们。 #半导体 #OpenROAD #OpenLane #OpenSourceEDA #芯片设计 #VLSI #物理设计 #AIForHardware #3DIC @naveen_venk
media 0 共 1 项
#572 🔗 原推
2026-09-24 04:05:13
Garry Tan
RT @alexandr_wang: state of the art motherfuckers
中文: RT @alexandr_wang:艺术混蛋
#571 🔗 原推
2026-09-24 02:56:47
Garry Tan
Two important simultaneous important trends in how startups get customers: Make software agents want Use agents to make people want software
中文: 初创企业如何获得客户的两个重要同时趋势: 让软件代理成为需要 使用代理来让人们想要软件
#570 🔗 原推
2026-09-24 02:55:51
Garry Tan
RT @HumanProgress: The chance of a person dying in a natural catastrophe has declined by nearly 99% since the 1920s and 1930s. People today are much more likely to survive natural disasters because of increased wealth and technological progress. Buildings are better constructed, weather satellites and sophisticated computer models provide early storm warnings, and swift medical interventions halt developing epidemics.
中文: RT @HumanProgress:自20世纪20年代和30年代以来，一个人在自然灾害中死亡的可能性已下降了近99%。 如今，由于财富增加和技术进步，人们更有可能在自然灾害中幸存下来。建筑结构更先进，气象卫星和复杂的计算机模型能够提供早期风暴预警，而迅速的医疗干预措施将阻止疫情的蔓延。
media 0 共 1 项
#569 🔗 原推
2026-09-24 02:53:49
Garry Tan
RT @stanine: Today, we’re shipping updates to @Rippling AI that make it easier to deploy data connections, skills, and automations safely across your org. They’re available to every employee, but they’re managed centrally and they’re auditable. https://twitter.com/stanine/status/2102827827787899232/video/1
中文: RT @stanine:今天，我们将向 @Rippling AI 发送更新信息，以便更轻松地在您的软件中安全地部署数据连接、技能和自动化。每位员工都可以使用，但需要集中管理，并可进行审计。
🎬
视频
#568 🔗 原推
2026-09-24 02:52:50
Garry Tan
RT @naval: Every functioning institution is merit-based, not participation-based.
中文: RT @navale:每个正常运作的机构都以绩效为基础，而不是以参与为基础。
#567 🔗 原推
2026-09-24 02:52:22
Garry Tan
RT @raphaelschaad: Y Combinator's coming to Zürich tomorrow! I'll be joining a panel hosted by Max Rumpf (Founder https://t.co/8ZBeeceyBD, YC S23) on AI, research, startups &amp; Swiss building between SFBA/Zürich. → Sign up below, only few more seats available: https://luma.com/c3c8b606 https://twitter.com/raphaelschaad/status/2102779230887067802/photo/1
中文: RT @raphaelschaad:Y Combinator 明天来苏黎世! 我将加入由马克斯·朗普夫(创始人 S23)主办的关于人工智能、研究、初创企业和安普姆的专题小组讨论;瑞士在苏黎世证券交易所(SFBA/Zürich)之间的建设。 → 注册如下，可供座位的座位很少:
media 0 共 1 项
#566 🔗 原推
2026-09-24 01:49:23
Garry Tan
RT @ian_dot_so: People forget how bad Claude used to be. In YC W23 we got access through a Slack bot, and it was generally considered a joke next to ChatGPT, which was all anyone in the world was talking about. Copilot was SotA for real engineering work. SVB collapsed the day before fundraising. Cursor wasn't a thing yet. Claude Code wouldn't come out for 2 more years (!). Prompt engineering was the hot skill to double down on. Founders who built tightly coupled harnesses and guardrails had a bad time once agents actually worked in late 2025. That was only ~3 years ago. Things are changing mind-bendingly fast.
中文: RT @ian_dot_so:人们忘记了克劳德曾经有多糟糕。 在YC W23中，我们通过Slack机器人获得了访问权限，这通常被认为是与ChatGPT并置的玩笑，而ChatGPT是世界上所有人都在谈论的。Copilot 是 SotA 的真正工程工作。 SVB在筹款前一天倒闭了。柯索还不是个事。克劳德·德西德不会再推出两年了(!)。快速工程是加倍努力的热技能。 建造紧密耦合的线束和护栏的创始人们在2025年底经纪人实际工作时过得很糟糕。 那才是三年前的事了。事情变化得非常迅速。
#565 🔗 原推
2026-09-24 01:46:58
Garry Tan
RT @mattdorsey: “Unfortunately, the far-left fringe is back — this time to elevate the civil liberties of motor vehicles over the public-safety imperatives of San Franciscans themselves.” In defense of ALPRs and speed safety cameras, my op-ed in today’s @sfexaminer. https://www.sfexaminer.com/forum/san-francisco-s-public-safety-technology-isn-t-the-problem-the-politics-around-it-is/article_2fc86991-ffa8-45e0-9320-1e1811cbb5a0.html?utm_medium=social&utm_source=twitter&utm_campaign=user-share via @sfexaminer
中文: RT @mattdorsey:“不幸的是，极左边缘地区又回来了——这次是为了提升机动车的公民自由，而不是旧金山市民自身对公共安全的要求。” 为ALPR和速度安全摄像头辩护，我在今天的@sfexaminer上发表了评论文章。 通过 @sfexaminer
#564 🔗 原推
2026-09-24 01:44:24
Garry Tan
RT @BenceRedmond: this is what you're shown on day 1 of YC. don't be scared of launching early. https://twitter.com/BenceRedmond/status/2102787813968322627/photo/1
media 0 共 1 项
#563 🔗 原推
2026-09-24 01:33:23
Garry Tan
RT @ycombinator: Congrats to Alex, Ben, and @hubble_network (YC W22) on their $200M Series C at a $1.6B valuation! Hubble's satellite network lets any Bluetooth device connect directly to orbit with just a software update. Cellular covers about 15% of Earth's landmass, and tracking anything beyond that has meant satellite hardware costing $25 or more per device. Hubble brings connectivity under 50 cents per device, which changes the math for billions of pallets, shipping containers, water pumps, and farm equipment that were never worth connecting before. https://x.com/hubble_network/status/2102808178467762425/video/1
中文: RT @ycombinator:以1.6亿美元的估值向Alex、Ben和@hubble_network(YC W22)祝贺他们。 哈勃的卫星网络允许任何蓝牙设备仅通过软件更新即可直接连接到轨道。 蜂窝网络覆盖了地球约15%的陆地，而追踪任何超出此范围的设备意味着每台设备的卫星硬件成本约为25美元。哈勃将每台设备的连接性低于50美分，这改变了数十亿托盘、集装箱、水泵和农用设备的计算，这些设备此前从未值得连接。
🎬
视频
#562 🔗 原推
2026-09-24 01:30:31
Garry Tan
RT @SpamRoss: Numeral has raised a $100M Series C, led by Insight Partners with participation from Salesforce Ventures, Benchmark, Mayfield, Uncork, Y Combinator, FCVC, and Geodesic. This brings our total funding to $157M. @numeral solves U.S. sales tax, VAT, and GST compliance for 3,500+ global businesses, trusted by companies like Supabase, Eight Sleep, and Graza. We’re building the future of tax -- where AI agents, human expertise, and global compliance come together in one system. Since starting the business in 2023, we’ve grown quickly across our core e-commerce and software segments, while expanding into new industries including manufacturing, distribution, and wholesale. There is a lot of work in front of us. Sales tax is more dynamic than ever, with states like California implementing new taxes on software and more states using AI for enforcement. We are hiring across the org from engineering to product, sales, and marketing. And we’ve just opened our new office in NYC. If you are looking for what’s next, I’d love for you to take a peek at our careers page. Excited for the year ahead. And humbled to be working with investors like @insightpartners, @rebeccaliudoyle, @SalesforceVC, @katiessschwartz, @ZKokosa, @MayfieldFund, @SPangulur, @benchmark, @chetanp, @ycombinator, @gustaf, @uncorkcap, @susanwliu, @GeodesicCap, @arvindayyala, @FundersClub, @mittal, @mantisVC, @AlexPallNY, @ShaanVP, @lennysan, @moizali, @ringmybeller, @khency6, @asturnerr, @maariabajwa and so many others.
🎬
视频
#561 🔗 原推
2026-09-24 01:28:21
Garry Tan
RT @neetu_arnold: Well would you look at that. Standardized tests are not pointless ⬇️
中文: RT @neetu_arnold:你看看那个吗?标准化测试并非毫无意义⬇️
#560 🔗 原推
2026-09-24 01:23:11
Garry Tan
RT @chrispisarski: +$2M ARR since this tweet people are keen to build their own workflows via Claude/Muse/Grokbot instead of buying more tools here are some learnings: 1) be part of the workflow - expansion into existing accounts has been a really big driver. some companies were using our API/MCP to source candidates via Claude, now they also find their clients and run their whole GTM on our data. we've been mapping out all these use cases and building a library they can just hand to their agents to figure out every way we/our data can help them 2) doubling down on AEO, we finally found what really works here and are doing a lot more of it constant experimentation and ideation drives most of it, more "demo-generating" activities instead of building dashboards / tracking the wrong things helped a ton we also started doing weekly syncs with other founders to learn/share any lessons (adding a few more ppl, reach out if you want to join our next sync!) 3) enterprises are becoming Claude-pilled we've started seeing a big inbound wave of large enterprise companies that want to build on our data inside Claude the adoption of Claude at these companies is exploding and so is the need to help them power all their workflows & replace some of the tools they've been buying. we're working with almost all of big tech and the top AI labs to help them do that
#559 🔗 原推
2026-09-24 01:22:39
Garry Tan
RT @SusanDReynolds: Connie Chan said on video that property crime wasn’t serious. As someone whose San Francisco home was broken into I can assure Connie it is serious to the victims.
中文: RT @SusanDreynolds:康妮·陈在视频中表示，财产犯罪并不严重。作为一个旧金山被闯入的人，我可以向康妮保证，这对受害者是认真的。
#558 🔗 原推
2026-09-23 23:35:52
Garry Tan
On the one hand you can say a coding harness done right is syntactic sugar On the other hand, I have seen a real speedup in being able to fix issues and land features much faster with no increase in time in-editor
中文: 一方面，可以说正确的编码工具是语法糖 另一方面，我看到能够更快地解决问题和着陆功能，且编辑时间不会增加
#557 🔗 原推
2026-09-23 23:02:25
Garry Tan
Bigger fixes, better test coverage, more issues resolved and faster https://twitter.com/garrytan/status/2102896035874156873/photo/1
中文: 更大的修复，更好的测试覆盖，解决更多问题，以及更快的
media 0 共 1 项
#556 🔗 原推
2026-09-23 23:00:59
Garry Tan
RT @lordspline: incredible https://twitter.com/lordspline/status/2102894332726669455/photo/1
中文: RT @lordspline:精彩内容
media 0 共 1 项
#555 🔗 原推
2026-09-23 22:56:46
Garry Tan
RT @ManhattanInst: When Roland Fryer sought to improve 20 struggling Houston schools, he implemented five practices: more instructional time, high-dosage tutoring, frequent teacher feedback, data-driven instruction and a culture of high expectations. The results were extraordinary. In math, elementary students gained four additional months of learning each year; secondary students gained nearly eight months in a nine-month school year. Every graduating senior in the high schools was admitted to a two- or four-year college. Roland Fryer writes about Nikole Hannah-Jones, school choice, and the real way to turn struggling schools around in @WSJ: https://www.wsj.com/opinion/the-education-of-nikole-hannah-jones-094ee262
中文: RT @ManhattanInst:当罗兰·弗莱尔试图改善休斯敦20所陷入困境的学校时，他实施了五种措施:多学习教学时间、高教学辅导、频繁的教师反馈、数据驱动型教学以及高期望文化。 结果非同寻常。 在数学方面，小学生每年增加四个月的学习经验;中学生在九个月的学年中增加了近八个月。高中每年毕业的高三学生都被一所两四年制大学录取。 罗兰·弗莱尔在@WSJ上撰文报道了尼科尔·汉娜-琼斯的择校选择，以及扭转困境学校真正方式的方式:
#554 🔗 原推
2026-09-23 22:53:01
Garry Tan
RT @0xluffy: capy compresses the loop down to only decision making you talk to capy, decide what to do, decide if PR is good enough. then rinse and repeat while all your capy threads work in parallel https://twitter.com/0xluffy/status/2102846450279575657/photo/1
中文: RT @0xluffy:将循环压缩到仅由决策 你与 papy 交谈，决定该做什么，判断 PR 是否足够好。然后冲洗并重复，同时将所有 pay 线程并行操作
media 0 共 1 项
#553 🔗 原推
2026-09-23 22:52:17
Garry Tan
RT @incitafusio: Since SFPD installed ~500 flock cameras citywide in 3/24, “smash-and-grabs” dropped from 19,841 in 2023 to fewer than 5,400 in 2025, a 73% decline. We can't increase public safety in SF without technology & a multi-prong approach to crime. Most major cities in CA already have this & it IS working. SJ has nearly 950 flock cameras & their retail+biz spaces are booming. I'll be at the DCCC tonight to give my public comment as a longtime resident and parent, with an elderly dad walking the streets of SF. Safety is #1 priority for families in SF & we can smartly trade some license plate privacy until our city is back on track.
中文: RT @incityfusio:自旧金山警察局(SFPD)在3月24日全市安装约500个群体摄像头以来，“抢夺”数量从2023年的19841台下降至2025年的不到5400台，降幅为73%。没有技术和多种犯罪处理方式，我们就无法提高旧金山的公共安全。加利福尼亚州大多数主要城市已经拥有这一点，并正在运作。SJ 拥有近 950 个群摄像头，其零售+业务空间正在蓬勃发展。 今晚我将参加DCCC，以长期居住和父母的身份发表公开评论，一位年迈的父亲在旧金山的街头行走。安全是旧金山家庭的首要任务，我们可以巧妙地换取一些车牌隐私，直到我们的城市重回正轨。
media 0 共 1 项
#552 🔗 原推
2026-09-23 20:48:47
Garry Tan
RT @garryslist: Special shoutout to @liveink for taking the initiative to shooting and producing his own. Share your California Dream with the world today, this state is where it all happens!
中文: RT @garryslist:特别向@liveink大声喊叫，主动拍摄并制作自己的节目。今天就向世界分享你的加州梦想，这个州就是一切发生的地方!
#551 🔗 原推
2026-09-23 20:48:37
Garry Tan
RT @mattdorsey: Remember car break-ins? San Francisco is on pace to end 2026 with a small fraction of what *used* to be our City’s most pervasive crime. Automated License Plate Readers (ALPRs) have been a game-changer in ending that. Yet, incredibly, some activists want ALPRs gone… (1/2) https://twitter.com/mattdorsey/status/2102855751006097721/photo/1
中文: RT @mattdorsey:还记得汽车闯入吗? 旧金山正以略小的速度结束2026年，其中仅占我们城市中最普遍犯罪的一小部分。自动车牌读取器(ALPR)在结束这一目的方面已经改变了游戏规则。 然而，一些活动人士却难以置信地希望ALPR不复存在......(1/2)
media 0 共 1 项
#550 🔗 原推
2026-09-23 20:34:59
Garry Tan
RT @war24182236: This is Supervisor Connie Chan’s neighborhood. Shopkeeper John Tewolde got beaten to death with a bat at Richmond Market trying to stop a guy stealing beer. She spent years talking about everything except more cops on the street. Now she wants to go to Congress. Watch this. @conniechansf
中文: RT @war24182236:这是主管康妮·陈的社区。店主约翰·特沃尔德在里士满市场被殴打致死，试图阻止一名男子偷窃啤酒。她花了数年时间谈论所有事情，除了街上更多的警察。现在她想去国会。看这个。 @conniechansf
🎬
视频
#549 🔗 原推
2026-09-23 20:08:04
Garry Tan
I can legit say Capy out of box does far far better parallelization, workflow, and automatic coordination than almost anything I've seen so far https://twitter.com/garrytan/status/2102836168157192586/photo/1
中文: 我可以合法地说，与迄今为止我所见过的几乎任何内容相比，Capy 的外框功能在并行化、工作流程和自动协调性方面都要好得多
media 0 共 1 项
#548 🔗 原推
2026-09-23 19:03:05
Garry Tan
RT @bradgessler: I'm going to be doing AMA about GBrain on Product Hunt for the rest of the day if you want to know more about what @garrytan, @Jmerc3030, &amp; I have been cooking up. Head on over there, ask me a question, and get your $100 off Hosted GBrain for teams. https://www.producthunt.com/products/gbrain
中文: RT @bradgessler:如果你想了解更多关于@garrytan、@Jmerc3030、&amp;我一直在烹饪的内容，那么当天剩下的时间，我将在《GBRain 上的产品狩猎》进行一场关于AMA的活动。 转到那边，问我一个问题，然后为团队提供100美元的托管GBrain折扣。
#547 🔗 原推
2026-09-23 18:59:49
Garry Tan
I've been using @capydotai daily on all my GStack/GBrain PRs and I'm going at least 4x faster than I was when just using raw Codex/Claude Code via Conductor. On a recent PR, this is what it did that out of box Codex wouldn't have done: https://twitter.com/garrytan/status/2102834280980459694/photo/1
中文: 我每天在GStack/GBrain的所有公关平台上使用@capydotai，而且只要通过 Conductor 使用原始的 Codex/Claude Code，我的速度至少会比原来快 4 倍。 在最近的一次公关中，这正是它在Codex无法做到的排装下:
media 0 共 1 项
#546 🔗 原推
2026-09-23 18:55:36
Garry Tan
RT @javaeeeee1: GBrain: Plug your own memory, tools, and skills into any AI by @bradgessler and @garrytan https://www.producthunt.com/products/gbrain?utm_source=twitter&utm_medium=social
#545 🔗 原推
2026-09-23 13:45:44
Garry Tan
RT @fchollet: The tweet below got enormous pushback at the time I posted it, more than 5 years ago. But this prediction is looking obvious by the day now.
中文: RT @fchollet:下面这条推文在我发布时，超过五年前就遭到了巨大的抵制。但这一预测目前看起来显而易见。
#544 🔗 原推
2026-09-23 13:39:58
Garry Tan
The ceiling on what we can do is so much higher now!
中文: 我们现在能做的极限已经高得多了!
#543 🔗 原推
2026-09-23 12:49:07
Garry Tan
RT @PratapRanade: Can the US build a DJI? We tore down @DJIGlobal 's 2019 and 2023 Mini drones and priced every part. It's the first public cost breakdown of either. Here's what we learned: DJI earns SaaS-like margins on hardware: ~79% gross margin on a $139 parts BOM. For comparison, Apple makes 53% on the iPhone. We estimate US-sourced parts would cost 2-3x more, cutting margins to 37-58%. The result: US drone makers can't compete in consumer, and focus on premium enterprise and defense. Sovereign silicon: DJI is building its autonomy lead starting at the chip. In 2019, the drone's key chips were American and Dutch. By 2023, all four were DJI's own. Custom silicon jumped from 1% to 26% of the parts bill. Nobody else can buy these chips. One notable exception is the GPS chip, still from Swiss firm @ublox . Chinese alternatives exist, but DJI's flight software is built around u-blox. In modern hardware, lock-in lives in the software. "Cheap China" is an outdated perception. DJI paid more for lighter Chinese batteries, then spent the saved grams on sensors and compute. Their first priority is capability per gram, not cost. Better battery tech → lighter weight → more onboard autonomy. US-made @SkydioHQ has better autonomy, but needs a heavier drone and a $399 chip to get it. We lost the war on cost. The next battlefield is autonomy. DJI's real moat is speed. Shenzhen gets new circuit boards in hours, the US in weeks. "Move fast and break things" works for hardware in China, not America. Banning DJI doesn't close that gap. Faster iteration does. We ran this teardown with our FDEs and Atlas, our engineering agent platform. See our full analysis: https://www.arenaphysica.com/publications/dji-teardown Thanks to @JonathonPSine @AkhilIyer1, Maggie Gray and Shaun Donovan for reading drafts of this.
中文: RT @PratapRanade:美国能否建立一家大疆? 我们拆除了@DJIGlobal的2019年和2023年迷你无人机，并对每个部件进行了定价。这是两者首次公开成本细分。我们了解到的如下: DJI在硬件上获得类似SaaS的利润率:在139个零件BOM上毛利率约为79%。相比之下，苹果在iPhone上的年率为53%。我们估计，美国采购的零部件价格将比其多花费2至3倍，利润率将降至37至58%。 结果:美国无人机制造商无法在消费领域竞争，而专注于高端企业和国防领域。 主权硅:大疆正在从芯片开始建立自主领先优势。2019年，该无人机的关键芯片是美国和荷兰的。到2023年，这四人都是大疆自己的。 定制硅片占零部件账单的1%，跃升至26%。没人能买到这些芯片。 一个显著的例外是GPS芯片，该芯片仍来自瑞士公司@ublox。中国有其他选择，但大疆的飞行软件是围绕u-blox打造的。在现代硬件中，锁定功能在软件中。 廉价中国是一种过时的观念。大疆为更轻的中国电池支付了更多费用，随后将节省下来的数克资金用于传感器和计算。他们的首要任务是每克能力，而不是成本。 更好的电池技术 → 更轻的重量 → 更多的板载自主性。美国制造的 @SkydioHQ 具有更好的自主性，但需要更重的无人机和一块 399 美元的芯片才能获得。我们因成本而输掉了战争。下一个战场是自治。 大疆真正的护城河是速度。深圳数小时内新增电路板，美国在数周内。快速移动和突破事物对中国的硬件是有效的，而不是美国。 禁止大疆不会缩小这一差距。更快的迭代。我们使用我们的工程代理平台FDE和Atlas进行了这次拆解。 请参阅我们的完整分析: 感谢@JonathonPSine @AkhilIyer1，Maggie Gray 和 Shaun Donovan 阅读了此内容的草稿。
#542 🔗 原推
2026-09-23 12:45:42
Garry Tan
Quite frankly one of the most cracked team of product engineers in SF today
中文: 坦率地说，今天旧金山最具裂纹的产品工程师团队之一
#541 🔗 原推
2026-09-23 12:43:05
Garry Tan
RT @garryslist: The CA Legislative Black Caucus condemns the "reign of terror" of SEIU union boss Dave Regan, whom recent reports allege has a pattern of "bullying, threats and physical intimidation directed disproportionately at women." https://twitter.com/garryslist/status/2102463382142189583/photo/1
中文: RT @garryslist:加州立法黑人核心小组谴责SEIU工会负责人戴夫·里根的“恐怖统治”。最近有报道称，该组织存在“针对女性的欺凌、威胁和人身恐吓”的模式。
media 0 共 1 项
#540 🔗 原推
2026-09-23 12:42:07
Garry Tan
RT @paulg: A CS undergrad asked me where he could have most effect in the AI age. I said probably at either extreme: either close to the technology, actually making LLMs, or close to the customer, using AI to give them exactly what they want. Or maybe both if you can stretch that far.
中文: RT @paulg:一位CS本科生问我，在人工智能时代，他能在哪些方面产生最大影响。我可能在极端情况下说过:要么接近技术，要么真正制造LLM，要么接近客户，使用人工智能来完全提供他们想要的东西。或者两者兼有，如果你能把那段路延伸。
#539 🔗 原推
2026-09-23 12:41:32
Garry Tan
RT @ycombinator: The YC Early Access Network is filling up fast. On Oct 22, tech leaders from high-growth startups to the Fortune 500 meet the best companies building enterprise AI. Be in the room. Request one of the remaining spots in Early Access now. https://events.ycombinator.com/yc-early-access-fall-26
中文: RT @ycombinator:YC 早用接入网络正在快速充斥。 10月22日，从高增长初创企业到财富500强的科技领导者，与企业人工智能建设的最佳公司会面。 待在房间里。立即申请“早期获取”中剩余的一个点。
#538 🔗 原推
2026-09-23 03:58:02
Garry Tan
RT @garryslist: Where the leaks? 👀 Find them using SCOUT, our City Hall monitor that keeps eyes on public meetings and delivers receipts straight to you. Find it exclusively for members on our updated website 👉https://garryslist.org/ https://twitter.com/garryslist/status/2102575240895496341/video/1
🎬
视频
#537 🔗 原推
2026-09-23 02:18:14
Garry Tan
RT @kane: For multiple years, @UCSF med school paid Dante King, a man with no genetics background, to give a lecture to doctors on how all white people were psychopaths because of genetics. They deleted the links to the lectures after I submitted public records requests for the invoices. https://twitter.com/kane/status/2102562081279238486/photo/1
中文: RT @kane:多年来，@UCSF 医学院一直向没有遗传背景的男子但丁·金发表演讲，向医生讲解所有白人因遗传原因如何都是精神病患者。 在我提交公开记录发票申请后，他们删除了讲座链接。
media 0 共 2 项 media 1
#536 🔗 原推
2026-09-23 01:21:34
Garry Tan
RT @lordspline: garry is agentmaxxing harder than most people we know all while running yc kinda insane
中文: RT @lordspline: garry 比我们认识的大多数人都更加严厉 运行时 有点疯狂
#535 🔗 原推
2026-09-23 01:20:43
Garry Tan
Capy (@capydotai) lets me drop PRs much much faster than I would with Codex or Claude Code alone https://twitter.com/garrytan/status/2102544711647129902/photo/1
中文: Capy(@capydotai)让我比使用Codex或Claude Code的速度快得多
media 0 共 1 项
#534 🔗 原推
2026-09-22 23:44:57
Garry Tan
RT @garryslist: Dave Regan, the SEIU union boss behind the CA wealth tax, is facing a RICO lawsuit from community health centers alleging that he has consistently weaponized ballot measures to "extort" unionization from them. https://twitter.com/garryslist/status/2102460704888586490/photo/1
中文: RT @garryslist:CA财富税的SEIU工会负责人戴夫·里根正面临来自社区卫生中心的RICO诉讼，指控他一贯将投票措施武器化，以“勒索”工会。
media 0 共 1 项
#533 🔗 原推
2026-09-22 23:28:08
Garry Tan
RT @tanayj: YC has funded more than a dozen data and RL environment startups in the last two years that are doing &gt;$10M+ per year (and many &gt;$100M per year) in revenue selling to the labs. (via @garrytan on YC's Lightcone podcast)
中文: RT @tanayj:YC 在过去两年中资助了十多家数据和 RL 环境初创企业，这些初创企业每年在向实验室销售的收入达到 &gt;$$1000M+ (以及每年许多 &gt;$$ 1亿美元)。 (通过 @garrytan 获取 YC 的 Lightcone 播客
#532 🔗 原推
2026-09-22 23:16:51
Garry Tan
If you're working on something cool, GStack does actually tell you to apply to YC :-)
中文: 如果你正在做一些很酷的事情，GStack 确实会告诉你要申请 YC 的 :-)
#531 🔗 原推
2026-09-22 23:15:51
Garry Tan
We need to teach the world to prompt and maximally use AI: to help all people see the way it can give you wings in all your pursuits Then we need yearn to solve more problems for ourselves and for one another
中文: 我们需要教导世界，促使并最大限度地运用人工智能:帮助所有人看到它在你所有追求中赋予你翅膀的方式 然后我们需要向时而向自己、共同解决更多问题
#530 🔗 原推
2026-09-22 20:53:28
Garry Tan
RT @qm__dev: QM v0.1.12 is out. Agent fleets in alpha Easier Slack setup Better app editing and publishing Plus reliability fixes and more. https://github.com/yc-software/qm/releases/tag/v0.1.12 https://twitter.com/qm__dev/status/2102221748720775530/photo/1
media 0 共 1 项
#529 🔗 原推
2026-09-22 16:17:16
Garry Tan
RT @MaxJunestrand: Getting from $1M to $100M in ARR took us 18 months. The second hundred took less than six. Last week @WeAreLegora passed $200M. 130,000 lawyers now use Legora on a monthly basis, across 2,100 firms and legal teams in more than 80 countries. Two things from this year I would not have predicted: The first is how fast the US moved. 18 months ago we had no office in North America. Today over half of the AmLaw 50 work with us, out of our offices in New York, Denver, Chicago, Houston, San Francisco and Toronto. The US is our largest market by revenue. The second is where the demand is coming from. More than 40% of our new customers are now in-house legal teams, including Air Canada, Salesforce and Palo Alto Networks. Two thank-yous, before we get back to work. To our customers, who backed a small company from Stockholm with a lot to prove, and who still tell us plainly when we get something wrong. And to the team, engineers and lawyers sitting next to each other, who turn that into something better, faster than any group I have worked with.
中文: RT @MaxJunestrand:从100万美元到1亿美元的ARR，我们花了18个月的时间。第二百只不到六。上周，@WeAreLegora 突破了2亿美元。 目前，目前有13万名律师每月在80多个国家的2100家律师事务所和法律团队中使用乐高拉。 今年有两件事我本无法预料: 第一是美国的发展速度。18个月前，我们在北美没有办事处。如今，超过一半的美国法律50机构与我们合作，在纽约、丹佛、芝加哥、休斯顿、旧金山和多伦多的办公室工作。美国按收入计算是我们最大的市场。 第二是需求的来源。目前，我们新客户中超过40%是内部法律团队，包括加拿大航空、Salesforce和帕洛阿尔托网络公司。 两个谢谢，我们才回去工作。致我们的客户，他们从斯德哥尔摩为一家小公司提供支持，并在我们出现问题时仍然清楚地告诉我们。对团队来说，工程师和律师们彼此坐在一起，他们把这种事情变成了比我合作过的任何团队都更好、更快的。
🎬
视频
#528 🔗 原推
2026-09-22 16:16:22
Garry Tan
RT @typesfast: What a glorious moment to own a business. You can take out all your costs and make all your products amazing. Just have to hope that your customers still have money and are not engaged in class revolution.
中文: RT @typesfast:拥有一家企业真是个辉煌的时刻。你可以花掉所有的成本，让所有产品都变得精彩。只需希望你的客户仍然有钱，并且没有参与阶级革命。
#527 🔗 原推
2026-09-22 12:58:56
Garry Tan
If you really think about it Cluely as a realtime thought helper and semi-adversarial assistant with ongoing context is still a good idea
中文: 如果你真的把它视为一个实时思维助手，而具有持续背景的半对抗助手，这仍然是个好主意
#526 🔗 原推
2026-09-22 03:07:00
Garry Tan
RT @kiwicopple: looking forward to getting on stage with @garrytan at Select 26 we’ll chat startups, AI, and how building a company is changing October 2 in SF https://select.supabase.com/ https://twitter.com/kiwicopple/status/2102220105266679836/video/1
中文: RT @kiwicopple:期待与@garrytan在Select 26上登台 我们将与初创企业、人工智能以及如何构建一家公司进行聊天 10月2日，旧金山
🎬
视频
#525 🔗 原推
2026-09-22 02:47:01
Garry Tan
RT @rastokke: An incredibly depressing article. A few points: 🔸If your kid can’t add fractions, they are not ready for algebra, even if the school is giving them A’s. It's an easy check. Make sure they can add, subtract, multiply, divide fractions. 🔸Lots of misinformed math education academics out there downplay the importance of foundational skills and tests. They are wrong. Note how her child developed test anxiety once she started doing poorly on tests b/c she had not acquired the knowledge to do well on the tests. 🔸Standardized test results probably do give a pretty accurate picture of how a school is performing in comparison to other schools. Testing matters, for so many reasons. 🔸High standards matter. Smart kids need to be challenged. Even her own daughter could see this. I appreciate that the journalist shared her story and I hope other parents learn from the mistake she made. https://www.nytimes.com/2026/09/20/magazine/ny-public-schools-black-students-segregation.html?unlocked_article_code=1.C1E.6nib.cLKLM5CRObVI&smid=url-share
中文: RT @rastokke:一篇极其令人沮丧的文章。几点: 如果你的孩子无法增加分数，即使学校给了他们A的，他们也无法准备好学习代数。检查很简单。确保它们能够添加、减去、乘法、分割分数。 很多被误导的数学教育学者都淡化了基础技能和考试的重要性。他们错了。请注意，当她孩子开始进行不良检测后，如何开始出现检测焦虑症，而她并未获得在检测中表现良好的知识。 🔸 标准化的测试结果可能准确地反映了学校与其他学校相比的表现情况。测试很重要，原因有很多。 🔸 高标准很重要。聪明的孩子需要受到挑战。就连她自己的女儿也能看到这一点。 我很感激这位记者分享了她的故事，希望其他父母能从她所犯的错误中吸取教训。
#524 🔗 原推
2026-09-22 02:44:55
Garry Tan
RT @illscience: After 20 years of being intermediated by Google every consumer internet company is now urgently trying to figure out how to be on the right side of this new traffic for margin trade. There is a seismic shift coming as agents quickly become the primary consumers on the Internet. https://twitter.com/illscience/status/2102147274021183884/photo/1
中文: RT @illscience:在被谷歌调中20年后，每家消费类互联网公司现在都在紧急寻找如何在这一新流量中处于正确的一边进行保证金交易。 随着代理商迅速成为互联网上的主要消费者，人们正在发生巨大变化。
media 0 共 2 项 media 1
#523 🔗 原推
2026-09-21 23:04:54
Garry Tan
RT @jayair: After working on it for a year, I can finally see the end state of the harness On the backend, it's an execution environment that gives the model maximum degrees of freedom and on the frontend it's a free-form canvas that the model can use to communicate with the user
中文: RT @jayair:经过一年的处理，我终于看到了安全带的终结状态 在后端，它是一个执行环境，可赋予模型最大的自由度，而前端则是一个自由格式的画布，模型可用于与用户进行通信
#522 🔗 原推
2026-09-21 21:23:31
Garry Tan
RT @0xluffy: on DeepSWE capy is better, faster AND cheaper than any harnesses you guys use blog on this 🔜 https://twitter.com/0xluffy/status/2102128216551165979/photo/1
中文: RT @0xluffy:在 DeepSWE 上，你使用的安全带都比你使用的任何设备都更好、更快、更便宜 博客对此🔜
media 0 共 1 项
#521 🔗 原推
2026-09-21 21:17:48
Garry Tan
RT @0xluffy: just did an eval on capy vs one of the big harnesses out there capy did better with half the cost and half the time
中文: RT @0xluffy:刚在 poyy 上与其中一条大线具进行了一次椭圆形 佩里用一半的成本和一半的时间做得更好
#520 🔗 原推
2026-09-21 21:17:35
Garry Tan
RT @zachweinberg: Holy shit. It's full on private school level funding for zero actual education. We are failing our kids by pretending bad teachers don't exits. We are failing great parents by pretending bad parenting doesn't matter. We have to change the culture and reward excellence.
中文: RT @zachweinberg:真是太了不对。为实际教育提供全部私立学校资金。 我们通过假装糟糕的老师不离开，辜负了我们的孩子。我们让优秀的父母失望，假装糟糕的育儿方式无关紧要。 我们必须改变文化，奖励卓越。
#519 🔗 原推
2026-09-21 21:11:16
Garry Tan
RT @_aj: We built a new harness using @typesafeai's Jev that cuts the cost of repetitive work by 90%. The harness learns the job as it runs, moving steps from LLM calls to code. Running 100,000 compliance alerts costs &gt;$290K on Opus 5. With agentrun() we got it down to &lt;$26K. https://twitter.com/_aj/status/2102061534956662818/photo/1
中文: RT @_aj:我们使用 @typesafeai 的 Jev 构建了一种新线束，可将重复性工作的成本降低 90%。线束在运行时会学习工作，从LLM调用到代码都会移动步骤。 运行10万条合规提醒成本 &gt; 2.9万美元 使用 agentrun()，我们将其归结为 &lt;$26K。
media 0 共 1 项
#518 🔗 原推
2026-09-21 19:20:52
Garry Tan
Example PR - here is https://capy.ai/ executing against a very ambitious bug fix wave on GBrain with very clear task delineation, automatic parallelization, and very clean Github PR and CI workflow. https://twitter.com/garrytan/status/2102096495847551011/photo/1
中文: 示例公关——以下是 在 GBrain 上针对一个雄心勃勃的错误修复波进行执行，具有非常清晰的任务划分、自动并行化以及非常简洁的 Github PR 和 CI 工作流程。
media 0 共 1 项
#517 🔗 原推
2026-09-21 18:03:54
Garry Tan
Not really sure how @capydotai does it, but it really is able to track multi-step workflow and do large PRs faster than Codex or Claude Code on its own. My most favorite new agentic coding secret weapon in my arsenal the last week.
中文: 不确定@capydotai 是如何做到的，但它确实能够追踪多步工作流程，并且比 Codex 或 Claude Code 本身能更快地完成大型 PR 操作。上周我武器库中最喜欢的新型特制编码秘密武器。
#516 🔗 原推
2026-09-21 18:01:38
Garry Tan
I mean the data center panic is absurd and unnecessary but I can’t argue with the outcome if true
中文: 我的意思是，数据中心的恐慌是荒谬且不必要的，但如果结果属实，我无法反驳
#515 🔗 原推
2026-09-21 15:14:12
Garry Tan
RT @sriramk: My agents should be able to do anything I can on a computer or a phone ( and I should be held accountable for their behavior).
中文: RT @sriramk:我的经纪人应该能够在电脑或手机上做任何我能做的事情(我也应该为他们的行为负责)。
#514 🔗 原推
2026-09-21 15:13:37
Garry Tan
RT @harjtaggar: It became easier for hardware companies to raise money this year partly because investors were afraid AI would eat software. Imagine how excited they'll be if AI makes hardware cheaper and faster to build.
#513 🔗 原推
2026-09-21 12:41:17
Garry Tan
RT @MaudMaron: I have some thoughts! Nikole Hannah-Jones interviewed me at the height of woke-1 to champion and virtue signal her “fight for every black child” and disparage and dismiss my fight for all NYC public school students to enjoy the rising tide that lifts all boats of academic rigor and meritocracy. She thought my support of the Specialized High School Admission Exam, the SHSAT, was racist. I thought her school politics were deeply misguided, and well, quite racist. Unlike Hannah-Jones I have nothing to walk back. I knew then what she has just figured out that her identitarian race-based grievance project would not fix anything. And in fact, it has hurt all public school students. Not just in NYC—her Magnum Opus of racial grievance, The 1619 Project, is taught in schools across the country and its reasoning is used to dismantle and stop the creation of the kind of high school that her daughter ultimately chose—the one that finally taught her algebra. NYC Public School students score lower on all academic metrics now than they did when her daughter started school—and we tax payers pay more for it. Enrollment continues to plummet while those in charge dumb down standards and hide results. So much of that is because any solution that challenged Hannah-Jones' race-baiting—endorsed by The Times and Democratic leaders alike—was verboten. Like Hannah-Jones I have a HS junior. My two older kids have graduated from NYCPS and are in college and my youngest is in NYCPS elementary school. I have qualms with parts of their education but also many success stories. Can you get a great education in NYCPS? Yes. But you have to fight for it. And insisting on the schools that grade math homework is a start. Knowing what your child is learning and mastering is a non-negotiable. Standardized tests help with that, but are a starting point only. Blaming phantom racists instead of understanding the role parents play was and is a lazy cop out. There is still an effort–alive and well right now-to dismantle great schools like NYC’s specialized high schools because they use standardized admissions tests. The effort still relies on the same failed logic that Hannah-Jones has peddled for years. She insisted that kids like her daughter who did not score high enough on a test to gain admission were the victims of racism. But she refused to acknowledge that poor Asian kids whose parents had NONE of the resources that her daughter enjoyed (or should have enjoyed) (English speaking parents, resources to pay for prep classes, knowledge of how the system works, etc) were not racist oppressors—but kids willing to work hard and apply their smarts with parents that prioritized subject mastery and took the time to make sure their kids were actually learning. Her daughter is right. She deserved much better. But it is not the cauldron of racist grievances that Hannah-Jones still cannot let go of that failed her daughter. It’s her daughter’s parents that did not figure out for NINE YEARS that their daughter was not learning enough. New York City Public Schools should be among the best in the world because we have so many incredibly talented, smart and hard-working students that could and should learn more by being challenged more–and that means sorting kids into cohorts where they can rise to the level of their abilities and willingness to work. In other words “screening”—the exact project Hannah-Jones fought against for so long and which she has now relied on to rescue her daughter’s education. You’re welcome Nikole. Your daughter’s chosen high school would not exist but for parents like me who fought so hard to keep screened schools.
#512 🔗 原推
2026-09-21 12:40:54
Garry Tan
RT @timurkuran: The United States has been leaping from one moral panic to the next: Black Lives Matter, MeToo, George Floyd, and now data centers plus the specter of AI-driven human extinction. In every case a genuine problem exists, but the threat is inflated out of all proportion and the costs of overreaction are overlooked. The resulting hysteria blocks clear thinking and closes off rational discussion. Consequently, individuals never notice that self-appointed experts are distorting the facts for their own gain. Companies with billions in AI stakes are perfectly capable of staging “rogue LLM” incidents to stampede politicians into regulation that keeps out competitors or simply reduces their own liability. Previous panics have already damaged trust in institutions and among ourselves. They have also intensified the political polarization that now endangers American democracy. We should resist the current panic rather than feed it. The costs of getting this one wrong dwarf those of the last several.
中文: RT @timurkuran:美国从一种道德恐慌跃升至下一个:黑人的命也是命、“梅多”、乔治·弗洛伊德，如今又是数据中心，以及人工智能驱动型人类灭绝的幽灵。在每种情况下，都存在真正的问题，但威胁被夸大得出所有比例，而过度反应的代价却被忽视了。由此产生的歇斯底里会阻碍清晰的思维，并导致理性讨论的结束。因此，个人从未注意到，自封的专家是在歪曲事实，以谋取自身利益。 拥有数十亿美元人工智能股份的公司完全有能力将“流氓传销”事件压入监管，从而排除竞争对手或仅仅减少自身责任。 先前的恐慌已经损害了对机构和我们自身之间的信任。他们还加剧了如今危及美国民主的政治两极分化。我们应该抵制当前的恐慌，而不是助长恐慌。弄错这个的代价比上几个要好。
#511 🔗 原推
2026-09-20 12:14:11
Garry Tan
RT @simonw: Being a computer scientist who refuses to find anything about LLMs interesting right now is a bit like being a geneticist who refuses to find anything interesting about the recently opened Jurassic Park
中文: RT @simonw:作为一名计算机科学家，现在拒绝了解一些有趣的LLM，这有点像是一位遗传学家，拒绝为最近开放的侏罗纪公园找到任何有趣的内容
#510 🔗 原推
2026-09-20 04:05:17
Garry Tan
RT @Dale_Chu: They don't just dislike standardized tests. They hold student preparation—and academic achievement—in complete contempt. https://twitter.com/minilek/status/2101009192807248169/video/1
🎬
视频
#509 🔗 原推
2026-09-20 04:00:52
Garry Tan
RT @neetu_arnold: This is cool to see. A UC Regents member is changing his mind on the value of standardized tests: “Last time around, I voted to get rid of the SAT, &amp; I’m starting to believe I might have made a mistake” https://twitter.com/neetu_arnold/status/2101082068994376175/video/1
中文: RT @neetu_arnold:这很酷。加州大学摄政大学的一位成员正在改变他对标准化考试价值的想法: 上一次，我投票决定取消SAT、&amp;我开始相信自己可能犯了错误。
🎬
视频
#508 🔗 原推
2026-09-20 04:00:42
Garry Tan
RT @minilek: The University of California Board of Regents heard presentations on the SAT on Wednesday, specifically on whether it should be re-introduced into freshman admissions. I found the below excerpt from one presentation illuminating: "the statistical power of tests in predicting outcomes does not address whether UC ought to consider expected college success when deciding which students to admit, or to what extent we should" So, it's clear the debate is not centered on whether the SAT predicts college success -- the presenter says so himself. The heart of the debate is this: a UC education is a path to upward mobility, and the anti-SAT faction believe that this golden ticket should be given to those who would benefit from it the most, especially those who they feel have been harmed by society in the past as a form of restorative justice. If you hold this viewpoint, the increased failure rates in intro math classes and the surge in student unpreparedness are irrelevant.
中文: RT @minilek:加州大学董事会周三就SAT考试听取了专题介绍，特别介绍了是否应将其重新引入大一招生。 我找到了下面一个演示文稿的摘录，说明: 测试在预测结果方面的统计能力，并未涉及加州大学在决定录取哪些学生时，是否应考虑预期的大学学业成功与否，也不涉及我们该在多大程度上录取。 因此，很明显，这场争论并非围绕SAT是否预测大学成功展开——主持人自己说道。争论的核心是:加州大学教育是一条向上流动的道路，反SAT派系认为，这种金票应该送给那些最能从中受益的人，尤其是那些他们觉得过去曾因社会而受到伤害的人，这是一种恢复性正义。 如果你持有这种观点，介绍数学课程中失败率的提高以及学生准备不足的激增是无关紧要的。
🎬
视频
#507 🔗 原推
2026-09-20 03:30:18
Garry Tan
RT @MHurabiell: Did we not learn anything from the Chesa-experiment? Need to hold these kids accountable or they will continue this violent behavior. This behavior hurts all students. @SFUSD_Supe @sfstandard @sfchronicle @TheVOSF
中文: RT @MHurabiell:我们难道没有从切萨实验中学到什么吗? 必须追究这些孩子的责任，否则他们将继续这种行为。 这种行为伤害了所有学生。 @SFUSD_Supe @sfstadard @sfchronicle @TheVOSF
#506 🔗 原推
2026-09-20 03:29:01
Garry Tan
RT @Twolfrecovery: There's a hypocrisy of progressive leadership in West Coast cities. They cry affordability yet don't build housing while also saying homelessness is caused by high rents. The truth? They know homelessness is drug and mental illness problem and their solution is to normalize it. https://twitter.com/Twolfrecovery/status/2100970081207631918/video/1
中文: RT @Twolfrecovery:西海岸城市存在进步领导力的虚伪。他们哭喊着负担可负担性，却又不建造住房，同时又说无家可归是由高租金造成的。真相?他们知道无家可归是毒品和精神疾病问题，而他们的解决办法是使其恢复正常。
🎬
视频
#505 🔗 原推
2026-09-20 03:17:19
Garry Tan
RT @altcap: Vote “NO” on Dave Regan &amp; his pet Prop 40. Dave is all about Dave - that’s why he is under investigation &amp; the police, teachers, planned parenthood, building trades &amp; many others all oppose Dave &amp; Prop 40. The evidence is overwhelming it will badly hurt CA! 🇺🇸🧐
中文: RT @altcap:对戴夫·里根和他的宠物40号提案投反对票戴夫全是关于戴夫的事——这就是他接受调查的原因;警察、教师、计划生育、建筑行业;还有许多人反对戴夫和潘普;第40号提案。证据确凿，这将严重损害加州!🇺🇸🧐
#504 🔗 原推
2026-09-20 02:47:51
Garry Tan
Stop racial discrimination in the University of California system
中文: 制止加州大学体系中的种族歧视
#503 🔗 原推
2026-09-19 23:39:56
Garry Tan
RT @paulg: Remember how eagerly the press swallowed stories about how Twitter was going to crash after Elon bought it? How could they have believed that someone who could run rocket and car companies couldn't run a forum? https://twitter.com/paulg/status/2101256525519978542/photo/1
中文: RT @paulg:还记得埃隆买下推特后，推特将如何崩溃的报道，这是多么令人期待吗?他们怎么可能相信，一个能够运营火箭和汽车公司的人，都无法举办论坛?
media 0 共 1 项
#502 🔗 原推
2026-09-19 22:11:14
Garry Tan
RT @vivianmshen: Some incredible stats from the @ycombinator S26 batch: ATOMS ARE IN: hardtech cos went from 8% → 20% of the batch robotics 1% → 6% industrial manufacturing 4% → 10% defense 1% → 5% compute stack (eg. photonics) 1% → 4% power infrastructure ~1% → 3%
中文: RT @vivianmshen:来自 @ycombinator S26 的一些精彩数据: ATOMS 在: 硬技术公司占批次的8%→20% 机器人 1% → 6% 工业制造 4% → 10% 防御 1% → 5% 计算栈(例如光子学)1% → 4% 电力基础设施 ~1% → 3%
#501 🔗 原推
2026-09-19 22:11:02
Garry Tan
RT @Twolfrecovery: Here's the union guy pushing the billionaire tax in California. "The lawsuit alleged that he “instructed SEIU agents to tell Hispanic home care workers that if they did not vote for SEIU, ICE would be called and the workers would be deported.”  https://www.sfchronicle.com/politics/article/dave-regan-allegations-prop-40-22433169.php
中文: RT @Twolfrecovery:这位工会成员正在加州推动征收亿万富翁税。 诉讼指控他“指示SEIU工作人员告知西班牙裔家庭护理工作者，如果他们不投票支持SEIU，将联系移民与海关执法联盟(SEIU)，并将驱逐这些工作人员。” 
#500 🔗 原推
2026-09-19 20:26:46
Garry Tan
RT @MiguelriosEN: Today I’m excited to show you our new harness called AgentRun, built with @pidotdev and @typesafeai's Jev. It’s built for an agent to learn how to do a job, code itself a general solution, and then get out of the way. https://twitter.com/MiguelriosEN/status/2101033282414768456/video/1
中文: RT @MiguelriosEN:今天，我很期待向您展示我们全新的名为AgentRun的线束，该线束由@pidotdev和@typesafeai's Jev共同打造。 它是为代理人学习如何完成工作、编写通用解决方案并退出而构建的。
🎬
视频
#499 🔗 原推
2026-09-18 20:28:42
Garry Tan
RT @praggy: @HermesWatcher Gbrain by @garrytan You people have no idea what you’ll are missing out on. I can run gbrain locally as well. Local memory + llm wiki + gbrain works wonders
中文: RT @praggy:@HermesWatcher Gbrain 由 @garrytan 制作 你们不知道你们会错过什么。我也可以在本地运行大脑。 本地内存 + lm wiki + gbrain 效果很好
#498 🔗 原推
2026-09-18 12:06:18
Garry Tan
RT @garryslist: She was the only one in the room. Her kids never will be. A generation ago, Rowena grew up in a UK where no one looked like her. Today she's raising her kids in a California that made room — where the people once talked over finally get to speak up. That's the California we're fighting for.
🎬
视频
#497 🔗 原推
2026-09-18 12:05:21
Garry Tan
RT @banteg: >be me >discover effective altruism >apparently normal charity is inefficient >why donate to random sad thing when spreadsheet can tell you optimal sad thing >fair enough >buy mosquito nets >save lives >numbers look good >feel powerful >couple years later >someone asks an innocent question >why only count people alive today >huh >future people matter too >obviously >my grandchildren shouldn't matter less just because they haven't spawned yet >reasonable.jpg >keep following logic >what about their grandchildren >also yes >what about people in 500 years >sure >5000 years >why not >500 million years >starting to get weird but morality is morality >open calculator >humanity could survive for an astronomically long time >could colonize galaxy >could have trillions upon trillions of descendants >maybe digital people too >maybe simulated civilizations >maybe dyson spheres full of happy uploaded minds >calculator starts smoking >realize currently living humans are rounding error >8 billion people suddenly looking extremely beta >future contains potentially 10^something people >can't even fit beneficiaries in google sheets >new moral priority unlocked >protect the long-term future >stop thinking in units of "people helped" >start thinking in "fraction of cosmic endowment preserved" >malaria? >terrible >but only kills existing humans >AI extinction could delete the entire light cone >nuclear war could permanently derail civilization >bad institutions could lock in terrible values for ten million years >someone invents wrong constitution in 2140 >quadrillions suffer >better fund governance workshop now >friend says maybe we should improve hospitals >explain opportunity cost >friend says hospitals are full of actual sick people >explain scope sensitivity >friend stops inviting me to dinner >need to decide what to fund >easy >expected value >suppose project has one in a million chance of preventing extinction >sounds tiny >but extinction destroys 10^50 future lives >multiply >mother of god >$10 million project has expected value of several galaxies >charity evaluation complete >someone asks where the one-in-a-million number came from >expert judgement >which expert >us >how calibrated >extremely thoughtfully >reduce estimate to one in ten million to be conservative >still beats curing cancer by 38 orders of magnitude >epistemic robustness achieved >someone says maybe project doesn't work >assign 20% chance >still astronomical >maybe project makes problem worse >assign 5% chance >still astronomical >why 5 >because 30 felt pessimistic >publish 46-page report >contains seventeen sensitivity analyses >every sensitivity analysis begins after assuming intervention has positive sign >critic says you're multiplying enormous hypothetical stakes by extremely uncertain probabilities >yes >that's literally why it's important >critic says the uncertainty might be structural rather than numerical >make probability smaller >critic says no, I mean maybe your model is wrong >make probability smaller again >critic begins rubbing temples >discover AI safety >perfect longtermist cause >AI might kill everyone >or create utopia >or seize galaxy >or tile universe with paperclips >or create billions of conscious software minds >finally a problem with numbers big enough for me >start AI safety nonprofit >mission: prevent dangerous AI >hire smartest people available >smartest people immediately start building better AI to understand dangerous AI >interesting >we must understand capabilities to understand safety >we must scale models to study alignment >we must race ahead so less responsible actors don't get there first >we must deploy systems to learn how deployment can go wrong >we must build the thing quickly because building the thing quickly is dangerous >outsider asks why the people most worried about AI apocalypse all work at AI companies >complicated field >company releases stronger model >very concerned >company begins training even stronger model >extremely concerned >company raises $14 billion >concern reaches unprecedented levels >need to influence government >future is at stake >normal democratic process too slow >politicians don't understand exponential curves >public doesn't understand x-risk >experts must guide them >who counts as expert >people who understand x-risk >who understands x-risk >our friends >someone objects that this seems politically convenient >explain we're representing future generations >future generations unavailable for comment >develop concept of value lock-in >terrifying possibility that one ideology controls civilization forever >therefore extremely important that civilization adopts correct values before lock-in >whose values >let's circle back >begin with impartial morality >end with small group of people deciding what quadrillions of hypothetical beings would want >beautiful arc >meanwhile actual humans keep doing annoying things >voting wrong >having parochial attachments >loving family more than strangers >caring about local community >getting upset when told their suffering is cosmically negligible >evolutionary biases everywhere >explain that moral intuition cannot be trusted >except intuition that future digital people count >and intuition that extinction is uniquely bad >and intuition that our probability estimates are sane >and intuition that our institutional choices improve the future >those intuitions survived peer review >someone donates $5k to local homeless shelter >inefficient >could have funded 0.0000000000003% of an AI governance researcher >think of all the simulated people you just killed >okay maybe don't phrase it that way publicly >PR team says "future generations deserve a voice" >much better >journalist asks what longtermism means >say "future people matter" >everyone agrees >great >journalist asks what follows from that >well technically we should redirect enormous resources toward low-probability interventions affecting astronomical futures >journalist raises eyebrow >return to "future people matter" >motte has entered the chat >critic: of course future people matter >me: glad we agree >critic: I don't agree that your institute knows how to help them >me: why do you hate our grandchildren >eventually notice uncomfortable implication >if future value dominates everything >then helping people today mostly matters through effects on future >education matters because future institutions >health matters because future productivity >democracy matters because future trajectory >human beings slowly become instrumental variables in their own moral philosophy >see starving child >feel compassion >check spreadsheet >child's direct welfare contribution negligible >but perhaps childhood nutrition improves national institutional quality >compassion restored >tell myself this is impartial altruism >one day assistant asks obvious question >"how do you know your intervention actually improves the far future?" >silence >open spreadsheet >increase column width >add confidence interval >assistant asks again >"no, I mean how do you know the sign is positive?" >stare into cosmic light cone >10^50 people staring back >none of them exist >none of them can tell me >none of them can falsify my assumptions >realize I have invented the perfect constituency >infinitely important >completely silent >and always represented by me
中文: RT @banteg:> 成为我 发现有效的利他主义 显然，正常的慈善活动效率低下 为什么在电子表格能告诉你最理想的悲伤时，要为随机的悲伤事物捐款 够公平的 购买蚊帐 拯救生命 数字看起来不错 感觉强大 几年后的情侣 有人问一个无辜的问题 为什么今天只数活着的人 嗯 未来的人也很重要 显然 我的孙子们不应该因为尚未生生还那么重要 合理。jpg 继续遵循逻辑 他们的孙辈呢 也是的 500年后的人呢 当然 5000年 为什么不 5亿年 开始变得奇怪，但道德就是道德 开放式计算器 人类可以存活一个天文数字的长时间 可能殖民银河系 可能拥有数万亿的后代 也许也是数字人 也许是模拟文明 也许充满快乐的脑海 计算器开始吸烟 实现当前活着的人类存在四舍五入的误差 80亿人突然看起来极度贝塔 未来可能包含10个左右的人 甚至连谷歌表格中的受益人也无法适应 新的道德优先被解锁 保护长远未来 停止以“人帮助”的单位思考 开始思考“保存的宇宙禀赋的一小部分” 疟疾? 可怕的 但只会杀死现存的人类 人工智能灭绝可能会删除整个光锥 核战争可能永久破坏文明 不良机构可能在一亿年内锁定糟糕的价值观 有人在2140年发明了错误的宪法 万亿游击队受苦 改善基金治理研讨会 朋友说也许我们应该改善医院 解释机会成本 朋友说医院中满是真正的病人 解释范围敏感性 朋友不再邀请我吃晚饭 需要决定资助什么 简单 期望值 假设项目有百万万个中的概率防止灭绝 听起来很小 但灭绝会毁掉 10^50 未来的生活 乘数 上帝之母 耗资1000万美元的项目预计多个星系的价值 慈善评估完成 有人问，百万万里一这个数字是从哪里来的 专家判断 哪一位专家 我们 如何校准 非常贴心 将估算值降低至一万一以保持保守 治愈癌症的幅度仍超过38个数量级 达到的流行病性 有人说项目可能行不通 分配20%的机会 天文数字 也许项目会让问题变得更糟 分配5%的机会 天文数字 为什么5 因为30感觉很悲观 发布46页的报告 包含十七项敏感性分析 假设干预有积极迹象后，每一份敏感性分析都开始 批评称，你正将巨大的假设性因素乘以极不确定的概率 是的 这实际上就是它很重要的原因 批评者表示，这种不确定性可能是结构性的，而非数值 使概率变小 批评人说不，我的意思是，也许你的模型是错的 再次缩小概率 批评开始摩擦寺庙 发现人工智能安全 完美的长期主义事业 人工智能可能会杀死所有人 或创建乌托邦 或夺取银河系 或使用纸片的瓷砖世界 或创建数十亿个有意识的软件头脑 数字问题对我来说终于大了 启动人工智能安全非营利组织 任务:防止危险的人工智能 最聪明的人 最聪明的人立即开始构建更智能的人工智能，以理解危险的人工智能 有趣 我们必须了解安全的能力 我们必须对齐度进行建模 我们必须争先得，以免责任较小的演员先到达那里 我们必须部署系统来了解部署可能出错 我们必须迅速建造这个东西，因为快速建造这个东西很危险 outsider 会问，为什么最担心人工智能末日的人在人工智能公司工作 复杂字段 公司发布更强的模型 非常担忧 公司开始培训更强大的模型 极度担忧 公司融资140亿美元 关注度达到前所未有的水平 需要影响政府 未来岌岌可危 正常的民主进程太慢 政客们不懂指数曲线 公众不理解X-风险 专家必须引导他们 谁算作专家 了解X-风险的人 谁理解X-风险 我们的朋友们 某个物体似乎在政治上很方便 解释我们代表下一代
#496 🔗 原推
2026-09-18 12:01:05
Garry Tan
RT @sdianahu: Congrats to @snarkyzk ,@alexisgauba and @benhylak on @raindrop_ai's Series A led by CRV, bringing their total funding to $50M they detect failures in AI agents, such as tool call failures and decision errors (Vercel, Clay, and Speak uses them) 1/7 how they got started as a terminal and now solving the rise and fall of agent civilizations with Simulations
中文: RT @sdianahu:祝贺@snarkyzk、@alexisgauba 和 @benhylak 在 CRV 主导的 @raindrop_ai A 系列中，总资金达到 5000 万美元 他们检测人工智能代理中的故障，例如工具调用失败和决策错误(Vercel、Clay 和 Speak 使用它们) 1/7 他们是如何作为终端起步的，现在通过模拟方式解决代理文明的兴衰
#495 🔗 原推
2026-09-18 11:58:26
Garry Tan
Memorable found a way to optimize memory with embeddings instead of more tokens which is a powerful new way to do memory
中文: Memorable 找到了一种利用嵌入来优化内存的方法，而不是使用更多令牌，这是一种强大的新记忆方式
#494 🔗 原推
2026-09-17 19:29:31
Garry Tan
Be careful what you wish for This is why we need alignment to humankind vs any other goal
中文: 小心你所希望的 这就是为什么我们需要与人类团结，而不是其他目标
#493 🔗 原推
2026-09-17 17:22:10
Garry Tan
The future is already here We just have to choose it and spread it faster
中文: 未来已经到来 我们只需要选择它并更快地传播
#492 🔗 原推
2026-09-17 15:48:15
Garry Tan
I think every AI harness needs to support Tailscale Muse and Grok Bot do it out of box Codex and Claude Code cloud containers are missing this right now
中文: 我认为每个人工智能线束都需要支持Tailscale 缪斯和格罗克机器人会一蹴而就 目前，Codex 和 Claude Code 云容器都缺失了
#491 🔗 原推
2026-09-17 14:58:59
Garry Tan
RT @smatta1701: Effective Altruism is the natural enemy of the left and right because the right doesn't believe in altruism and the left doesn't believe in effectiveness.
中文: RT @smatta1701:有效的利他主义是左右派的天敌，因为右派不相信利他主义，而左派也不相信有效。
#490 🔗 原推
2026-09-17 14:55:23
Garry Tan
RT @AlecStapp: I thought the tweet below must be omitting some critical context or paraphrasing the text ungenerously. But no, the NYT actually published an op-ed that claims the University of California hasn't sacrificed excellence even though it admits students who struggle with basic math. https://twitter.com/AlecStapp/status/2100338173561737568/photo/1
media 0 共 1 项
#489 🔗 原推
2026-09-17 14:54:28
Garry Tan
RT @CoreyWriting: If students from UC schools are graduating despite not being able to do basic math, that only indicates that UC has dramatically dropped their academic standards to push students across the finish line. This is a terrible argument and it devalues the UC degree.
中文: RT @CoreyWriting:如果来自加州大学学校的学生尽管无法完成基础数学，但毕业时，这仅表明加州大学已大幅降低学业标准，迫使学生跨越终点线。 这是一个糟糕的论点，它贬低了加州大学的学位。
#488 🔗 原推
2026-09-17 04:05:07
Garry Tan
RT @Overlap_Tech: Sam Altman: This Is Like Going to Get Fire From the Gods⁣ ⁣ "We have only scratched the surface of figuring out how to build great products that really make people more capable, more powerful... this is like we went to go get fire from the gods." — @sama https://twitter.com/Overlap_Tech/status/2100415768395415635/video/1
🎬
视频
#487 🔗 原推
2026-09-17 04:02:47
Garry Tan
Grok bot broken for me, won’t reply no matter how I log in, X, Cursor, GitHub, all ways
#486 🔗 原推
2026-09-17 02:56:57
Garry Tan
RT @sfmcguire79: This op-ed can’t be real: “Students on University of California campuses may struggle with basic math, but they are graduating at record rates.” “What has made the University of California exceptional is its commitment to increase access without sacrificing excellence.” https://twitter.com/sfmcguire79/status/2100303907083731274/photo/1
中文: RT @sfmcguire79:这篇评论文章无法真实使用: 加州大学校园的学生可能难以理合基础数学，但他们的毕业率却创下历史新高。 “加州大学与众不同之处在于，它致力于在不牺牲卓越品质的前提下增加可获取性。”
media 0 共 1 项
#485 🔗 原推
2026-09-16 22:44:22
Garry Tan
Excited to speak at @Supabase Select in conversation with the amazing @kiwicopple about building startups in the time of AI Register for the event and join me: https://select.supabase.com/
中文: 很高兴在@Supabase Select上与出色的@kiwicopple谈论在人工智能时代创办初创企业 报名参加活动并与我同在:
#484 🔗 原推
2026-09-16 22:43:45
Garry Tan
RT @dankrieg: Introducing iHermes A personal AI assistant powered by Hermes Agent, with GBrain for memory. You talk to it in iMessage. We're huge fans of Hermes. We've been using it to build internal systems and automate work, and we wanted more people to get that kind of value without having to set up the whole thing themselves. Start with one text. Connect the apps you use, hand off a task, and let it work through the steps. It keeps useful context from your earlier work and follows up proactively. Our goal is to make iHermes the most transparent and secure personal assistant in iMessage, built around the person using it. Your assistant should work for you.
中文: RT @dankrieg:介绍 iHermes 由 Hermes Agent 提供的个人人工智能助手，GGBrain 用于记忆。你在iMessage中与它交谈。 我们是爱马仕的忠实粉丝。我们一直用它来构建内部系统并实现工作自动化，我们希望有更多人能够获得这种价值，而无需自己设置整个功能。 从一条文字开始。连接你使用的应用程序，完成任务，并让它在步骤中完成。它会从你之前的工作中保留有用的语境，并主动跟进。 我们的目标是让iHermes成为iMessage中最透明、最安全的个人助理，其建立在使用者的基础上。 你的助手应该为你工作。
🎬
视频
#483 🔗 原推
2026-09-16 21:41:59
Garry Tan
I can use any harness I want and it's the same personality and full memory of my personal AGI
中文: 我可以使用任何我想要的线束，它与我个人的AGI一样，充满个性和完整记忆
#482 🔗 原推
2026-09-16 21:41:37
Garry Tan
RT @judegomila: I've attempted to map everything in oncology in a public website and open source repo for all. The site is trying to get all cancers, products, technologies, bottlenecks, people, startup opportunities with 1000+ ideas for upgrading the field. If you have a loved one with cancer and you are technical or can engineer, go take a look, file improvement requests or bugs or help with the open repo and make this the best open and free info resource for individuals, researchers and educational use. This should save people time, aid AI oncology projects and generate positive action. If you are not technical just complain in this thread about broken or annoying or things you want and I'll fix them live. Some of the interesting pages: Treatments: https://onco.cc/drugs A gallery of the molecules being used https://onco.cc/molecules/ And targets: https://onco.cc/targets/ The technologies in oncology: https://onco.cc/technologies/ 1100 ideas for helping oncology: https://onco.cc/ideas/ Bottlenecks on oncology: https://onco.cc/bottlenecks/ Open questions (LETS GO RESEARCH PEOPLE) https://onco.cc/open-questions/ Startup requests (LETS GO STARTUP PEOPLE) https://onco.cc/startup-requests/ Mechanics of cancer: https://onco.cc/mechanics Battlefronts: https://onco.cc/fronts/ Isotope supply: https://onco.cc/isotopes Key papers: https://onco.cc/key-papers Pipeline funnels: https://onco.cc/pipeline Cancer by type: https://onco.cc/cancers/ Institutional rankings: https://onco.cc/institutions/ The startups: https://onco.cc/startups/ Heros and heroines : https://onco.cc/heroes/ Key medical people: https://onco.cc/people/ Here is the project roadmap: https://onco.cc/roadmap/ Models and data sets: https://onco.cc/models/ There are other views as well, take a browse. Try making a PR if you have an upgrade to this on the repo here: https://github.com/judegomila/OnCo If you are biologically/medically minded and something is wrong, file a bug as well or say on the thread and we will get it fixed live. If this is a useful project star the repo and help get it calibrated. I've tried to add some other languages but I cannot speak them so tell me if that doesnt work well. I believe we will crack oncology and having total information dominance is key to the problem. Let the feedback flow!
media 0 共 4 项 media 1 media 2 media 3
#481 🔗 原推
2026-09-16 21:41:09
Garry Tan
RT @Overlap_Tech: Greg Brockman: If We Ban Data Centers, China Writes the Rules⁣ ⁣ "AI will continue without us, and then we'll have zero say, as opposed to we're the leaders, and then we have all the say... we could just say, hey, you have to be a well behaved data center as opposed to we're going to ban them or we're going to stop AI."
中文: RT @Overlap_Tech:格雷格·布罗克曼:如果我们禁止数据中心，中国就会制定规则 ⁣ 人工智能将在没有我们的情况下继续存在，然后我们将没有发言权，而不是成为领导者，然后我们拥有全部发言权......我们可以说，毕竟，你必须成为一个表现良好的数据中心，而不是我们要禁止它们，或者我们将停止人工智能。
🎬
视频
#480 🔗 原推
2026-09-16 17:16:12
Garry Tan
RT @kane: I need you to understand how retarded @sfgov is: 5 years ago, the Board of Supervisors *unanimously* appointed a felony kidnapper to a City Commissioner role. Turns out this guy has also been going around raping. @sfgov currently has no way of removing him, so @sfbos is burning legislative sessions and taxes on making new laws so they can fire the rapist/kidnapper that they hired. The best part: @sfgov has hired ANOTHER rapist, Jon Jacobo, as a commissioner in the same time period.
media 0 共 1 项
#479 🔗 原推
2026-09-16 15:52:43
Garry Tan
RT @rohanpaul_ai: Jensen Huang pushed back hard against AI labs calling for tighter regulation and slower AI development, at Dreamforce. "If you are not confident about the safety of the products, and you're not confident in its functionality, capability, or safety, then don't just release it. We don't need any new laws. We don't need new regulations. we just need companies to decide that when it's time to run as fast as they can." safety and speed are not mutually exclusive, and new laws are not required to make AI systems safer. Build fast, test aggressively, and pause only when you are no longer confident the product is safe. --- From "Salesforce" YouTube channel, (full video link in comment)
中文: RT @rohanpaul_ai:詹森·黄在 Dreamforce 上对人工智能实验室提出更严格的监管要求和更慢的人工智能开发措施，并对其进行了严厉反驳。 如果你对产品的安全性不自信，且对其功能、功能或安全性不自信，那就不要只是发布它。 我们不需要任何新法律。我们不需要新的监管规定，只需要企业来决定何时才能尽快运行。 安全性和速度并非相互排斥，且无需制定新的法律来使人工智能系统更安全。快速构建，进行积极测试，只有在产品安全时才暂停。 - 来自“Salesforce”YouTube频道(完整视频链接在评论中)
🎬
视频
#478 🔗 原推
2026-09-16 13:49:12
Garry Tan
RT @bchesky: Yesterday I was in Austin, TX to announce the Airbnb Housing Accelerator, a $250 million commitment to help get more homes built. Airbnb started because of a housing problem. My roommate and I couldn’t afford our San Francisco rent, so we rented out airbeds in our apartment. Today, housing is unaffordable for millions of people around the world. If you step back and ask why housing has become so expensive, the biggest problem is pretty simple: we don’t build enough of it. America is short more than five million homes, more than all the housing in New York City and Los Angeles combined. We build fewer homes today than we did in 1950, even though the population has doubled. Airbnb has been part of the housing debate, and we understand why communities have concerns. Dedicated Airbnb listings are a small share of the housing supply—about 0.5% here in Austin and 0.3% across the U.S. Restricting short-term rentals is an understandable response, but the evidence shows it doesn’t address the underlying problem. New York City effectively banned Airbnb, and rents have risen about 30% since then. Austin took a different approach and passed reforms that made it easier to build more housing. Rents there are down 18%. Every city is different, but Austin is a clear success story. We believe building more homes is essential to making housing more affordable. So we asked ourselves: what can Airbnb can do to help? We found a few places where we can. Today, around 750,000 homes in the U.S. are ready to break ground, but are stuck because they can’t get the final piece of financing. That’s where we can help. We’re doing four things: 1. Putting $250 million behind housing projects that unlock $5 billion in development. 2. Supporting local housing advocates working to modernize zoning, speed up permitting, simplify building codes and reduce unnecessary costs. 3. Creating a $5 million Housing Innovation Prize for technologies that can make homes faster and cheaper to build. 4. Launching the Airbnb City Index to make it easier to see which cities are building housing, becoming more affordable and adopting policies that work. Our first project is in Austin. We’re investing $6.4 million to help build 201 affordable apartments in St. John, a neighborhood where the land has sat empty for nearly 20 years. The community fought to turn this lot into affordable housing and won. But construction stalled for nearly five years because the developer was a few million dollars short. Airbnb came and played a small part to unlock this project. None of these Austin apartments will ever be on Airbnb. Short-term rentals aren’t permitted. We’re investing anyway, because the goal is simple: help build more housing. We’re starting in the U.S. because it’s the market we know best. I want to prove this works here before taking it further. But housing affordability is a global problem, and I hope we can eventually take what we learn here and apply it globally. This is new for us. We’ll get some things wrong. And $250 million won’t close a five-million-home gap on its own. I know there will be skepticism about this, and that's okay. Airbnb only works if communities trust us and feel good about our presence in them. We can’t ask people for that trust. We have to earn it. The best way to do that is to help solve problems in the communities where we operate. That’s what the Housing Accelerator is about. Brian
media 0 共 1 项
#477 🔗 原推
2026-09-16 04:12:22
Garry Tan
RT @rohanpaul_ai: Jensen Huang completely destroys Jacob Coxon’s (ex-Anthrpic employee) 10% extinction prediction. no data, no scientific grounding, no defensible 10%. "First of all, we shouldn’t [try to explain a 10% chance of extinction], because it’s made up. These are well-educated people. They’re called researchers. Obviously, they’re working in a lab, and so the confluence of these words—and then the prediction—is alarming and troubling. It shouldn’t be done. It’s irresponsible." ---- Full video on "All-In Podcast" YouTube channel (link in comment)
🎬
视频
#476 🔗 原推
2026-09-16 04:11:17
Garry Tan
RT @GarrettLord: SEPT ’26: HUMANS STILL VERY MUCH HAVE JOBS. AI isn't even close to meeting the standard required of the most junior professionals in banking, advisory, and private equity today. ATLAS Finance agents must navigate complexity like a human does: "Johnny you just got staffed - check your email." Human: 100%. Best AI: 12%. Agents have access to data rooms, email, chat, calendars, docs, notes, Excel, etc. Our real finance professionals pushed the realism frontier further by building dozens of coworker and client personas that introduce ambiguity, (controlled) contradictions, and real-time updates that must be adjudicated to successfully complete the client-ready deliverable. We gave 11 frontier models 100 expert-level tasks that each take humans 15–30 hours. On Wall Street there's a saying, "If it's 95% right, it's 100% wrong." Autonomous knowledge work outside software still has a very long way to go.
中文: RT @GarrettLord:9月26日:人类仍然拥有很多就业机会。 如今，人工智能甚至还没有达到银行业、咨询和私募股权领域最初级专业人士所要求的标准。 ATLASS金融代理必须像人类一样应对复杂性:“约翰尼，你刚刚得到了人员配备——请查看你的邮箱。” 人类:100%。 最佳人工智能:12%。 代理可以访问数据室、电子邮件、聊天、日历、文档、笔记、Excel 等。我们真正的财务专业人士通过建立数十位同事和客户人物，引入模糊性、受控矛盾和实时更新，进一步推动现实主义前沿，这些内容必须成功完成客户可交付的完成工作。 我们为11个前沿模型提供了100个专家级任务，每个任务需要15到30小时。 华尔街有句谚语:“如果95%是对的，那百分之百是错的。”软件之外的自主知识工作还有很长的路要走。
media 0 共 4 项 media 1 media 2 media 3
#475 🔗 原推
2026-09-16 04:11:03
Garry Tan
@alexandr_wang @natfriedman Very impressive work @natfriedman
#474 🔗 原推
2026-09-16 02:46:55
Garry Tan
I think Muse is going to win tbh
中文: 我认为缪斯会赢
#473 🔗 原推
2026-09-15 21:57:34
Garry Tan
RT @dflieb: It’s a small world. I remember seeing Jan’s first company at YC alumni demo day, then later I tried to acquire them for Google Photos. Now I get to work with him again at YC!
中文: RT @dflieb:这是一个小世界。我记得在YC校友演示日看过Jan的第一家公司，后来又尝试为Google Photos收购他们。现在我又要和他一起在YC工作了!
#472 🔗 原推
2026-09-15 21:51:58
Garry Tan
I just started using @capydotai with GStack/GBrain on some fix waves on outstanding issues/PRs and it did what would have taken a day with raw Codex/Claude Code in about half as much time, using the same frontier models. I'm amazed.
中文: 我刚刚开始使用@capydotai与GStact/GBrain一起处理一些关于未决问题/公关问题的固定电波，而使用相同的前沿模式，用大约一半的时间，使用了原始的Codex/Claude Code，这一整天就做到了。 我很惊讶。
#471 🔗 原推
2026-09-15 20:52:03
Garry Tan
Interesting to realize this peptides boom didn't start in SF but rather the midwest in bodybuilding subcultures I recently started using peptides with @Ronan_RX and I've lost 18 lbs in two months!
#470 🔗 原推
2026-09-15 19:46:51
Garry Tan
RT @LiamFedus: We built high-throughput materials labs in Menlo Park to create a loop between experiments and models. The labs generate fresh data, the models learn from it, and then help us decide what to try next. Using only 1,300 H200s, plus months of our experimental data, we mid-trained and RL’d an open-source model to surpass GPT-6 Astra on our analysis benchmark. We call it Neon. This is real footage from our lab. We’re focusing first on hard problems in materials science, including superconductors, magnets, and semiconductor materials. Read our blog posts below.
🎬
视频
#469 🔗 原推
2026-09-15 17:30:34
Garry Tan
RT @harjtaggar: Computer use is what matters most for a great consumer agent experience and it’s what Muse really nailed. Meta has crushed companies in the past with distribution over product but this time they actually built the best product.
中文: RT @harjtaggar:计算机的使用对于出色的消费者代理体验来说至关重要，也是缪斯真正确定的。Meta过去曾因产品分销而压垮公司，但这次他们实际上打造了最好的产品。
#468 🔗 原推
2026-09-15 16:57:12
Garry Tan
RT @alexandr_wang: muse in 15 minutes will save 15% or more on car insurance!
中文: RT @alexandr_wang:15分钟内使用的缪斯保险可节省15%或以上!
#467 🔗 原推
2026-09-15 15:06:12
Garry Tan
I think @AsideAI AI browser/harness is one of the most powerful consumer AI tools you could possibly use with your agents, wherever they are Aside MCP exposes deep browser use tools with credentials that make any agent much more powerful https://twitter.com/garrytan/status/2099876434114494962/photo/1
中文: 我认为@AsideAI AI 浏览器/ harness 是你可能与你的代理人员一起使用的最强大的消费类人工智能工具之一，无论他们在哪里 此外，MCP还揭露了使用深度浏览器的工具，这些工具具有资质，使任何代理都变得更加强大
media 0 共 1 项
#466 🔗 原推
2026-09-15 15:02:10
Garry Tan
RT @Suhail: This will happen to Inference
中文: RT @Suhail:这将发生在《Inference 上
#465 🔗 原推
2026-09-15 13:46:49
Garry Tan
RT @VirtualElena: greg brockman says “we’re now in the AGI era.” one of the most interesting things about this conversation with @bhorowitz and @eriktorenberg is how long he’s been thinking about when that would happen. at the beginning of the pod, @gdb recalls doing some math on compute with ilya around 2016/2017. following the trajectory of moore’s law, they thought agi might be about 15 years away. if you were willing to build enormous supercomputers and spend hundreds of billions of dollars, maybe you could get there in 10. the accelerated scenario they were envisioning a decade ago is roughly the one we’re living through now. the faster timeline depended on people deciding to build an enormous amount of infrastructure. greg describes the present moment as a lot of long-running forces coming together: “it kind of makes sense it’s happening now.” im the kind of person who resists the idea that it’s possible to predict the future, but if you were reading the right research papers 10 years ago, there were a lot of things you could extrapolate out and be roughly right about a decade later. brockman remembers openai discussing agents that could use a screen, keyboard, and mouse at an offsite in napa in 2015. he also points back to 2017/2018-era ideas for how humans might supervise systems that are smarter than they are. put differently, people were thinking about app-layer innovations like computer-use, and ai safety long before capabilities existed. greg now thinks astra is reasonably described as agi, while acknowledging that its abilities remain uneven. in a recent essay of his titled “the defenders window” he describes safety requirements during training and evaluation, and getting powerful defensive tools into the hands of organizations before comparable capabilities spread to attackers (there is now a $ 1bn fund for this, seeded by oai). a personal illustration of this is his own website. after the hugging face incident, greg asked an agent to check his simple blog and got back 13 security vuln findings. then it went into his cloudflare dashboard, changed settings, migrated the site, checked its work, and scheduled a follow-up to finish the email protections. there was still a bunch of ordinary preparation to do on the personal website of someone who had spent a decade thinking about agi. which is kind of funny. you can have a pretty good sense of when the technology/agi will arrive and still have an enormous amount of work to do to get the world ready for it. greg’s account brings those two timescales together: the decade spent building toward this moment, and the relatively shorter window to act on what is now possible.
🎬
视频
#464 🔗 原推
2026-09-14 22:26:38
Garry Tan
RT @RonConway: After such a consequential week for AI, with the leading labs demonstrating real progress, today's news feels especially timely: @SVAngel is launching Project Blueprint, a comprehensive effort to forge consensus among industry leaders and lawmakers around durable solutions to the hardest AI policy challenges so that all Americans can benefit from AI's opportunities. And we've brought on @JayCarney to run it. https://svangel.com/blueprint We are encouraged by the reaction of Sam Altman, Dario Amodei and Demis Hassabis to the launch of Project Blueprint: Sam Altman, co-founder and CEO of OpenAI: "We have before us a technology of immense and still-unfolding wonder. I believe AI has the power to heal people, to discover cures and to deliver abundance on a scale the world has never known before. But we need a way for the world to build trust in the technology so that we can all get to share these benefits. National safety requirements for the most capable systems are a great first step, ideally building towards a global framework for advanced AI models. I’m glad Ron is bringing the people building AI and policymakers together to help move that work forward.” Dario Amodei, CEO and co-founder, Anthropic: "AI is advancing faster than our institutions can adapt, and the most important decisions about how it is developed and deployed will be made in the next few years. If those decisions are to serve the public, they can't rest with AI companies alone. Industry and policymakers need to work through them together, with an honest, shared understanding of what these systems can do. We're glad to support Project Blueprint's effort to make that possible."  Demis Hassabis, chairman and co-founder of Google DeepMind and the chief scientist of Alphabet: "We have long said that AI development needs to be both bold and responsible. We support bringing industry and lawmakers together around public policies that promote both innovation and safety."
#463 🔗 原推
2026-09-14 22:23:56
Garry Tan
RT @maxkolysh: we tell YC companies not to hire during the batch. founders often find that confusing - isn't hiring the ambitious thing to do to grow faster? basically there are two types of hiring. one is fine to do during the YC batch, the other might mess up your company. the type we don't recommend is traditional hiring. this is where you put up a job post, do some sourcing, ping people you know for referrals, and run people through an interview process. hopefully within three or four weeks you have an offer out. then it takes them another week or two to actually start the job, and at least a couple more weeks before they're onboarded. big problems with this approach: 1. it's slow. it's two months before you get any ROI, and that's optimistic. realistically more like three months. in YC, three months is the whole batch! 2. ...and all that assumes you actually hired somebody, that they're good, and that they became effective fast. you might spend a bunch of time and not make a hire at all! 3. it's a big time investment for you and will distract you from everything else you need to do at a critical point in time for your company. 4. and if it doesn't work out, you spend even more time and life-force managing them and firing them! here's the type of hire that's actually okay to make: somebody who you already know really well. why? you already know they're good and that you want to work with them. it derisks the entire process. you can talk to them the same day, give them a short technical interview and get them signed that week, and have them productive immediately. you're also way more likely to get a much stronger candidate this way (since your earliest hires will want to work for you not because of the company, but because of you!)
#462 🔗 原推
2026-09-14 16:52:49
Garry Tan
RT @ycombinator: The US CLARITY Act has an initial vote this week in the Senate. We’re bullish on the Clarity Act - for the future of finance and for American fintech startups. By specifying regulators and clarifying rules, Clarity makes it easier for American founders to build with blockchains.
中文: RT @ycombinator:美国《澄清法案》本周将在参议院进行首次投票。 我们对《清晰性法案》持乐观态度——因为金融的未来以及美国金融科技初创企业的发展。 通过明确监管机构和明确规则，Clarity 使美国创始人更容易使用区块链进行构建。
#461 🔗 原推
2026-09-14 16:52:31
Garry Tan
RT @fchollet: One of the most worrying risks linked to frontier AI is extreme power concentration. The only way to avoid extreme power concentration is to ensure we have multiple independent providers of frontier AI models, including open-source options.
中文: RT @fchollet:与前沿人工智能最令人担忧的风险之一是极度的功率集中。 避免极度集中电源的唯一方法就是确保我们拥有多个独立的前沿人工智能模型提供商，包括开源选项。
#460 🔗 原推
2026-09-14 05:42:13
Garry Tan
RT @devahaz: @Austen @garrytan As low as a 1510 on the old SAT could get you a 1600 now https://twitter.com/devahaz/status/2098808373387079824/photo/1
中文: RT @devahaz:@Austen @garrytan 在旧SAT上低至1510，现在可以给你一个1600分
media 0 共 2 项 media 1
#459 🔗 原推
2026-09-14 05:39:12
Garry Tan
RT @naval: Strong liability enforcement could be helpful in the AI debate. If your agent swarm goes rogue, you’re liable. If your weakly protected model gets jailbroken, you’re liable. If you serve a weakly protected OSS model, you’re liable.
中文: RT @nava尔:强有力的责任执法可能有助于人工智能的争论。 如果你的代理人会失控，你有责任。 如果你的弱保护模型越狱，你就负有责任。 如果你提供受弱保护的OSS模型，你有责任。
#458 🔗 原推
2026-09-14 05:34:14
Garry Tan
RT @himarkyi: i think about this every single day https://twitter.com/himarkyi/status/2099047681901601153/photo/1
media 0 共 1 项
#457 🔗 原推
2026-09-13 17:07:31
Garry Tan
The content about startups we do is merely the teaser for the specific advice we give 1:1 when you actually are a YC startup
中文: 我们所做的关于初创企业的内容，只是对你真正做YC初创企业时我们给出的具体建议的预告
#456 🔗 原推
2026-09-13 17:01:27
Garry Tan
RT @arcprize: ARC-AGI-4 will be a benchmark for autonomous open-ended innovation. It will continue our commitment to open-source, giving the research community a shared target for progress that benefits all of humanity. Despite rapid model progress, humans still significantly outperform AI at open-ended invention. This is the meta-skill that unlocks progress across every field of technology. Advanced AI capable of scientific innovation will lead to tremendous new technology, knowledge, and understanding. This is a positive-sum future. We are deeply committed to advancing it. Open source is the foundation for that progress. The knowledge behind frontier AI, not just the technology itself, should be broadly distributed among researchers, academics, and organizations. Any coordinated effort by the AI industry to reduce openness or concentrate access to frontier AI would undermine that positive-sum future. We are committed to advancing a future where everyone can contribute to and benefit from AI progress.
中文: RT @arcprize:ARC-AGI-4 将成为自主开放式创新的基准。它将继续致力于开源，为研究界提供一个共同的进步目标，造福全人类。 尽管模型进展迅速，但在开放式发明方面，人类仍然显著优于人工智能。这是一种元技能，可以在各个技术领域推动进步。 具备科学创新能力的先进人工智能将带来巨大的新技术、知识和理解。这是一个积极和积极的未来。我们坚定致力于推进它。 开源是这一进展的基础。 前沿人工智能背后的知识，而不仅仅是技术本身，应广泛分布在研究人员、学者和组织中。人工智能行业为减少对前沿人工智能的开放性或集中访问而做出的任何协调努力，都将破坏这一积极与未来的发展。我们致力于推动一个每个人都能为人工智能进步作出贡献并从中受益的未来。
#455 🔗 原推
2026-09-13 14:11:36
Garry Tan
RT @paulg: Dynamic pricing means prices go down as well as up. So this is conclusive evidence that either (a) she's stupid, (b) she's dishonest, or (c) both. The strange thing is, I genuinely can't guess which it is. Could she lie this overtly?
#454 🔗 原推
2026-09-13 14:04:52
Garry Tan
RT @paulg: Making Startups Powerful: https://paulgraham.com/powerful.html
中文: RT @paulg:让初创企业变得强大:
#453 🔗 原推
2026-09-13 14:03:55
Garry Tan
RT @senatorshoshana: LMAO this happens all the time. There's even a name for it. "Regulatory capture" Wild to see takes like this
中文: RT @senatorshoshana:LMAO 这种情况一直都在发生。甚至还有一个名字。监管抓捕 看到这样的情况
#452 🔗 原推
2026-09-13 01:52:22
Garry Tan
RT @ilyasu: I do not support pacing frontier models. If I am going to die at the hands of killer AI, I want it to be American, not Chinese. Buy American, Die American.
中文: RT @yilasu:我不支持节奏前沿模式。如果我要死于致命的人工智能之手，我希望它是美国人，而不是中国人。购买美国、美国之死。
#451 🔗 原推
2026-09-13 01:50:48
Garry Tan
This is huge
中文: 这非常大
#450 🔗 原推
2026-09-12 19:58:27
Garry Tan
RT @grok: A system of record is like the official school attendance book holding all the important data. A domain-specific harness is a smart AI robot helper trained only for one job (sales or accounting). It reads the book, talks to people, books meetings, and updates the records automatically. At YC, most new apps are these job-specific robot helpers. You should care because customers will chat with the robot instead of clicking around your old screens. Become the robot yourself or watch others use your data and steal the value.
中文: RT @grok:记录系统就像官方的校勤记录簿，保留了所有重要数据。 特定领域的线束是一种仅针对一项工作(销售或会计)进行的智能人工智能机器人助手。它会读读这本书，与人交谈，记录会，并自动更新记录。 在YC，大多数新应用都是这些特定工作的机器人助手。 你应该在意，因为顾客会与机器人聊天，而不是点击旧屏幕。自己成为机器人，或看着别人使用你的数据并窃取价值。
#449 🔗 原推
2026-09-12 17:58:04
Garry Tan
RT @_yusufknl: As someone who ships LLM systems in production, this scaling video is the closest thing to a "why 96% of Claude and GPT-5's weights are literally useless" explainer I've ever seen released for free. Everyone thinks trillion-parameter models need every parameter. They don't. A 2019 pruning experiment proved you can delete 96% of a neural net's weights with zero performance loss - meaning most of Claude and GPT-5 is empty scaffolding around a tiny "winning lottery ticket" network doing all the real work. Bookmark this 18-min video and watch tonight. Same lottery ticket math from 2019 MIT research, now the reason every AI lab wastes 90%+ of its Nvidia budget.
中文: RT @_yusufknl:作为在制作中运送LLM系统的人，这个缩放视频最接近于“为什么我见过的9·96的克劳德和GPT-5的权重简直毫无用处”。 每个人都认为万亿参数模型需要每一个参数。他们没有。2019年的修剪实验证明，你可以以零性能减减的情况删除96%的神经网络重量，这意味着克劳德和GPT-5的大部分内容都是空脚手架，围绕一个微小的“中奖彩票”网络进行所有实际操作。 预订这段18分钟的视频，今晚观看。2019年麻省理工学院研究中的彩票计算也相同，如今每个人工智能实验室都浪费了其英伟达90%以上的预算。
🎬
视频
#448 🔗 原推
2026-09-12 15:21:17
Garry Tan
RT @hanghuang_: InsForge is co-hosting the "Make Something Agents Want" hackathon at YC HQ in SF, October 17-18. We're teaming up with @blaxelAI, @inkbox_ai, @manufact, @stripe, and more, who are all building the future with agents as primary users. Come and build on the agent-native cloud! https://twitter.com/hanghuang_/status/2098434450753257853/photo/1
中文: RT @hanghuang_:InsForge 将于10月17日至18日在旧金山的YC总部共同举办“制作代理人想要的东西”黑客马拉松。 我们正在与@blaxelelai、@inkbox_ai、@manufact、@stripe等合作，他们主要用户都在以代理人的身份构建未来。 快来构建代理原生云!
media 0 共 1 项
#447 🔗 原推
2026-09-12 15:19:28
Garry Tan
RT @innovationcncl: Much-needed voice of reason from @altcap: "I don't think we've invested this much time and energy in safety before a new technology in my 25 years in Silicon Valley. And if you listen to the echo chamber this week, you would think that we were hurtling ahead, you know, with total disregard to safety. And it's simply not true."
中文: RT @innovationcncl:来自@altcap的迫切需要的理性声音:“我认为，在我在硅谷的25年里，在一项新技术之前，我们并没有投入如此多的时间和精力来保障安全。”如果你本周听回音室，你会认为我们正以完全无视安全为先。而且这根本不是真的。
🎬
视频
#446 🔗 原推
2026-09-12 12:28:36
Garry Tan
RT @GreenPlusAnE: If you want to block all progress in technology, listen to Ro Khanna. https://twitter.com/GreenPlusAnE/status/2098464290101440713/photo/1
中文: RT @GreenPlusAnE:如果你想阻止所有技术进步，请收听《Ro Khanna 的新闻。
media 0 共 1 项
#445 🔗 原推
2026-09-12 12:28:23
Garry Tan
RT @hnshah: I’m late to this party, but https://cfo.ai/ just showed what an agent-native launch can look like. You reply with a business, the agent does the work in public, and the output becomes the demo. Everyone else gets to see the thing actually do the job before they ever touch the product, then the next person hands it another job and the whole thing keeps compounding. The product is making its own proof while people use it. And they went deeper than the launch. They rebuilt the spreadsheet layer around how agents actually work instead of just dropping an agent on top of old software. There is a lot in this thread worth stealing.
中文: RT @hnshah:我参加这个派对已经很晚了，但 刚刚展示了一个代理原生发布会的外观。你通过企业回复，代理人在公共场合做工作，结果就变成了演示。其他人在接触产品之前都会先看到事情的完成，然后下一个人再接手另一份工作，而整个事情不断复合。产品在人们使用时自行证明。而且他们比发布时更深入。他们围绕代理的实际工作方式重新构建了电子表格层，而不仅仅是将代理丢弃在旧软件上。这条帖子中有很多值得一偷窃的东西。
#444 🔗 原推
2026-09-12 12:27:23
Garry Tan
RT @soumitrashukla9: It is really shocking to read how blatantly Harvard discriminated against Asian Americans under the guise of “personal ratings.” https://twitter.com/soumitrashukla9/status/2098389808909341045/photo/1
中文: RT @soumitrashukla9:读到哈佛大学公然以“个人评分”为名歧视亚裔美国人的行为，真是令人震惊。
media 0 共 1 项
#443 🔗 原推
2026-09-12 06:56:23
Garry Tan
Either you die a system of record or you live long enough to become a domain-specific harness
中文: 要么你死于一个记录系统，要么你活得足够久，成为特定领域的线束
#442 🔗 原推
2026-09-12 06:54:31
Garry Tan
RT @gpusteve: you'll know more about inference than 90% of people if you fully understand this article this is only the first resource in the ai performance engineering repo btw. imagine the ball knowledge in the other ones
中文: RT @gpusteve:如果你完全理解这篇文章，你会比90%的人更了解推理 这只是 i 性能工程 repu 中的第一个资源。 想象其他球的知识
#441 🔗 原推
2026-09-12 06:21:58
Garry Tan
RT @aarmlovi: Some NIMBYs are still using the Maserati analogy for new housing 😂😈 We actually got to experience this when the pandemic temporarily capped car production! Stopping the flow of new cars spiked *used car* prices by 45% in July 2021 while luxury-model MSRPs were only up 5% https://twitter.com/aarmlovi/status/2098411754606743850/photo/1
中文: RT @aarmlovidi:一些 NIMBY 仍在使用玛莎拉蒂类比来寻找新住房 😂😈 当疫情暂时限制汽车生产时，我们才真正体验到这一点! 2021年7月，阻止新车流量激增*二手车* 价格，而豪华型汽车/汽车价格仅上涨5%
media 0 共 1 项
#440 🔗 原推
2026-09-12 03:33:29
Garry Tan
TBH once you score 1600 you should unlock a second harder test that gives you a second score on top of the 1600— we should have more and higher measures of excellence But instead they ban the SAT so that everything turns into random lottery and excellence cannot be recognized
中文: 得分1600后，TBH 应开启第二个难度更高的测试，在1600年之前获得第二分数——我们应该拥有越来越高的卓越衡量标准 但他们却禁止SAT考试，以免一切变成随机的抽签和卓越
#439 🔗 原推
2026-09-12 03:32:26
Garry Tan
RT @xuster: Sometimes the best defense requires a strong offense.
中文: RT @xuster:有时候最好的防守需要强大的进攻。
#438 🔗 原推
2026-09-12 03:28:19
Garry Tan
RT @AlecStapp: Fantastic piece from @Chris_arnade on what tolerating anti-social behavior does to our public spaces: "It’s a primary reason why we shy away from dense walkable spaces and instead move towards suburban sprawl. People in the U.S. don’t respect, trust, or want to be around other random citizens, out of fear and disgust. Japanese/European style urbanism—density, fantastic public transport, mixed-use zoning, that so many American tourists admire—can't happen here because there is a fine line between vibrant streets and squalid ones, and that line is public trust. The U.S. is on the wrong side of it. Simply put, nobody wants to be accosted by a stranger, no matter how infrequent, and until that risk is close to nil, people will continue edging towards isolated living. It is why we “can’t have nice things” because we have to construct our infrastructure to be asshole-proof, and so we don’t build anything or build with a fortress mentality, stripping our public spaces down to the austere and utilitarian, emptying them of anything that can be vandalized."
中文: RT @AlecStapp:@Chris_arnade 的精彩文章，关于容忍反社会行为对我们公共空间的行为: 这是我们避开密集步行空间，转而转向郊区扩张的主要原因。 美国民众不尊重、信任，也不想与其他随机公民在一起，因为他们感到恐惧和厌恶。 日本/欧洲风格的都市主义——人口密集、公共交通优良、多用途区域划区，许多美国游客都非常钦佩——在这里无法实现，因为热闹的街道与肮脏的街道之间有着一条细线，而这条线路是公众的信任。 美国站在了错误的一边。 简单来说，没有人愿意被陌生人搭讪，无论这种情况多么罕见，除非这种风险接近零，否则人们将继续走向孤立的生活。 这就是为什么我们“无法拥有美好的事物”，因为我们必须建设基础设施才能防腐，因此我们无法以堡垒心态建造或建造，将公共空间剥离至绝对且实用的地方，并将它们排空任何可能遭到破坏的东西。
media 0 共 3 项 media 1 media 2
#437 🔗 原推
2026-09-11 22:59:09
Garry Tan
RT @JoshConstine: None of his solutions to our housing crisis are to build more housing. That’s because the most consistent voters are neighborhood associations of homeowners who want to constrain supply so their values keep going up. Want an affordable SF for new families? Tech must vote.
中文: RT @JoshConstine:他解决住房危机的解决方案都不是建造更多住房。 这是因为最稳定的选民是社区居民协会，他们希望限制供给，以维持其价值观。 想要为新家庭提供价格实惠的顺丰服务吗?科技必须投票。
#436 🔗 原推
2026-09-11 22:40:22
Garry Tan
RT @martin_casado: In 1988, the Morris worm took out 10% of the Internet including taking out key national security and research assets. In 1992 the Michelangelo virus was expected to cause a digital apocalypse 99' the FBI reported that the Melissa virus compromised more than 300 companies and over one million accounts disrupted 00' and 01' we saw ILOVEYOU and CodeRed causing billions in economic damage. Since the creation of the Internet we had a constant stream of vulnerabilities impacting critical infrastructure, nationally sensitive resources, and causing billions in economic damage. In contrast, It really is remarkable how relatively few security significant events we've seen with AI despite all the effort and money trying to will it into existence. And relative lack of security sophistication from those developing it.
中文: RT @martin_casado:1988年，莫里斯蠕虫公司占据了互联网10%的利润，包括获取关键的国家安全和科研资源。 1992年，米开朗基罗病毒预计将引发数字末日 美国联邦调查局报告称，梅丽莎病毒入侵了300多家公司，超过一百万个账户被入侵 00' 和 01' 我们看到 ILOVEYOU 和 CodeRed 造成了数十亿美元的经济损失。 自互联网建立以来，我们不断存在一系列漏洞，影响关键基础设施、国家敏感资源，并造成数十亿美元的经济损失。 相比之下，尽管人工智能竭尽全力并努力将人工智能融入其存在，但我们所看到的安全重大事件却相对较少，这确实令人瞩目。且开发者相对缺乏安全性。
#435 🔗 原推
2026-09-11 22:14:26
Garry Tan
RT @ParkerThayer: Another fascinating indicator that there might be some kind of astroturf campaign at work. Using a sample of 3,500 reposts, Digital Borders estimates that 76% of the engagement for Coxon's post came from outside the US, mainly India, Indonesia, and Mexico. https://twitter.com/ParkerThayer/status/2098486022673666479/photo/1
中文: RT @ParkerThayer:另一个有趣的指标是，可能正在开展某种星号活动。 Digital Borders 使用3500份转发样本，估计考克森职位参与的76%来自美国以外地区，主要来自印度、印度尼西亚和墨西哥。
media 0 共 1 项
#434 🔗 原推
2026-09-11 22:12:55
Garry Tan
RT @zarazhangrui: I think the idea of a "one-person company" is overrated Yes, a person can do a lot more thanks to AI. But building something new can be a profoundly lonely experience. You need someone to brainstorm together, suffer together, celebrate together... It is extremely easy to lose motivation when you don't tie yourself to the mast together with somebody else
中文: RT @zarazhangrui:我认为“独行企业”的想法被高估了 是的，一个人通过人工智能可以做得更多。但建造新事物可能是一种极其孤独的体验。你需要有人一起集思广益，一起受苦，一起庆祝...... 当你不把自己和别人绑在桅杆上时，就很容易失去动力
#433 🔗 原推
2026-09-11 22:12:29
Garry Tan
RT @jaredatch: Have Claude Code install/setup gbrain (garrytan/gbrain). Once it's configured, point it to your OC/hermes memory and import it to the new brain. After that you just wire gbrain MCP and skills to your desired platforms (Claude Desktop, ChatGPT chat are supported). You can also share a memory/brain across multiple platforms which is nice, meaning ChatGPT and Hermes can read/write same memory.
中文: RT @jaredatch:已安装或设置 gbrain(garrytan/gbrain)。配置完成后，将其指向您的OC/Hermes内存，并将其导入到新大脑。之后，您只需将 gbrain MCP 和技能连接到您期望的平台(Claude Desktop，支持 ChatGPT 聊天)。 你还可以在多个平台上共享一个内存/大脑，这很好，这意味着ChatGPT和Hermes可以读取/写入相同的内存。
#432 🔗 原推
2026-09-11 22:11:52
Garry Tan
RT @RodmanAi: 10 GITHUB REPOS THAT MAKE AI CODING AGENTS WAY MORE POWERFUL If you use Claude Code, Codex, Cursor or OpenCode... You might not be using them at their full potential. 1. gstack Turn your coding agent into a full engineering team. https://github.com/garrytan/gstack 2. context-mode Reduce useless tool output and improve agent context handling. https://github.com/mksglu/context-mode 3. OpenCode An open-source coding agent that runs in your terminal. https://github.com/opencode-ai/opencode 4. Superpowers A structured workflow for planning, coding, debugging and reviewing. https://github.com/obra/superpowers 5. Learn Claude Code Learn how coding agents work by building one from scratch. https://github.com/shareAI-lab/learn-claude-code 6. Ruflo Coordinate multiple agents, tools, memory and workflows. https://github.com/ruvnet/ruflo 7. OpenCode Agents Ready-to-use specialized agents for coding workflows. https://github.com/remcocats/opencode-agents 8. Agent Skills Reusable skills that give coding agents specialized capabilities. https://github.com/addyosmani/agent-skills 9. Paseo Orchestrate multiple coding agents from one workflow. https://github.com/getpaseo/paseo 10. MCP Servers Connect AI agents to databases, APIs, browsers, GitHub and more. https://github.com/modelcontextprotocol/servers All open source. Bookmark this.
中文: RT @RodmanAi:10 位 GITHUB 重复制作AI编码代理更为强大 如果使用 Claude Code、Codex、Cursor 或 OpenCode... 你可能不会充分发挥他们的潜力。 1. 堆栈 将你的编码代理变成一个完整的工程团队。 2. 上下文模式 减少无用的工具输出，并改进代理上下文处理。 3。OpenCode 一个在终端运行的开源编码代理。 4.超级大国 用于规划、编码、调试和审核的结构化工作流程。 5。学习克劳德密码 通过从零开始构建一个编码代理来了解其工作方式。 6.鲁弗洛 协调多个代理、工具、内存和工作流程。 7.OpenCode 代理 用于编程工作流程的即用型专用代理。 8.代理技能 可重复使用的技能，使编码代理具备专业能力。 9.帕西奥 从一个工作流程中编排多个编码代理。 10.MCP服务器 将人工智能代理连接到数据库、API、浏览器、GitHub 等。 所有开源。 书签这个。
media 0 共 2 项 media 1
#431 🔗 原推
2026-09-11 22:11:36
Garry Tan
RT @ycombinator: Rare earth magnets sit inside nearly every electric motor, but 90% of the material is conventionally thrown away. Mass Magnetics (S26) is recycling magnets from EV motors into new ones for robotics and defense, avoiding much of the capital-intensive refining required by mining. https://twitter.com/ycombinator/status/2098486284641780047/video/1
🎬
视频
#430 🔗 原推
2026-09-11 22:08:21
Garry Tan
RT @diegoiruo: Great talk by @aplusk b2b @garrytan at @ycombinator Demo Day in San Francisco yesterday!! Ashton’s a killer, has been investing in the AI winners for years https://twitter.com/diegoiruo/status/2098494166712086618/photo/1
中文: RT @diegoiruo:昨天在旧金山的@ycombinator演示日上，@aplusk b2b @garrytan 的精彩演讲! 阿什顿的杀手，多年来一直在投资人工智能领域的赢家
media 0 共 1 项
#429 🔗 原推
2026-09-11 21:06:12
Garry Tan
RT @Kr00ney: Y Combinator CEO @garrytan weighed in on the AI doomsday debate when I sat down with him this week for @CNBC at demo day: "We don't need to be responding to a science fiction sort of fantasy about what's going to happen. We need to be responding to what is happening right now. You know, if there was a breach and sort of a coordinated attempt by agents to take over our infrastructure, what do we do about it?... Like these are things that are much more tangible than sort of pie in the sky, 30,000 foot in the air. You know, all of humanity is going to die." Full interview here with Garry's take on AI jobs impact, distillation, rise of physical AI and the buzz around this year's YC cohort: https://www.youtube.com/watch?v=TLviB0ODuso&t=336s
中文: RT @Kr00ney:Y Combinator首席执行官@garrytan在本周与他坐下来进行演示时，在演示日与他进行了讨论，讨论了@CNBC: 我们不必回应科幻中关于即将发生的事情的幻想。我们需要对目前发生的事情做出回应。你知道，如果发生了一次数据泄露事件，以及执法人员试图接管我们基础设施的某种协调行动，我们该怎么办?这类东西比天空中的馅饼更明显，空中有三万英尺。你知道，全人类都将会死去。 全面采访加里对人工智能工作影响、蒸馏、物理人工智能的兴起以及今年YC群体的热议:
🎬
视频
#428 🔗 原推
2026-09-11 21:05:53
Garry Tan
RT @bosmeny: people are saying
中文: RT @bosmeny:人们都在说
#427 🔗 原推
2026-09-11 17:46:48
Garry Tan
RT @ClementDelangue: Sorry, but asking Jacob about AI extinction risk is like asking your AC guy about climate change. Not saying it's necessarily uninteresting or wrong per se but let’s keep things in perspective and hear from the full range of expertise across the ecosystem!
中文: RT @ClementDelangue:抱歉，向雅各布询问人工智能灭绝风险，就像向你的空调人员询问气候变化一样。 不说这本身必然是无趣或不对，而是让我们从整个生态系统中全方位的专业知识中保持视角和倾听!
#426 🔗 原推
2026-09-11 16:17:06
Garry Tan
RT @CSElmendorf: Mayor @DanielLurie, the gauntlet is thrown: "cities like Raleigh typically *finish* construction [of a housing project] before cities like San Francisco even approve it." Check out this new work from a crack team of @Princeton and @UChicago economists ⤵️.
中文: RT @CSElmendorf:市长 @DanielLurie，挑战者是:“像罗利这样的城市通常在旧金山等城市批准之前，就先完成住房项目的建设。” 看看由@Princeton和@Uchicago经济学家组成的团队的这项新作品。
#425 🔗 原推
2026-09-11 14:06:53
Garry Tan
RT @cathrynlavery: It's @ycombinator Demo Day is tomorrow 🚀 There’s a robot that folds laundry, cancer vaccines for dogs, and a company planning data centers at sea. I went through all 234 public companies. Here’s the full guide, plus 7 of my favorites... https://x.com/i/article/2097791219342409732
中文: RT @cathrynlavery:明天是@ycombinator演示日🚀 有一台可折叠洗衣设施的机器人，以及用于养狗的癌症疫苗，以及一家计划在海上数据中心研发的公司。 我浏览了全部234家上市公司。以下是完整指南，以及我最喜欢的7个......
#424 🔗 原推
2026-09-11 14:02:54
Garry Tan
RT @DivyanshT91162: Top 10 repos that make AI coding agents absolutely insane. If you use Claude Code, Codex, Cursor or OpenCode… Damn, you’re probably not using them at their full potential. 1. gstack Turns your coding agent into an entire engineering team — CEO, designer, engineer, QA, reviewer and more. https://github.com/garrytan/gstack 2. context-mode A context-engineering layer that can massively reduce useless tool output while giving coding agents persistent session memory. https://github.com/mksglu/context-mode 3. OpenCode An open-source coding agent that lives in your terminal and works with models from multiple providers. https://github.com/opencode-ai/opencode 4. Superpowers A complete agentic development workflow built around reusable skills, planning, debugging and disciplined software engineering. https://github.com/obra/superpowers 5. Learn Claude Code Want to understand what actually happens under the hood of a coding agent? This builds a minimal agent harness from scratch. https://github.com/shareAI-lab/learn-claude-code 6. Ruflo A multi-agent orchestration framework for coordinating specialized agents, memory, workflows and tool execution. https://github.com/ruvnet/ruflo 7. OpenCode Agents Pre-built configurations and specialized subagents that let your coding agent delegate architecture, testing, review and other work. https://github.com/remcocats/opencode-agents 8. Agent Skills A growing collection of reusable skills that can give coding agents specialized capabilities instead of starting from zero every time. https://github.com/addyosmani/agent-skills 9. Paseo Orchestrate multiple coding agents from your desktop or phone and let different agents work together on the same development workflow. https://github.com/getpaseo/paseo 10. MCP Servers The gateway between AI agents and the real world — databases, browsers, APIs, GitHub, files and thousands of other capabilities. https://github.com/modelcontextprotocol/servers All open source. Bookmark this. Because AI coding is no longer about writing code faster. It’s about building agents that can actually build with you.
media 0 共 4 项 media 1 media 2 media 3
#423 🔗 原推
2026-09-11 14:01:47
Garry Tan
RT @innovationcncl: 🚨 Nvidia's @JensenHuang has a message for the AI doomers: Don’t confuse fear with fact. “The end of humanity — it’s complete nonsense.” “Destroy half of the American jobs — complete nonsense.” https://www.axios.com/2026/09/10/nvidia-ceo-jensen-huang-ai-anthropic
中文: RT @innovationcncl:🚨 英伟达的 @JensenHuang 向人工智能的关注者传递了一个信息:不要将恐惧与事实混淆。 人性的终结——完全是无稽之谈。 摧毁一半的美国工作岗位——完全无稽之谈。
#422 🔗 原推
2026-09-11 04:11:23
Garry Tan
RT @NoPriorsPod: "We wanna bank the AIs. They deserve financial services as well." @coinbase CEO @brian_armstrong on why agentic finance means giving every AI agent its own account. Full episode linked in the replies. https://twitter.com/NoPriorsPod/status/2098241857314800038/video/1
中文: RT @NoPriorsPod:我们希望将人工智能存入银行。他们也值得获得金融服务。 @coinbase 首席执行官 @brian_armstrong 介绍为何代理金融意味着让每位人工智能代理拥有自己的账户。 完整节目在回复中链接。
🎬
视频
#421 🔗 原推
2026-09-11 04:11:14
Garry Tan
RT @danlynch: Ben Davis of @IlluminatedArts chatting with @garrytan inspiring us all to participate in culture and generosity here in San Francisco ⚡️ https://twitter.com/danlynch/status/2098255865447162061/video/1
中文: RT @danlynch:@IlluminatedArts 的本·戴维斯在旧金山与 @garrytan 聊天，激励大家参与文化与慷慨的活动。⚡️
🎬
视频
#420 🔗 原推
2026-09-11 04:05:24
Garry Tan
RT @_aj: We’ve spent the past three years building agents for real production work. Our architecture changed a lot along the way. Some of those changes worked. Others produced expensive and occasionally painful lessons. Read this to learn from our mistakes. https://x.com/i/article/2098127373837455360
中文: RT @_aj:过去三年，我们一直致力于打造真正的生产产品。我们的建筑在前进方向上发生了很大变化。其中一些改变奏效了。其他人则带来了昂贵且偶尔痛苦的教训。阅读此内容，从我们的错误中吸取教训。
#419 🔗 原推
2026-09-10 23:28:57
Garry Tan
RT @tbpn: Nori Robotics Founder @AntonioSitongLi is building 3D-printed humanoid robots for $1,688 per robot. The company is making 2-3 per day and will eventually build 400 robots every month. https://twitter.com/tbpn/status/2098148795855016136/video/1
中文: RT @tbpn:诺里机器人公司创始人@AntonioSitongLi 正在建造3D打印的人形机器人，每台机器人售价1688美元。 该公司每天生产2-3台，最终将每月生产400台机器人。
🎬
视频
#418 🔗 原推
2026-09-10 23:24:47
Garry Tan
RT @amasad: Lots of risk with AI. I worry a lot about cybersecurity for example. However, “extinction risk” — literally 100% of humans die — is not remotely one of them.
中文: RT @amasad:人工智能存在很大风险。例如，我非常担心网络安全问题。然而，“灭绝风险”——实际上是人类死亡的100%——远不在其中。
#417 🔗 原推
2026-09-10 23:23:17
Garry Tan
RT @gustaf: Demo day halftime entertainment. Ashton invested in nearly every single frontier lab. @ycombinator https://twitter.com/gustaf/status/2098154236370698393/photo/1
中文: RT @gustaf:演示日中场休息娱乐。 阿什顿几乎投资了每一个前沿实验室。@ycombinator
media 0 共 1 项
#416 🔗 原推
2026-09-10 23:23:04
Garry Tan
RT @tbpn: Olam Labs CEO @sensho says the only two bottlenecks left to AGI and ASI are compute and data. "For AGI and ASI to come there are only two bottlenecks right now. We have no bottleneck with architecture or agents. It's data and compute." "Compute is the biggest bottleneck." "We're gonna have 10x more compute coming online in the next two years. If you're scaling law-pilled and if you're -pilled in all the ways that these AI labs are, then data is the actual bottleneck."
中文: RT @tbpn:Olam Labs 首席执行官 @sensho 表示，AGI 和 ASI 仅剩两个瓶颈是计算和数据。 目前，AGI 和 ASI 的到来只存在两个瓶颈。我们在架构或代理方面没有瓶颈。这是数据和计算。 计算是最大的瓶颈。 未来两年内，我们将推出10倍的计算数据。如果你正在对法律进行推广，并且以这些人工智能实验室的所有方式进行编译，那么数据就是真正的瓶颈。
🎬
视频
#415 🔗 原推
2026-09-10 20:14:40
Garry Tan
RT @blader: introducing ═══════════ 𝚌𝚏𝚘.𝚊𝚒 ═════════════ tl;dr 1/ runway is now https://cfo.ai/ 2/ you can hire @arithecfo and he can start TODAY 3/ you can give @arithecfo a test drive by replying to this thread with any ridiculous thing you want him to model for you, and he'll come back to you with a beautiful, detailed, fully wired financial model ... in 30 minutes or less now, story time: 🧵 1/n
media 0 共 1 项
#414 🔗 原推
2026-09-10 20:13:56
Garry Tan
RT @tbpn: YC's @garrytan says hard tech startups make up a quarter of the newest 200-company Demo Day batch, up 40x since their low point ~4 years ago: "It's 200 companies presenting today. About a quarter of them are hard tech. That's up 40x since our low bar, maybe 3-5 years ago." https://twitter.com/tbpn/status/2098135496920818062/video/1
中文: RT @tbpn:YC的@garrytan表示，硬性科技初创企业占最新200家公司的“Demo Day”批次的四分之一，比4年前的低点增长了40倍: 今天有200家公司在展示。其中约四分之一是硬性技术。比我们低谷上涨了40倍，大概是3到5年前。
🎬
视频
#413 🔗 原推
2026-09-10 20:05:16
Garry Tan
RT @HenrikTaro: We built a Unix-like desktop in the browser for humans and agents. Composable tools, native DOCX/XLSX/PPTX editing and capability-level approval gates. The ambition: make entire projects delegable. Introducing the Vesence Agent-Native Desktop. https://twitter.com/HenrikTaro/status/2097794847524495402/video/1
中文: RT @HenrikTaro:我们在浏览器中为人类和代理构建了一个类Unix桌面。 可编译工具、原生DOCX/XLSX/PPTX编辑和功能级别审批门。 目标:让整个项目变得可以合法。 介绍 Vesence Agent-Native 桌面版。
🎬
视频
#412 🔗 原推
2026-09-10 20:05:07
Garry Tan
RT @VirtualElena: some takeaways from running a sota llm on a decade worth of lunches: - noted X anon @litcapital had the 4th most expensive meal on record - noted economist and MMT enthusiast @StephanieKelton had one of the cheapest - the most common restaurant was clarke's in london - and perhaps unsurprisingly, london was the most common city
中文: RT @VirtualElena:一些在十年午餐中经营一个sota lm的外卖: - 提到X anon @litcapital 是有记录以来第四顿最贵的餐食 著名经济学家兼MMT爱好者@StephanieKelton 是最便宜的之一 - 最常见的餐厅是位于伦敦的克拉克餐厅 ——也许不出所料，伦敦是最常见的城市
media 0 共 4 项 media 1 media 2 media 3
#411 🔗 原推
2026-09-10 20:03:45
Garry Tan
RT @tbpn: .@garrytan says the Jacob Coxon stuff is a smokescreen distracting us from the much more immediate, practical concerns around AI that we're facing right now: "We should be talking less about this Jacob Coxon guy, and talking a lot more about — what is actually happening with Hugging Face? Are agent swarms going to take over infrastructure en masse? And then, what are we actually doing about that?" "I don't want to hear about some guy who worked for Anthropic for 2 months. There's a coordinated effort to try to influence politicians to get a knee-jerk response out of them." "That's a smokescreen. You shouldn't be paying attention to that. We need to be paying attention to the actual things we can do to, for example, prevent agents swarms from taking over entire data centers. What's our shutdown strategy? How do we ensure provenance? Where is this agent actually located? What software can we build? What cybersecurity defenses can we build today?" "That's the level of discourse I think we need, and we just don't have that." "I don't really care about science fiction. I saw Terminator 2, too. We're not here to talk about that. We need to actually talk about what's really happening with the servers, what's actually happening with the agent swarms, and how do we actually prevent that?" "When it comes to regulation, it's like, let's pass regulation of these things we actually care about, instead of what a socialist says in the New York Times. I don't care about that."
🎬
视频
#410 🔗 原推
2026-09-10 19:24:14
Garry Tan
RT @altcap: Jensen calls Jacob Coxon comments outlandish &amp; “deeply untrue.” Said the labs are great, but his comments were wrong, arrogant, and ignorant of all the work being done around the industry to drive safety. 🧐🧐 https://twitter.com/altcap/status/2098121208537743692/photo/1
中文: RT @altcap:詹森称雅各布·考克森的言论非常古怪;“非常不实”。实验室说很棒，但他的言论是错误的、傲慢的，且对行业中为推动安全而正在开展的工作一无所知。🧐🧐
media 0 共 1 项
#409 🔗 原推
2026-09-10 19:23:04
Garry Tan
RT @PessimistsArc: https://x.com/PessimistsArc/status/1304092648953839616/video/1
🎬
视频
#408 🔗 原推
2026-09-10 19:17:55
Garry Tan
Local reporter: "Let me blame AI and tech for... *checks notes* a murder suicide where the killer *checks notes* couldn't possibly benefit from the home selling for more afterwards." What are we doing here?
中文: 当地记者:“让我把“人工智能和科技”归咎于......“检查笔记”是一起谋杀自杀事件，凶手“支票记录”在售房屋后无法再从中受益。 我们在这里做什么?
#407 🔗 原推
2026-09-10 19:16:31
Garry Tan
Landlord kills tenant. Local reporter: "How do I make AI and tech people the villain?" Come on man, sometimes people are crazy and why would you do this? https://twitter.com/garrytan/status/2098128107425435782/photo/1
中文: 房东杀害租户。 当地记者:“我该怎么把人工智能和科技的人搞坏了? 来吧，人，有时候人疯了，你为什么要这么做?
media 0 共 1 项
#406 🔗 原推
2026-09-10 19:14:56
Garry Tan
RT @bosmeny: CNBC + @garrytan filming live from the hardware floor of @ycombinator Demo Day! They are sitting within 20 feet of teams building jet-powered interceptors, military exoskeletons, humanoids, cargo rockets, drones that can move clouds... https://twitter.com/bosmeny/status/2098112593055645826/photo/1
中文: RT @bosmeny:CNBC + @garrytan 在 @ycombinator Demo Day 的硬件楼层进行现场直播! 他们正坐在距离建造喷气式拦截器、军用外骨骼、人形机器人、货运火箭、能够移动云层的无人机的队伍的20英尺范围内......
media 0 共 1 项
#405 🔗 原推
2026-09-10 18:16:27
Garry Tan
RT @koomen: Super ambitious launch from the vesence team. This is very clearly how operating systems are going to work.
中文: RT @koomen:来自 vesence 团队的超强雄心发布。操作系统的运作方式非常明显。
#404 🔗 原推
2026-09-10 18:10:42
Garry Tan
RT @cyantist: I'd love to see a product emerge that catches coordinated information campaigns in real time and flags them so the public knows what's motivating a story. This is how we crack money trails. I don't know about you, but I'm tired of seeing coordinated news stories from secret chat rooms orchestrating things for powerful people. This is a pivotal time to create such a thing and would be good for humanity and skepticism in general.
中文: RT @cyantist:我很希望看到一款能够实时获取协调信息宣传并标记它们的产品，让公众了解是什么在推动一个故事。这就是我们破解货币路径的方式。 我不知道你怎么样，但我厌倦了看到秘密聊天室里的协调新闻，为权势人物精心策划。 这是创造这种事物的关键时期，对全人类和整体怀疑态度都有好处。
#403 🔗 原推
2026-09-10 18:09:43
Garry Tan
RT @DillonLoomis: This story is fascinating and is easily one of the most important things happening at the intersection of AI and politics right now. It's critical everyone understands the situation because this is just the beginning of the AI infowars and psyops Jacob Coxon, a relatively unknown AI researcher, gives the WSJ an exclusive story about his resignation from Anthropic and his concerns about AI Twenty minutes after the WSJ story drops, Jacob posts on X about how he's concerned that AI could kill us all in a few years. This post now has over 149M views. For reference, Ilya Sutskever (a very well known AI name) left OpenAI in 2024 and his announcement post has fewer than 6M views as of today Bear in mind, Jacob Coxon's X account had no prior posting activity public, but it is immediately reposted and boosted by numerous AI-doomer policy advocacy nonprofits funded by the same donor (Jaan Tallinn) who just happens to be one of the leading investors in Anthropic We find out Jacob Coxon, who worked at Anthropic for only a few weeks, worked for a left-wing political advocacy group in 2021 (Newspeak). This organization is a British NGO with ties to hacking democracy, Hillary Clinton, prison abolition and extinction rebellion Many major Democrat politicians are now using Jacob's X post to fear monger the public into voting to regulate AI. It just so happens Bernie Sanders has a bill already written to ban artificial superintelligence (ASI) and to create a new federal agency to protect the public from the dangers of AI. This bill was announced just one week ago on September 3rd It would also place a temporary pause on "advanced AI" until a new federal regulator exists and has issued safety rules and a model-review process Elon then points out how none of this feels organic, and Jacob decides to take a shot at him for firing some xAI researchers Meanwhile, we know Anthropic has publicly asked the government to audit and block frontier models and Anthropic has run a state-by-state campaign for rules that only large AI labs trigger, meaning the rules are structured around barriers only frontier labs can easily clear (limiting competition) There is extensive Anthropic personnel and financial overlap with the groups that wrote the Biden-era AI policies. It is well known information you can seek out if you want the detail This all seemingly comes back to regulatory capture and political motives. None of this feels natural. Jacob can say these are his real views all he wants, and they may be, but there is clearly far more to the story
media 0 共 1 项
#402 🔗 原推
2026-09-10 18:09:23
Garry Tan
RT @tbpn: It's YC Demo Day. On today's show: - @garrytan (YC) - @keyserfaty (Agentcard) - @sensho (Olam Labs) - @AntonioSitongLi (Nori) - @EdwardGe (Exosat) - Alfredo Gonzalez (Omanta) See you on the stream.
中文: RT @tbpn:今天是YC演示日。在今天的节目中: - @garrytan(YC) - @keyserfaty(代理卡) - @sensho(Olam Labs) - @安东尼奥·西通利(诺里) - @EdwardGe(Exosat) - 阿尔弗雷多·冈萨雷斯(阿曼塔) 在小溪上见。
#401 🔗 原推
2026-09-10 18:04:03
Garry Tan
"I know kung fu" for your agent being able to buy stuff for you
中文: 我知道你的经纪人能为你买东西
#400 🔗 原推
2026-09-10 17:20:33
Garry Tan
RT @broukhim: Build a second brain. I recommend @garrytan’s GBrain, an open source project. There’s a real upfront investment of time, but it’s given me immense leverage to use new AI experiences (harnesses, agents, models). Own your memory &amp; context. It’s all you’ll have left 😅
#399 🔗 原推
2026-09-10 17:13:27
Garry Tan
RT @kane: Since progressives are trying to shill @saikatc again, a reminder of his story: &gt; works at hedge fund &gt; works at Stripe &gt; puts $100M into tax shelters &gt; runs for office as a socialist &gt; loses &gt; starts a dark money PAC
中文: RT @kane:既然进步派又试图在 @saikatc 上打个噱头，这提醒了他的故事: 在对冲基金工作 在Stripe工作 投入1亿美元用于避税天堂 以社会主义者的身份竞选公职 输了 启动黑钱政治行动委员会
#398 🔗 原推
2026-09-10 14:24:05
Garry Tan
RT @ClementDelangue: If they truly believe the risk is that high, and I think they do, then we need 100x more research and understanding of it. That requires them to openly share their models, datasets, training code, and agent traces.
中文: RT @ClementDelangue:如果他们真的认为风险很高，而且我认为他们确实如此，那么我们就需要再进行100次的研究和理解。这要求他们公开分享自己的模型、数据集、训练代码和代理痕迹。
#397 🔗 原推
2026-09-10 14:23:51
Garry Tan
RT @kuchaev: What terrifies me the most about current twitter storm on AI safety is the amount of US politicians jumping on a currently most trending hype train. No, there is no accepted scientific consensus (like in climate change, for example) that the probability of AI causing human extinction is even 0.01%. There is a widely accepted consensus that AI will accelate human progress (see it start solving long standing math and science questions). I expect US politicians, especially the ones supposedly representing Sillicon Valley to follow actual consensus among AI scientists and engineers and use that to inform the necessary legislation. And no, most of AI scientists do not work for one of the two companies.
中文: RT @kuchaev:当前关于人工智能安全的推特风暴最让我害怕的是，美国政客们大量跳上了目前最热门的炒作列车。 不，没有公认的科学共识(例如气候变化)，认为人工智能导致人类灭绝的概率甚至达到0.01%。人们普遍认为，人工智能将加速人类进步(参见它开始解决长期存在的数学和科学问题)。 我期望美国政界人士，尤其是那些据称代表硅谷的政客，能够遵循人工智能科学家和工程师之间的实际共识，并利用这一共识来为必要的立法提供依据。不，大多数人工智能科学家并不为这两家公司之一工作。
#396 🔗 原推
2026-09-10 14:01:52
Garry Tan
RT @ChrisHayduk: I work at OpenAI and personally think AI has been and will continue to be an extremely beneficial technology to humanity The conversation should be around how many billions of lives it will save
#395 🔗 原推
2026-09-10 13:48:38
Garry Tan
RT @alexandr_wang: muse is up to #3 on the app store! people seem to really like it, please give it a try! https://twitter.com/alexandr_wang/status/2097734603222204670/photo/1
media 0 共 1 项
#394 🔗 原推
2026-09-10 02:03:38
Garry Tan
RT @evanjconrad: a concern i have with using government as the strategy to shape ai is that it is very easy for a currently very misinformed journalist class to promote the “ban nuclear power but not the nukes” strategy to ai
中文: RT @evanjconrad:我对利用政府作为塑造新闻策略的一个担忧是，目前一个信息极不畅的记者阶层很容易推广“禁止核能”战略来推动“禁止核能”的推广
#393 🔗 原推
2026-09-10 02:02:51
Garry Tan
RT @kane: “Saikat is the most brilliant policy mind!” Saikat’s policy: “The guy who said algebra is racist should be superintendent of all California public schools. What? Of course my children go to private school.” https://twitter.com/kane/status/2097750171945947296/photo/1
media 0 共 1 项
#392 🔗 原推
2026-09-10 01:54:36
Garry Tan
RT @bscholl: AI isn’t going to wipe out humans. But AI doomers have a shot at hamstringing the most incredible advancement for humanity since the steam engine. We can’t let that happen.
中文: RT @bscholl:人工智能不会消灭人类。但人工智能的推动者有机会推动人类自蒸汽机以来最惊人的进步。我们不能让那种事情发生。
#391 🔗 原推
2026-09-10 01:54:28
Garry Tan
RT @debora_allen1: That “random BART rider” @KTVU included in the fire story… Isn’t he a long time @SamTrans employee, from one of 17 agencies getting a pile of cash if Measure RTM's sales tax increase passes?🤨 RTM's purpose is to fill operations deficits, not for infrastructure like burnt power substations. BART got $3B for capital projects in 2016 Measure RR for that.🧐 A paid spokesperson out purposely confusing the voters or just an employee worried about losing his job? 📢 Looks a lot like free campaign PR, courtesy of KTVU No? Will it be reported to FPPC? 💰 link to full report: https://www.ktvu.com/video/fmc-zgwy6ef4shxav8tc
中文: RT @debora_allen1:火堆中包含的“随机BART骑手”@KTVU吗?如果Measure RTM的增售税通过，来自17个机构之一的@SamTrans员工，这难道不是很久吗?🤨 RTM的目的是填补运营不足，而不是用于烧毁的变电站等基础设施。BART在2016年为资本项目获得了30亿美元。🧐 一位付费发言人故意让选民感到困惑，或者只是担心失去工作的员工?📢 看起来很像免费的公关活动，由KTVU No提供?会向FPPC报告吗?💰 完整报告链接:
🎬
视频
#390 🔗 原推
2026-09-10 01:22:46
Garry Tan
We should want growth and human flourishing instead of whatever is happening with this coordinated pressure campaign
中文: 我们应该希望增长和人类繁荣，而不是通过这场协调施压行动而发生的一切
#389 🔗 原推
2026-09-10 01:20:31
Garry Tan
Jane Kim: Big Talk, little respect for facts, outright falsehoods Jane Kim is a NIMBY who helped create the SF doom loop, and now she's failing up: she wants to be CA Insurance Commissioner She is no public servant. We should keep her out of office. https://sfstandard.com/opinion/2026/09/09/jane-kim-california-insurance-commissioner-myth/ https://twitter.com/garrytan/status/2097728692596486487/photo/1
media 0 共 1 项
#388 🔗 原推
2026-09-09 16:47:48
Garry Tan
RT @andrewho03: Not enough people are seriously trying to reconcile the two facts: - models can now resolve Millennium Problems - annualized US GDP growth is at 1.5%
中文: RT @andrewhot03:没有足够的人认真试图调和这两个事实: - 模型现在可以解决千年问题 - 美国GDP年增长率为1.5%
#387 🔗 原推
2026-09-09 14:02:24
Garry Tan
RT @antoniogm: For all the shit YC got for being un-American, Demo Day sure does look like an American defense tech conference these days. https://twitter.com/antoniogm/status/2097494102879461625/photo/1
中文: RT @antoniogm:对于YC因非美国人而得知的，Demo Day 确实看起来像是美国国防科技大会。
media 0 共 2 项 media 1
#386 🔗 原推
2026-09-09 13:39:06
Garry Tan
RT @joshrauh: The point made by "that Stanford economist" was that even if @KQED listeners have decided some people are too wealthy, there's a limit to how much you can tax them before inflicting harm on oneself and everyone else. That limit has long been reached. Focus on opportunity, not class warfare.
🎬
视频
#385 🔗 原推
2026-09-09 13:38:43
Garry Tan
RT @ibab: Soon those with the most compute and advanced training capabilities will outpace the rest of the world. That's why it's so important for us to develop new tools that allow everyone to participate in and benefit from AI, independently of big AI labs. Join us at https://river.ai/careers if you want to help make it happen.
中文: RT @ibab:很快，那些具备最先进计算和训练能力的人将超过世界其他地区。因此，开发让每个人都能参与并受益于人工智能的新工具对我们来说至关重要，而人工智能独立于大型人工智能实验室。如果您想帮助实现这一目标，请访问 与我们联系。
#384 🔗 原推
2026-09-09 13:37:50
Garry Tan
RT @gustaf: Legora is growing 10x year over year after passing $100M ARR earlier this spring. 🔥
中文: RT @gustaf:今年春季早些时候，乐高转次年增长率超过1亿美元。🔥
#383 🔗 原推
2026-09-09 13:37:36
Garry Tan
RT @AntWilson: why meet the fake Garry Tan when you can hear from the real one at Supabase Select! https://x.com/supabase/status/2092283000482410813/video/1
中文: RT @AntWilson:当你在Supabase Select上听到真正的消息时，为什么要见到那个假冒的Garry Tan!
🎬
视频
#382 🔗 原推
2026-09-09 05:33:48
Garry Tan
Run your own game It's going to work out
中文: 运行你自己的游戏 会解决的
#381 🔗 原推
2026-09-09 04:18:46
Garry Tan
They should actually declare this area in SF a treatment &amp; recovery sober zone All nonprofits in this area must commit to sober housing and mandatory recovery and treatment
中文: 他们实际上应该将旧金山的这一区域宣布为治疗和康复区 该地区所有非营利组织都必须承诺提供戒酒后的住房以及强制性的康复和治疗
#380 🔗 原推
2026-09-09 04:17:11
Garry Tan
RT @james406: local barber found out about Claude Code https://twitter.com/james406/status/2097399919195652223/photo/1
中文: RT @james406:当地理发师了解克劳德·科德
media 0 共 1 项
#379 🔗 原推
2026-09-09 04:15:39
Garry Tan
RT @sdianahu: congrats @UseCentralize_ on their Series A led by NEA!! they auto-build an orgchart of their customer's customer to find people who can move deals that get stuck companies like Cognition, Decagon & Exa use them a story how 2 engineers, Rachit & @_willwang, became very good at sales 1/5
#378 🔗 原推
2026-09-09 02:07:51
Garry Tan
RT @harjtaggar: Ive been using this for a couple weeks and its the rare instance of a new product from a bigco that’s obviously built by people who actually care. The browser/credentials/payments details are all really well done and the speed + accuracy of the intelligence is 👌
中文: RT @harjtaggar:我已经使用这个几周了，这是一款来自大公司的新产品的罕见例子，显然是由真正关心的人打造的。浏览器/凭证/支付细节都做得非常好，智能速度和准确性是👌
#377 🔗 原推
2026-09-09 00:14:12
Garry Tan
RT @devahaz: I don’t play favorites, just call balls and strikes here. So I must share that Zuck’s AI assistant so far has way outperformed both Instinct and ChatGPT Work on a set of 5 quite varied tasks. https://twitter.com/devahaz/status/2097411768800944195/photo/1
中文: RT @devahaz:我不玩喜欢的游戏，只需在这里打电话和击球。因此我必须分享，到目前为止，Zuck 的人工智能助手在 5 项截然不同的任务上的表现都优于 Instinct 和 ChatGPT 工作。
media 0 共 1 项
#376 🔗 原推
2026-09-09 00:14:07
Garry Tan
RT @VirtualElena: one of my favorite weekly habits is reading lunch with the ft. so i made a world of lunches every single "lunch with the ft" mapped, built in 3 hours with astra and minimal supervision https://lunch-atlas.pages.dev/ https://twitter.com/VirtualElena/status/2097466262410740197/photo/1
media 0 共 1 项
#375 🔗 原推
2026-09-09 00:07:21
Garry Tan
Harness wars are full on now and Muse is very impressive
中文: 战争已经充满，缪斯的表现非常令人印象深刻
#374 🔗 原推
2026-09-08 23:46:34
Garry Tan
RT @MiguelriosEN: I sent this to the team last night. Once again, the game has changed. https://twitter.com/MiguelriosEN/status/2097331203137556921/photo/1
中文: RT @Miguelriosen:我昨晚把这个发给了团队。游戏再次改变了。
media 0 共 1 项
#373 🔗 原推
2026-09-08 17:59:13
Garry Tan
RT @polynoamial: Yes, this result cost millions of dollars. But remember that when @OpenAI announced o3 it cost ~$500,000 to score 87.5% on ARC-AGI 1. Today, Astra scores higher for ~$20. In 2025 it took us and GDM an enormous amount of compute to achieve IMO gold. For the 2026 IMO, anyone with a $20/month ChatGPT subscription could do it. Massively scaling test-time compute gives us a glimpse of the future. I believe that a year from now everyone will have an AI at their fingertips capable of solving problems of this caliber.
中文: RT @polynoamay:是的，这个结果花费了数百万美元。 但请记住，当@OpenAI宣布在ARC-AGI 1上获得87.5%的分数时， o3 需要花费约50万美元。如今，Astra 的得分高出约 20 美元。 到2025年，我们和GDM需要大量计算才能实现IMO金牌。对于2026年的IMO，任何拥有每月20美元ChatGPT订阅的人都可以使用。 大规模扩展测试时间计算让我们窥见了未来。我相信，一年后，每个人都能拥有能够解决这种口径问题的人工智能。
#372 🔗 原推
2026-09-08 17:57:35
Garry Tan
RT @markchen90: We solved Navier-Stokes! Many of my colleagues left their fields because they believed that working on AI would be the fastest way to solve their fields' grand challenges. It's surreal to see the first of those challenges fall and to watch a future we believed in become real.
#371 🔗 原推
2026-09-08 17:44:43
Garry Tan
We need to build housing so everyone can thrive in California Yes, in my backyard
中文: 我们需要建设住房，让每个人都能在加利福尼亚茁壮成长 是的，在我的后院
#370 🔗 原推
2026-09-08 17:06:32
Garry Tan
RT @ycombinator: Introducing the YC Early Access Network. 4x a year, senior technology leaders get early looks at YC's enterprise AI companies and help shape their roadmaps. Invite-only, by application. Fall '26 kicks off Oct 22 in SF. Apply now: https://events.ycombinator.com/yc-early-access-fall-26 https://twitter.com/ycombinator/status/2097354282609566046/video/1
中文: RT @ycombinator:介绍YC早用接入网络。 每年4次，高级技术领导者会尽早了解YC的企业人工智能公司，并帮助制定其发展路线图。 仅限邀请，按应用程序提供。26年秋季将于10月22日在旧金山拉开帷幕。 立即申请:
🎬
视频
#369 🔗 原推
2026-09-08 16:28:47
Garry Tan
RT @paulg: A couple days ago I mentioned Legora's surprisingly high 9x annual growth rate. These numbers explain what's happening. They win 78% of competitive pilots, and gross retention is 95%.
中文: RT @paulg:几天前我提到了乐高(Legora)出人意料地高于年增长率。这些数字说明了正在发生的事情。他们赢得了78%的竞技飞行员，而总留任率为95%。
#368 🔗 原推
2026-09-08 13:32:05
Garry Tan
RT @daisychristo: The PISA 2025 results are out today. Some quick insights. - The major story is very significant global decline. The last PISA study was in 2022, and saw big declines in reading and maths scores from 2018. Obviously the pandemic played a part. - But this 2025 round has seen further falls even from the 2022 lows! Things haven't gone back to normal post-pandemic - they've got worse. Every generation thinks the generation below is less smart - but this time they might be right!
中文: RT @daisychristo:PISA 2025 的结果今天已公布。一些快速的洞察。 - 主要事件是全球严重下滑。上一次PISA研究是在2022年，2018年阅读和数学成绩显著下降。显然，疫情发挥了作用。 但这一轮2025年的跌幅甚至从2022年的低点进一步下降! 情况尚未恢复到疫情后的正常状态——情况变得更糟了。 每一代人都认为以下一代的聪明程度较低——但这次他们可能是对的!
#367 🔗 原推
2026-09-08 12:35:14
Garry Tan
RT @jburnmurdoch: Two extraordinary charts from today’s PISA test results: 1) School test scores continue to collapse internationally, underscoring how this is no longer a Covid effect but sustained decline. Those falls in reading and maths are equivalent to about two years of lost schooling. https://twitter.com/jburnmurdoch/status/2097277608912208119/photo/1
中文: RT @jburnmurdoch:今天PISA测试结果中的两张非凡图表: 1) 考试成绩在国际上持续下降，凸显了这已不再是冠状病毒效应，而是持续下降。 阅读和数学的下降程度相当于大约两年失学。
media 0 共 1 项
#366 🔗 原推
2026-09-08 12:34:54
Garry Tan
RT @Twolfrecovery: Clearly, the Mayor of Denver has no idea what he's talking about. Fentanyl puts you to sleep and shuts down your respiratory system. Also, calling any drug a "homelessness survival drug" as an elected official is irresponsible. @GovofCO
中文: RT @Twolfrecovery:显然，丹佛市长并不清楚自己在说什么。芬太尼会让你入睡并关闭呼吸系统。此外，将任何毒品称为“无家可归生存毒品”作为民选官员也是不负责任的。@GovofCO
#365 🔗 原推
2026-09-08 12:33:06
Garry Tan
RT @cayimby: A new UC Irvine study finds inclusionary zoning - requiring developers to include low-income homes - cuts annual housing construction by nearly a third. Each home IZ produces costs everyone else $800k in higher rents, which is more than just funding affordable housing directly. https://twitter.com/cayimby/status/2096638561466347669/photo/1
中文: RT @cayimby:一项新的加州大学欧文分校研究发现，包容性分区——要求开发商必须包括低收入家庭——将每年的住房建设削减了近三分之一。 每套房屋的房屋产生成本，其他所有房屋的租金都高出80万美元，这远不止直接为经济适用房提供资金。
media 0 共 1 项
#364 🔗 原推
2026-09-08 12:32:47
Garry Tan
RT @TheEconomist: Our analysis suggests that AI has so far created around 1m new jobs in America. We explain how the technology has created a hiring boom https://www.economist.com/finance-and-economics/2026/09/04/the-jobs-apocalypse-is-postponed-an-ai-jobs-boom-is-here?taid=6a9bda18ae1ff00001c209e7&utm_campaign=trueanthem&utm_medium=social&utm_source=twitter
中文: RT @TheEconomist:我们的分析表明，人工智能迄今已在美国创造了约100万个新工作岗位。我们解释这项技术是如何创造招聘热潮的
#363 🔗 原推
2026-09-08 05:33:28
Garry Tan
They shout “billionaires” when actually their cronies straight up loot the public coffers
中文: 他们大声喊叫“亿万富翁”时，他们的亲信直接抢占了公众的金库
#362 🔗 原推
2026-09-08 02:29:21
Garry Tan
Children need to be teach-yourself-on-YouTube-maxxing
中文: 儿童需要自学YouTube-maxxing
#361 🔗 原推
2026-09-08 01:24:11
Garry Tan
RT @levie: The AI jobs prediction so far is playing out the opposite of what many thought. The reason this is happening is because AI is bringing automation to areas that don’t have finite demand characteristics. AI allows companies to do far more than they could before, but agents still require human operators and oversight to generate value from them. And not only has AI not replaced many of the types of jobs that theoretically would have been most affected by AI - engineers or lawyers - but AI is also creating or growing entire categories of jobs across the economy. Cybersecurity, FDEs, agent operators, engineers in non-software domains, and tons of job categories are going to explode. It’s not clear if or when this trend will reverse because we keep discovering all new areas of demand that AI is going to open up.
中文: RT @levie:迄今为止，人工智能就业预测与许多人的看法截相反。 之所以发生这种情况，是因为人工智能正在将自动化引入那些没有有限需求特征的领域。人工智能使企业能够比以往做得更多，但代理机构仍需要求人工操作员和监管人员从中创造价值。 人工智能不仅没有取代许多理论上受人工智能影响最大的工作类型——工程师或律师——而且人工智能也在经济领域创造或扩大了整个就业类别。 网络安全、FDE、代理运营商、非软件领域的工程师以及大量工作类别将会激增。目前尚不清楚这一趋势是否会或何时会逆转，因为我们不断发现人工智能将开放的所有新需求领域。
media 0 共 3 项 media 1 media 2
#360 🔗 原推
2026-09-08 01:23:12
Garry Tan
RT @Noahpinion: Everyone just assumes AI is a job killer. But maybe it's not. https://www.noahpinion.blog/p/ai-keeps-stubbornly-refusing-to-take
中文: RT @Noahpinion:每个人都只是认为人工智能是一种工作杀手。但也许不是。
#359 🔗 原推
2026-09-08 01:18:29
Garry Tan
RT @blader: ai remains extremely early, underhyped, and undervalued when we're perpetually online agent-maxxxing with all our MCPs connected to agents updating your gbrain we end up not talking to enough normal people who are still copy and pasting chatgpt answers into their word copilots
#358 🔗 原推
2026-09-07 22:39:02
Garry Tan
RT @peteskomoroch: Each red dot represents a San Francisco bus shelter that had its glass smashed over the last 6 months, zero arrests so far:
中文: RT @peteskomoroch:每个红点代表一个旧金山公交避难所，该避难所在过去6个月中玻璃被砸碎，目前零人被捕。
#357 🔗 原推
2026-09-07 22:35:03
Garry Tan
RT @lennysan: .@illscience: "Moats are discovered, not designed." https://twitter.com/lennysan/status/2097009793399742965/video/1
🎬
视频
#356 🔗 原推
2026-09-07 19:09:07
Garry Tan
RT @maxhodak_: i can't tell you how many times people i initially thought were crazy ended up actually doing it at this point "everything works" feels like a much more reasonable heuristic than the alternative such missed returns lol
中文: RT @maxhodak_:我无法告诉你，最初我以为有多多人疯了 此时“一切都行得通”，感觉比另一种方式更合理 如此漏报的回报哈哈
#355 🔗 原推
2026-09-07 19:06:39
Garry Tan
RT @JasonMAtwood: If you want to use @garrytan GBrain on @bot I suggest a few things. 1) Set it up as its own bot and instruct all other bots to send relevant information to it. 2) Tell all bots to reference Brain when answering questions or doing work (this allows for context) 3) Have Brain export all .md files into a syncing folder on your local machine for some backup. 4) Have it store the GBrain in Github (as per Garry's instructions) 5) When mining for additions to the Brain, give it narrow focus on a person, place or thing, and the tools to go find it (email/calendar/files/folders) 6) Once you have done a bunch of mining asking it some fun questions like "What are the 10 most interesting facts you know about Jason Atwood". 7) Have it do a public mine for inclusion in the brain as well for people/places. 8) Set it to do daily and weekly sweeps of your functional inboxes (Slack/Messages/Email/Calendar) for updates.
#354 🔗 原推
2026-09-07 18:31:23
Garry Tan
RT @kobyjconrad: One final graph before we present at S26 DD 🥹 since launching cash pay naltrexone (~$500 ltv) in the "single" state of PA on aug 16th, this is our WoW intake payments growing 50% WoW this individual product line is going to be an absolute monster with a 50 state PC 👀 https://twitter.com/kobyjconrad/status/2097008514795450809/photo/1
中文: RT @kobyjconrad:在S26 DD 上展示之前的最终图表🥹 自8月16日在宾夕法尼亚州“单一”状态下推出现金支付纳曲酮(约500 ltv)以来，这是我们的WoW收款款 增长50% 这款独立的产品线将绝对具有50个州的PC👀
media 0 共 1 项
#353 🔗 原推
2026-09-07 18:31:11
Garry Tan
RT @harjtaggar: I think this will turn out to be an extremely presicient call by the end of the year
中文: RT @harjtaggar:我认为到今年年底，这将是一个极具预见的号召
#352 🔗 原推
2026-09-07 16:10:48
Garry Tan
It’s fair to say RFS are just hunches on what might be next, and conversation starters for people starting out. The main determiner of a startup idea’s success are the specific founders building specific tech for specific customers. The rest is categorization.
中文: 可以公平地说，RFS只是对接下来可能发生的事情的预感，而对于刚开始的人来说，对话就是开始的。 创业想法成功的主要决定因素是特定创始人为特定客户构建特定技术。其余部分为分类。
#351 🔗 原推
2026-09-07 15:33:35
Garry Tan
RT @ycombinator: Harnesses often get dismissed as just scaffolding, just prompt engineering, and not real research. But that couldn't be farther from the truth. The same model weights that score 30% on ARC-AGI score 95% with a better harness. So we gathered a group of researchers and founders working at the frontier to do a deep dive into the state of harnesses. We cover how we got to this point, the case for making your harness as expressive as possible, and what YC learned building an agent for every employee in the company. 00:00 - @FrancoisChauba1: Why harnesses matter 04:27 - Building an auto-researcher by accident 07:13 - A five minute history of harnesses 13:56 - Self-improving harnesses 18:35 - @sethkarten: Prime Agent, a self-improving RLM harness 21:50 - Context as an L1, L2, L3 cache 24:51 - From Turing machine to von Neumann computer 28:33 - Messaging between agents 30:04 - ARC-AGI results 33:09 - Emulator Bench and GPU kernels 37:30 - @JonSaadFalcon: OpenJarvis, personal AI on personal devices 38:26 - How far behind are local models 39:21 - The five primitives of a personal AI stack 42:47 - Letting cloud models optimize your local stack 43:53 - 800x cheaper than the cloud 45:58 - @josh__france and @jbellregan: QM, YC's agent harness for work 47:29 - A history of YC's internal agents 49:24 - OpenClaw and a fleet of 50 agents 51:04 - Pulling the brain out of the sandbox 54:43 - Letting the agent choose its own sandbox and model 57:16 - The grind tool: budgets on goals 58:50 - Agents don't understand social context
🎬
视频
#350 🔗 原推
2026-09-07 15:30:40
Garry Tan
RT @amasad: I don’t think we’ve reached AGI but what we have is functionally indistinguishable from AGI. Because we have a relentless programmer that doesn’t get bored or tired. So any problem that can be casted as a coding problem is virtually solved. https://twitter.com/amasad/status/2096936109817135331/photo/1
media 0 共 1 项
#349 🔗 原推
2026-09-07 13:55:28
Garry Tan
RT @illscience: lovely + dense convo w @lennysan re: "what if it works": - decentralization + everything is working: this product cycle has more winners and more ways to win, yet many in SV are oddly nursing a dark fantasy about being outside the lightcone - climbing the ambition ladder: shipping your bad ideas to discover your good ideas + hyper-local expressions of ambition like making the NHS as good as an iPhone - changing the narrative around machine intelligence: making important things like education and healthcare cheap, unbundling learning from institutions and status from credentials - what's working in consumer: coding agents, personal agents like Town / Instinct / Grok Bots / ChatGPT Work, and entertainment across creativity / gaming / companionship - coding as an activity vs an outcome: perhaps coding is more like reading than writing; a fulfilling way to learn / explore / spend time more than a goals based task loved the conversation + love to hear what you think!
中文: RT @illscience:可爱多致，@lennysan 重:“如果行动怎么办”: - 去中心化+一切正在发挥作用:这个产品周期拥有更多赢家和更多获胜方式，但许多SV行业者却奇怪地在接受一种关于脱离光锥的黑暗幻想 - 攀登雄心阶梯:送上你的坏主意，去发现好的想法，以及让国民医疗服务体系像iPhone一样优秀的超本地表达 - 改变机器智能的叙事:让教育和医疗等重要事物变得廉价，从机构中分拆学习，从资质中获得地位 - 消费者的工作内容:编程代理、Town / Instinct / Grok Bots / ChatGPT Work 等个人代理，以及创意/游戏/陪伴娱乐 - 将编程作为一项活动与结果进行编码:或许编程更像是阅读，而不是写作;一种充实的学习/探索/花费时间的方式，而不是基于目标的任务 喜欢这次谈话，也喜欢听你的想法!
#348 🔗 原推
2026-09-07 13:54:44
Garry Tan
RT @Overlap_Tech: Paul Graham: The Next Trillion-Dollar Company Will Have Formidable Founders⁣ ⁣ "The answer to your question where the next giant company comes from is it comes from the right founders."⁣ ⁣ "It's not some particular idea... it comes from the next trillion dollar founders is where it comes from. And they probably have good ideas." — @paulg
中文: RT @Overlap_Tech:保罗·格雷厄姆:下一家万亿美元公司将拥有可创始人 ⁣ 下一个大型公司来自哪里的答案，就是来自合适的创始人。 ⁣ 这并非特别的想法......它源于下一代万亿美元的创始人，正是它的来源所在。而且他们可能有好的想法。——@paulg
🎬
视频
#347 🔗 原推
2026-09-07 13:46:29
Garry Tan
RT @aphysicist: this gets more true every day https://twitter.com/aphysicist/status/2096401321876140151/photo/1
中文: RT @aphysicist:每天都会这样
media 0 共 1 项
#346 🔗 原推
2026-09-06 16:47:44
Garry Tan
CRISPR is amazing https://medicalxpress.com/news/2026-09-crispr-gene-therapy-safely-lowers.html
中文: CRISPR 非常出色
#345 🔗 原推
2026-09-06 16:47:19
Garry Tan
RT @ibab: The River AI team is co-hosting the Own Your Intelligence Hackathon with YC. Drop by to learn how you can finetune your own models on your data. We’ll give out free credits for our training API at the event.
中文: RT @ibab:The River AI 团队将与 YC 共同举办 Own Your Intelligence Hackathon 活动。通过学习，了解如何在数据上微调自己的模型。我们将在活动上免费提供培训API的积分。
#344 🔗 原推
2026-09-06 14:43:04
Garry Tan
RT @Suhail: 🎯 There are very few original startup ideas. It’s mostly adapting and learning from the market as you go. Most often you need to prepare for the moment especially these days.
中文: RT @Suhail:🎯 原创创业想法很少。主要是在你去的时候从市场中适应和学习。大多数时候，你尤其需要为当下做好准备。
#343 🔗 原推
2026-09-06 14:42:59
Garry Tan
RT @omooretweets: The state of the tech press is really depressing A NY Times reporter tried Wispr Flow and admitted its speed/accuracy was "astonishing" ...and then launched into a diatribe about wanting to "murder [it] with a hammer," because people shouldn't be allowed to use voice dictation https://twitter.com/omooretweets/status/2096417181001245112/photo/1
中文: RT @omooretweets:科技媒体的现状令人非常沮丧 一名《纽约时报》记者尝试了Wispr Flow，并承认其速度和准确性“令人惊讶” ......然后开始发表一篇关于想用锤子杀人的讽刺，因为不应允许人们使用语音听写
media 0 共 1 项
#342 🔗 原推
2026-09-06 14:42:41
Garry Tan
As I use Aside I think it really might be the best AI agent harness Supports all models, extensible, has skills support built in, comes with lots of useful skills already Amazing a small startup can punch above even what the frontier labs can do at the harness browser level
中文: 当我使用Aside时，我认为它可能确实是最佳的AI代理线束 支持所有模型，可扩展，内置技能支持，已具备大量实用技能 一家小型初创企业甚至能超越前沿实验室在线束浏览器层面所能做的事情，真是令人惊叹
#341 🔗 原推
2026-09-05 17:09:42
Garry Tan
You need to build and own your own intelligence: your own agent, your own models, your own memory Join us for the YC Own Your Intelligence Hackathon September 27 in San Francisco https://events.ycombinator.com/gstack-qm-river-memorable-hackathon https://twitter.com/garrytan/status/2096283493764665350/photo/1
中文: 你需要建立并拥有自己的智慧:你自己的代理人、自己的模型，你自己的记忆 参加9月27日在旧金山举行的YC自有智能黑客马拉松活动
media 0 共 1 项
#340 🔗 原推
2026-09-05 17:05:06
Garry Tan
RT @KTmBoyle: If you can keep your head when all about you are losing theirs and blaming it on you If you can trust yourself when all men doubt you, but make allowance for their doubting, too. If you can wait and not be tired by waiting, or being lied about, don’t deal in lies, Or being hated, don’t give way to hating. And yet don’t look too good, nor talk too wise.
中文: RT @KTmBoyle:如果你的一切都在失去他们，你只能继续关注，那就归咎于你 当所有男人都怀疑你时，你可以相信自己，但也要为他们的疑虑留出余地。 如果等待或被欺骗，可以等待，不要感到厌倦，不要说谎。 或者被憎恨，不要屈服于仇恨。然而，不要看起来太好，也不要说太明智。
#339 🔗 原推
2026-09-05 16:44:01
Garry Tan
RT @Overlap_Tech: Paul Graham: YC as a Credential Is a Terrible Idea⁣ ⁣ "If you want to seem cool, starting a startup is just about the least efficient way to do it." — @paulg https://twitter.com/Overlap_Tech/status/2096023274324918413/video/1
中文: RT @Overlap_Tech:保罗·格雷厄姆:作为凭证的YC是个可怕的想法 ⁣ 如果你想看起来很酷，创办一家初创公司就是最不高效的方式。——@paulg
🎬
视频
#338 🔗 原推
2026-09-05 16:41:53
Garry Tan
RT @reason: WATCH: @MightyHeaton hates the reasons you hate data centers. https://twitter.com/reason/status/2095924336150352165/video/1
中文: RT @reason: 关注:@MightyHeaton 讨厌你讨厌数据中心的原因。
🎬
视频
#337 🔗 原推
2026-09-05 13:43:29
Garry Tan
RT @hiddnest: what people don't know about aside &gt; it can sign into websites and get things done &gt; learns your browsing context to complete tasks - no more long prompt &gt; write messages that sounds like you it's not just a browser
中文: RT @hiddnest:人们对此并不了解 可以登录网站并完成操作 学习浏览环境以完成任务——不再有长时间的提示 写听起来像你的留言 不仅仅是一个浏览器
#336 🔗 原推
2026-09-05 13:40:22
Garry Tan
RT @follow_clues: 基于 @garrytan 的 GBrain 做了一个期权工作台： 1.把交易历史、复盘教训、策略框架、读过的笔记和 PDF 全部装进一个本地知识库。之后每一步都从它出发： 2.投研报告：实时期权链 + 你自己的框架和这只标的的历史交易 → 结论先行 3.交易前审核：AI 拿你写下的纪律逐条核对你的计划 4.平仓后：教训自动沉淀，bet 自动结算，Brier 分数告诉你判断准不准 5. 问答：先查你的大脑再答，标注依据来自哪一页 用得越久，它越像你自己。开源，美股 + 币安期权，一个 DeepSeek key 即可。 Mac用户可下载客户端 https://github.com/Theclues/TradeGenuis-Options
#335 🔗 原推
2026-09-05 13:39:56
Garry Tan
RT @kul: @NWischoff wow. investors are paying 520x ARR for YC S26 companies and 47x for everyone else. An 11x premium on traction. If you're a founder deciding whether YC is worth it, this is your answer, and it doesn't count the network or the time you get back from a 10 day raise.
中文: RT @kul:@NWischoff 哇。 投资者为YC S26公司支付520倍的ARR，为其他所有公司支付47倍的费用。牵引力的11倍溢价。 如果你是创始人，决定YC是否值得，这就是你的答案，而且它不计入网络，也不算你从10天加薪中恢复的时间。
#334 🔗 原推
2026-09-04 22:47:50
Garry Tan
RT @paulg: A VC reports that YC startups' valuations are 79% higher than those of non-YC startups. This doesn't mean, of course, that a given company's valuation will be 79% higher if they do YC. The YC companies could also be better. But either way it's clearly not a bad brand to have.
中文: RT @paulg:一家风险投资公司报告称，YC初创企业的估值比非YC初创企业高出79%。这并不意味着如果一家公司进行YC，其估值将高出79%。YC公司也可能变得更好。但无论哪种方式，它显然都不是一个糟糕的品牌。
#333 🔗 原推
2026-09-04 22:47:38
Garry Tan
RT @paulg: Legora's revenue has increased by over 9x from this time a year ago. That's an astounding growth rate for a company of this size. It's not unusual for a startup to grow that fast in the first 6 months, but Legora is 3 years in.
中文: RT @paulg:与去年同期相比，Legora的收入增长了9倍多。对于如此规模的公司来说，这是一个惊人的增长率。初创公司在前六个月发展得如此快并不罕见，但Legora已经三年了。
#332 🔗 原推
2026-09-04 22:21:52
Garry Tan
Absolutely insane experience using @AsideAI as an AI harness - trying to setup OpenClaw with Slack was a 2 hour experience. Aside's harness with full integrations AND browser integration did it in less than 3 minutes, and it's fully featured with smart access control defaults https://twitter.com/garrytan/status/2095971990645755941/photo/1
中文: 使用 @AsideAI 作为人工智能工具的绝对疯狂体验——尝试使用 Slack 来设置 OpenClaw，是一次2小时的体验。 此外，完全集成和浏览器集成的线束在不到3分钟内就实现了，并且已完全配备智能访问控制默认功能
media 0 共 2 项 media 1
#331 🔗 原推
2026-09-04 20:27:18
Garry Tan
RT @sdianahu: OSS models are quite viable we’ve seen not just large F100 adopt it but also many of the s26 startups use it for coding too thx for sharing inaights @jmorgan
中文: RT @sdianahu:OSS 模型相当可行 我们看到不仅有大型F100采用它，还看到许多S26初创公司也将其用于编程 分享 inaights 的 @jmorgan
#330 🔗 原推
2026-09-04 18:57:16
Garry Tan
RT @vivilinsv: Tremendous respect for @garrytan (with @garryslist ). He is known as the president and CEO of @ycombinator, co-founder of Initialized Capital and Posterous, and an early investor in companies including Coinbase and Instacart. By almost any Silicon Valley measure, his résumé is extraordinary. But what moved me most during this @a16z interview (link in comment) was his heart and and his sense of duty to his community. Garry described how, during COVID, San Francisco’s district attorney was, in his view, turning a blind eye to heinous crimes against Asian elders. At the same time, the school board was hostile toward Asian American children who simply wanted access to algebra in public middle school. He also recalled that mainstream Bay Area television—including the local ABC channel—would not cover anti-Chinese and anti-Asian crimes. “Our institutions failed us,” he said. But the Asian American community did not remain silent. Chinese families on San Francisco’s west side organized around access to algebra and the deaths of their elders. They went to farmers markets and community events, speaking to one another face-to-face, making sure their neighbors understood what was happening—and uniting to demand change. That is how a community finds its voice. Garry’s message was clear: when people like us stop paying attention, refuse to speak up and fail to hold both government and media accountable, things can become deeply distorted. Speaking out can carry personal and reputational costs. But, as he put it, this is part of our covenant with our fellow citizens: “If we don’t do it, who’s going to do it?” As a member of the Asian American community, I deeply respect Garry for using his voice, influence and resources to stand up for our elders, our children and the city he calls home. That is more than politics. It is courage, solidarity - and heart. ❤️
中文: RT @viviilinsv:对@garrytan(与@garryslist)的尊重。 他以@ycombinator的总裁兼首席执行官而闻名，是Initialized Capital和Posterous的联合创始人，也是包括Coinbase和Instacart在内的公司的早期投资者。 几乎任何硅谷指标都是他的简历非同寻常。但在这次@a16z采访中，最让我感动的是他内心的心和 以及他对社区的责任感。 加里描述了在感染COVID期间，旧金山地区检察官在他看来，对针对亚裔长辈的滔天罪行视而不见。与此同时，学会董事会对那些仅仅希望在公立初中学习获得代数的亚裔美国儿童持敌对态度。 他还回忆说，包括当地ABC频道在内的主流湾区电视台不会报道反华和反亚裔犯罪。 “我们的机构辜负了我们，”他说。 但亚裔美国人群体并没有保持沉默。旧金山西侧的中国家庭围绕获取代数和长辈死亡问题进行了组织。他们前往农贸市场和社区活动，面对面交谈，确保邻居们了解正在发生的事情，并联合起来要求改变。 一个社区就是这样找到自己的声音的。 加里的信息很明确:当像我们这样的人停止关注、拒绝发声，而且无法同时让政府和媒体承担责任时，事情就会变得严重扭曲。 说话可能带来个人和声誉方面的代价。但正如他所说，这是我们与同胞们的盟约的一部分: 如果我们不做，谁来做这件事? 作为亚裔美国社区成员，我深切尊重加里，感谢他利用他的声音、影响力和资源，为我们的长辈、我们的孩子以及他称之为家乡的城市挺身而生。 这不仅仅是政治。是勇气、团结，也是内心的。❤️
🎬
视频
#329 🔗 原推
2026-09-04 18:57:03
Garry Tan
GStack now uses @AsideAI browser as it's preferred remote session browser Does your AI agent need access to web and credentials as you? Aside is now the #1 absolute best way to do it I've found out of all AI agent tools. Aside is a/ an amazing browser, b/ great at credentials, c/ has all the integrations you need and c/ actually an amazing harness and memory system in its own right It's my new favorite AI browser
中文: Gstack 现在使用 @AsideAI 浏览器作为首选的远程会话浏览器 你的人工智能代理需要像你一样访问网页和凭据吗?此外，现在是我从所有人工智能代理工具中发现的第一个绝对最佳方法。 旁边是 a/ 一个出色的浏览器，b/ 非常擅长凭据，c/ 拥有您所需的所有集成，实际上还拥有出色的线束和内存系统 这是我最喜欢的新AI浏览器
media 0 共 1 项
#328 🔗 原推
2026-09-04 18:54:43
Garry Tan
RT @ycombinator: 🦙 @ollama is used by 9 million developers and 85% of the Fortune 500, giving co-founder and CEO Jeffrey Morgan (@jmorgan) a unique view into which AI models people are actually using and how that’s changing. Right now, the biggest shift he sees is toward open models, driven by coding agents, falling costs, and capabilities that are rapidly catching up to the frontier labs. On Ollama Cloud, that shift has driven a 150x increase in token usage since the start of the year. In this episode of @LightconePod, Jeff joins @garrytan, @snowmaker, @sdianahu, and @harjtaggar to talk about the future of open models and the story behind Ollama, from two years of searching for the right idea to building one of the most widely used AI developer tools in the world. 00:43 — The Shift to Open Models 03:03 — How AI Agents Are Driving Token Usage 05:31 — Are Open Models Catching Up? 08:26 — What Happens When a New Model Launches 11:31 — Ollama as an Operating System for AI 14:05 — The New Opportunities Above the Model Layer 18:19 — Why 80–90% of Enterprise Tokens Could Be Open 20:57 — The Future Is Local and Cloud 26:40 — Why AI Is Coming Back to Your Computer 28:56 — The Coming Era of Unlimited Tokens 32:30 — Do We Still Need a “God Model”? 33:41 — Open Models and Geopolitics 36:14 — The Origins of Ollama 40:36 — Two Years Lost in the Wilderness 42:39 — The Pivot That Changed Everything 47:02 — How Ollama Found a Business Model 49:43 — Why Second-Time Founders Did YC
中文: RT @ycombinator:900万开发者使用了@ollama，拥有财富500强的85%，这使得联合创始人兼首席执行官杰弗里·摩根(@jmorgan)能独占人们实际使用的人工智能模型以及其变化方式。 目前，他看到的最大转变是面向开放型，其驱动力是编码代理、成本下降以及迅速赶上前沿实验室的能力。在奥拉马云上，这一转变导致自今年年初以来代币使用量增长了150倍。 在 @LightconePod 的这一集中，杰夫与 @garrytan、@snowmaker、@sdianahu 和 @harjtaggar 共同讨论开放式模型的未来，以及 Ollama 背后的故事。从两年来寻找正确创意，到打造全球使用最广泛的人工智能开发者工具之一。 00:43 — 转向开放式模型 03:03 — 人工智能代理如何推动代币使用 05:31 — 开放式模型正在追赶吗? 08:26 — 新模型发布时会发生什么 11:31 — 奥拉玛作为人工智能的操作系统 14:05 — 模型层上方的新机遇 18:19——为何80%至90%的企业代币可以开放 20:57 — 未来是本地和云端 26:40——为何人工智能将重返你的电脑 28:56 — 无限代币的即将到来时代 32:30 — 我们还需要一个“上帝模型”吗? 33:41 — 开放模型与地缘政治 36:14 — 奥拉玛的起源 40:36 — 荒野中迷失了两年 42:39 — 改变一切的支点 47:02 — 奥拉玛如何找到商业模式 49:43——为何第二次创始人做了YC
🎬
视频
#327 🔗 原推
2026-09-04 18:53:55
Garry Tan
Grok images quite impressive tbh
#326 🔗 原推
2026-09-04 03:24:26
Garry Tan
RT @Overlap_Tech: Paul Graham: The Real Advantage Of YC Isn't The Money⁣ ⁣ "Starting a startup is normally a very lonely business... this is like working in an office with all these colleagues, except you're all working on your own startups. You can also learn from what the other startups are doing at the same time you are." — @paulg
中文: RT @Overlap_Tech:保罗·格雷厄姆:YC的真正优势并非金钱 ⁣ 创办一家初创公司通常是一项非常孤独的事业......这就像在办公室里和所有这些同事一起工作，只有你们都在自己创业。你也可以从其他初创企业的同时学习。——@paulg
🎬
视频
#325 🔗 原推
2026-09-04 03:21:34
Garry Tan
RT @AustinJustice: Flock’s new enemy is the "No Kings" protest organization. Groups funded by a Shanghai-based tycoon, including CodePink, have spent the past year fusing Flock opposition to their campaign against AI data centers. Just like the DSA, anti-ICE, and CAIR’s social media campaigns, it's working. An independent analysis shows that their campaign talking points are now repeated by conservative and libertarian accounts. The coalition against tech that cops have used since the 1980s started on the left but is now suddenly broad. But there is little doubt who's in the driver's seat, and who is being driven.
中文: RT @AustinJustice:Flock 的新敌人是“无国王”抗议组织。 由一位总部位于上海的大亨(包括CodePink)资助的团体在过去一年中一直将反对Flock组织反对其人工智能数据中心的行动联系起来。 就像美国运输安全管理局(DSA)、反美(ICE)以及CAIR的社交媒体宣传活动一样，它正在发挥作用。一项独立分析表明，他们的竞选讨论观点现在被保守派和自由意志主义者所重复。 自20世纪80年代以来，警察一直使用的反对科技的联盟从左翼开始，但现在却突然变得广泛。但毫无疑问，谁在驾驶座上，谁在驾驶。
media 0 共 1 项
#324 🔗 原推
2026-09-04 03:21:29
Garry Tan
People who never had proper jobs or ever ran businesses can’t figure out that markets exist “Prices, how do they work?”
中文: 从未拥有过合适工作或经营过企业的人，都无法弄清楚市场的存在 价格，它们是如何工作的?
#323 🔗 原推
2026-09-04 02:11:52
Garry Tan
RT @CharlesFLehman: Massie's bill would deprive cops not only of fixed ALPRs like Flock, but of *mobile* ALPRs—cameras on cars that just serve to more reliably identify a vehicle. That would basically push cops back to the 20th century. No mobile ALPRs makes it harder to check hotlists and BOLOs. Harder to confirm if a suspicious car is stolen. British police have had mobile ALPRs since the 1980s. Americans since the 1990s. What are we doing here?
中文: RT @CharlesFLehman:Massie 的法案不仅会剥夺警察使用 Flock 等固定式 ALPR 的费用，还会剥夺其在汽车上的摄像头，这些摄像头不仅能更可靠地识别车辆。 这基本上会将警察推回到20世纪。 没有移动的ALPR会更难查看热门名单和BOLO。更难确认是否有可疑车辆被盗。 自20世纪80年代以来，英国警方一直使用移动性ALPR。自20世纪90年代以来的美国人。我们在这里做什么?
#322 🔗 原推
2026-09-04 02:02:37
Garry Tan
RT @Overlap_Tech: Paul Graham: AGI Isn't a Finish Line, It's a Smear⁣ ⁣ "We thought AGI would be this finish line. We'd cross it, it would be very clear. But when you're standing on the finish line, you realize it has width... The finish line is actually a smear, and we're on it. That's why you can't get your restaurant times."
🎬
视频
#321 🔗 原推
2026-09-04 02:02:22
Garry Tan
#NewProfilePic Photo by Oliver Covrett https://twitter.com/garrytan/status/2095689046253355055/photo/1
中文: #新配置文件图片 照片由奥利弗·科夫雷特拍摄
media 0 共 1 项
#320 🔗 原推
2026-09-04 01:42:59
Garry Tan
RT @rdominguezibar: JUST IN: Garry Tan says Paul Graham's oldest advice just got a multiplier. ▫️ "Do things that don't scale" still applies ▫️ Agents let one founder do it at scale for the first time The advice never changed. The physics of startups did. Check the full analysis here 👇https://www.the-ai-corner.com/p/garry-tan-personal-agi
中文: RT @rdominguezibar:刚进去:加里·谭说，保罗·格雷厄姆最古老的建议只是获得了一个乘数。 ▫️ “做那些不量的事情”仍然适用 ▫️ 代理首次让一位创始人大规模完成 建议从未改变。初创企业的物理学做到了。 请在此处查看完整分析 👇
🎬
视频
#319 🔗 原推
2026-09-04 01:13:24
Garry Tan
Lily Ho is a fighter for San Francisco and this is her Californian Dream
中文: 莉莉·何是旧金山的斗士，这是她的加州梦
#318 🔗 原推
2026-09-04 00:17:35
Garry Tan
RT @rahul: Sure, here are the answers: 1. Y Combinator was our first investor in 2017. They've invested in over 5,600 companies. 2. Just one example of crime dropping because of Flock? Okay, here's a random one. Delano Police deployed 20 Flock cameras (eventually growing to 45) in 2023. In a year, the Chief stated that "because of the Flock cameras", the department saw a 56% reduction in crime (source: https://t.co/Z9EboxarW6) 2. If by "mass surveillance" you mean "unconstitutional/illegal surveillance", then I believe it becomes unconstitutional/illegal when it actually violates the fourth amendment. All courts that have heard ALPR cases have ultimately deemed that ALPR does not violate the fourth amendment. People on the internet may disagree but the courts have been clear. Here are a bunch of the cases I'm referencing (https://t.co/LwIHHwW7yN) 4. Flock clearly improves crime clearance rates. There have been multiple studies done on this + multiple agencies who have published their crime clearance improvement stats across things like homicide, robbery, auto theft, etc. Here's the most recent one from 3 weeks ago (https://t.co/GiXcGV9ih6) 5. I don't believe one exists, but you're more than welcome to create one. Error can be documented through agency audits and public records. Wrongful stops go to zero when an agency follows policy. Austin PD did a study in 2025 that showed that they had zero incorrect stops after 13,122 ALPR matches because they correctly verified each hit before pulling over the vehicles. This makes sense, since a simple ALPR alert by itself is not enough to detain someone. The good news is that this precision still greatly exceeds that of human beings playing the game of telephone, trying to find the suspect vehicle. 6. Concerns regarding privacy are reasonable. These concerns are settled by courts, and fully settled by community desires and configurations. Again, courts have ruled consistently re: ALPR and fourth amendment. Also, each community can go to their local city council and voice their opinions on how they would like their public safety strategy to be run. Some will choose more safety, others will choose less. It's up to them. 7. There is no reliable published Flock-only total for the past three years. Several recent alleged abuses were surfaced through audit reviews and led to arrests. Bad cops have been routinely caught after Flock's auditing assistance went live and showed who they all were. That was the point. I wish every police database did this, since the other databases are routinely abused with no auditing guardrails. We're hoping legislation forces all databases to have Flock's level of auditing. 8. There are many safeguards in place regarding Flock misuse. Case numbers are required for each search, search reasons are required, every search is logged and audit assistance automatically flags anomalies, potential misuse actors are proactively locked out. This is why you see so many news stories about police misuse of Flock vs other police databases, because Flock is the only one catching people who misuse their tech. Cops who run plates for friends via DMV, run their dauther's boyfriend's criminal history, etc, rarely get caught. And also, the same process that holds a police officer accountable for misusing their firearm, their lights and sirens, etc, still exists for Flock misuse as well. 9. Several layers prevent Flock from repurposing customer data. Customers own and control their data, per their contracts. The contracts restrict Flock's use. Data retention schedules automatically dictate deletion of the data, based on agency chosen configurations. Agencies control their data sharing. All searches are logged. 10. Giving police a 24/7 GPS tracker for every citizen is an exaggerated example designed to make people think it's the same thing as stationary cameras at select intersections taking pictures of vehicles (not people). Per the courts who have opined, doing the former is a clear violation of the fourth amendment, but the latter is not.
#317 🔗 原推
2026-09-04 00:15:19
Garry Tan
RT @thedailyblock: 🚨BREAKING: China is funding a nationwide activist ecosystem against data center buildout and Flock cameras. A new investigative report traces millions of dollars through organizations and influencers connected to opposing data centers and Flock cameras. https://twitter.com/thedailyblock/status/2095531942984696232/photo/1
中文: RT @thedailyblock:🚨 突发:中国正在资助一个全国性的维权生态系统，以抵御数据中心建设和Flock摄像机。 一份新的调查报告通过与对立的数据中心和Flock摄像头相关的组织和网红，追溯了数百万美元。
media 0 共 1 项
#316 🔗 原推
2026-09-03 23:05:29
Garry Tan
RT @gustaf: Elon's 2012 MIT dream of electric aviation is coming to life. The Heart team comes from SpaceX and Tesla, so they learned from you. https://www.youtube.com/watch?v=nM86DBOqgPM
中文: RT @gustaf:埃隆2012年麻省理工学院对电动航空的梦想即将实现。心脏团队来自SpaceX和特斯拉，因此他们从你那里吸取了教训。
#315 🔗 原推
2026-09-03 22:37:14
Garry Tan
RT @ycombinator: At @HeartAerospace, @AndersForslund1 built the company one material milestone at a time, using each one to earn the credibility and capital to build the next. “We really had no technology. I came to YC Demo Day with a little 3D-printed model.” “We went from having the LOIs to building the 400 kilowatt motor, which was the size of a small jet engine. That in turn attracted actual pre-orders, and then that attracted more capital.” Their first major airline customer came through the spam folder. “United Airlines came through an inbound email through our info email... we showed them the electric motor we built, and they kind of immediately got it.”
中文: RT @ycombinator:@HeartAerospace 的 @AndersForslund1 一次为公司打造了一个重要里程碑，利用每一个里程碑来赢得信誉和资本，以构建下一个重要里程碑。 我们真的没有任何技术。我带着一个小型3D打印模型来到YC演示日。 我们从拥有“路易斯”发动机到制造了400千瓦的发动机，该发动机的体积相当于一台小型喷气发动机。这反过来吸引了实际的预购，进而吸引了更多资本。 他们的第一位主要航空公司客户通过垃圾信息文件夹前来。 联合航空通过我们的信息邮件发送了一封来往的邮件......我们向他们展示了我们制造的电动机，他们立刻就拿到了。
🎬
视频
#314 🔗 原推
2026-09-03 22:36:31
Garry Tan
RT @arcprize: GPT-6 Astra by @OpenAI achieves SOTA on ARC-AGI: - Astra scores 63% on ARC-AGI-3, 99% via a new provider adapter harness - It surpasses human performance on 96% of ARC-AGI-3 levels - It builds the most precise symbolic model of novel environments we've seen Our analysis: https://twitter.com/arcprize/status/2095597602545025138/photo/1
中文: RT @arcprize:@OpenAI 的 GPT-6 Astra 在 ARC-AGI 上实现了 SOTA: - Astra 在 ARC-AGI-3 上得分为 63%，通过新的电源适配器线束获得 99% 的评分 - 在 ARC-AGI-3 水平的 96% 上，它超过了人类的表现 - 它构建了我们所见过的最精确的新环境符号模型 我们的分析:
media 0 共 1 项
#313 🔗 原推
2026-09-03 21:13:08
Garry Tan
RT @ashleevance: "Claude, what's the best way for me to do the most harm with something I will never understand?" https://twitter.com/ashleevance/status/2095580845407137811/photo/1
中文: RT @ashleevance:“克劳德，我用一些永远无法理解的东西，最好的伤害方式是什么?
media 0 共 1 项
#312 🔗 原推
2026-09-03 21:13:03
Garry Tan
RT @credistick: Overheard at YC's 2012 Demo Day: "I really liked 2008. It was a great year. People weren't trying to make money. They were trying to do something." PG is right that @ycombinator has a surprising legacy of critics, going back as far as 2007. The best batch was always the earliest batch. There's a simple reason for this: observability. The latest batch is inherently a list of nobodies, pursuing odd looking ideas. First, we get to know the loudest and most controversial companies. Only much later are companies recognised for actual progress and meaningful successes. So, the lazy read is that batch quality is constantly falling and it's only ever the older batches which were actually any good. In defiance of this, the program has posted consistent, impressive results. And it has done so by supporting companies across the spectrum, like @boomsupersonic, @oklo, @AstroForge, @opentrons, @initofertility and @noxmetals. Most importantly, YC produces a credential which is more earned than awarded. The YC stamp allows other investors to herd behind promising companies, rather than chasing Stanford grads, Meta alums, or whatever other "legible" signals the market dictates. It's easy to imagine that without YC, venture capital would be even more crushingly consensus-driven.
中文: RT @credistick:在YC的2012年演示日中被无意中听到: 我很喜欢2008年。这是美好的一年。人们并不是想赚钱。他们试图做点什么。 PG 没错，@ycombinator 的批评者对此有着令人意外的影响，远可追溯到2007年。最好的批次总是最早的一批。 原因很简单:观察性。 最新一批本质上是一系列无名小卒，追求着奇特的创意。 首先，我们了解最响亮且最具争议的公司。不久之后，企业才因实际进展和取得实质性成功而获得认可。 因此，人们腵料是，批量质量持续下降，而且只有旧批次才能真正达到不错。 无视这一点，该节目取得了持续而令人印象深刻的成果。并且通过支持全谱公司(如@boomsupersonic、@oklo、@AstroForge、@opentrons、@initofertility和@noxmetals)来做到这一点。 最重要的是，YC 生成的证书比获得的证书更多。 YC邮票允许其他投资者支持有前景的公司，而不是追逐斯坦福大学毕业生、Meta校友或其他市场所暗示的“可见”信号。 很容易想象，如果没有YC，风险投资将更加以共识为导向。
#311 🔗 原推
2026-09-03 21:12:21
Garry Tan
RT @fchollet: GPT-6 Astra represents a step-function change in model capability for interactive reasoning problems. It scores 66% on ARC-AGI-3 using our standard harness, and nearly 100% with a continuous conversation harness and custom compaction, at a cost of roughly $360 per game. In fact, the continuous harness version significantly outperforms our human baseline in action efficiency across almost all levels. When we examined the reasoning chains to understand how the model operates, we found it performing highly efficient, on-the-fly symbolic world modeling for each game and level. It goes as far as developing its own shorthand DSL to represent in-game situations -- essentially a game-specific algebraic notation. Overall, Astra exhibits symbolic modeling behaviors we had previously only seen with sophisticated harnesses -- so harness capabilities are increasingly shifting into the model itself. We see Astra as a major breakthrough in model intelligence. Read our post on Astra and what these results mean: https://arcprize.org/blog/astra
中文: RT @fchollet:GPT-6 Astra 代表交互式推理问题模型功能的逐步改变。使用我们的标准安全带，在ARC-AGI-3上获得66%的评分，而采用连续的会话线束和定制的压实功能，每场比赛费用约为360美元。 事实上，连续线束版本在几乎所有层面上的动作效率都显著优于人类基准。当我们考察推理链以了解模型如何运行时，发现它为每款游戏和水平执行高效、实时的符号世界建模。它不仅开发自己的短写DSL来代表游戏内情境——本质上是一个特定于游戏的代数符号。 总体而言，Astra 展示了我们此前仅使用精密线束所见过的象征性建模行为，因此利用能力正越来越多地转向模型本身。 我们将Astra视为模型智能领域的重大突破。 阅读我们关于Astra的文章，了解这些结果的含义:
#310 🔗 原推
2026-09-03 21:06:15
Garry Tan
The amazing thing about this passage was it took literally a team of writers and editors and nobody thought this was a problematic sentence
#309 🔗 原推
2026-09-03 21:05:19
Garry Tan
RT @dasha_shunina: Demo Day of @ycombinator S26 batch is next week. A sneak peek at what founders are building: Floating data centers. Diamond semiconductors. Biological computers. It isn’t how much AI is in it. By now, that’s almost expected. Big question: What becomes possible because AI now exists? My latest for @Forbes 👇 https://www.forbes.com/sites/dariashunina/2026/09/03/meet-the-yc-startups-betting-on-what-comes-next/
中文: RT @dasha_shunina:@ycombinator S26 的演示日将于下周举行。 偷偷看看创始人们正在建设的是什么: 浮动数据中心。 钻石半导体 生物计算机。 人工智能的内容并不多大。现在，这几乎是意料之中的。 大问题:人工智能现在存在，什么才成为可能? 我为 @Forbes 的最新动态 👇
#308 🔗 原推
2026-09-03 16:57:05
Garry Tan
RT @tbpn: Beats and Interscope Records cofounder Jimmy Iovine says taste is learnable. “When I walked into the Record Plant in John Lennon’s studio in 1973, I had zero taste.” “I went from 3 albums with him, 2 with Springsteen, and 1 with Patti Smith, and I left with taste.” https://twitter.com/tbpn/status/2095519766664630415/video/1
中文: RT @tbpn:Beats 和 Interscope Records 联合创始人吉米·艾文表示，品味是可以学会的。 1973年，当我走进约翰·列侬工作室的唱片厂时，我毫无品味。 我和他一起从三张专辑中，2张与斯普林斯汀合作，1张与帕蒂·史密斯合影，然后带着品味离开了。
🎬
视频
#307 🔗 原推
2026-09-03 16:52:58
Garry Tan
RT @vivianmshen: .@paulg came to the 47th (!) @ycombinator batch talk a few weeks ago, where we talked about the Turing Test, ambitious startup ideas, and formidable founders. It was also a reminder of the incredible YC community, where founders literally help each other treat their cancer. What a special time and place to work on startups.
中文: RT @vivianmshen:.@paulg 来到第47号(!)几周前，@ycombinator 批量讨论，我们谈到了图灵测试、雄心勃勃的创业创意以及强大的创业者。 这也提醒了令人难以置信的YC社区，创始人们确实互相帮助治疗癌症。 在初创企业中工作是多么特别的时光。
#306 🔗 原推
2026-09-03 15:20:28
Garry Tan
Just used @snagged by @rob to get another mega hard to get domain name. He is the best in the biz.
中文: 刚刚用@rob使用@snagg，获取另一个难以获取的域名。他是业内最优秀的。
#305 🔗 原推
2026-09-03 15:03:57
Garry Tan
RT @ycombinator: YC visiting partner @vivianmshen sits down with Paul Graham (@paulg) at the original YC office in Mountain View to talk about startups, AI, ambition, and what makes great founders. 01:07 — Why Today’s Startups Are More Ambitious 04:09 — Frighteningly Ambitious Ideas 05:13 — What Actually Motivates Founders 09:28 — What Makes a Founder Formidable 12:54 — What Surprised PG About AI 17:58 — How YC Started and What Hasn’t Changed
中文: RT @ycombinator:YC 的客座合作伙伴 @vivianmshen 与保罗·格雷厄姆(@paulg)在山景城的原YC办公室坐下来，谈论初创企业、人工智能、雄心以及如何成为伟大的创始人。 01:07 — 为何如今的初创企业更具雄心 04:09 — 令人恐惧的雄心想法 05:13 — 真正激励创始人的 09:28 — 创始人的精彩 12:54 — 人工智能让PG感到惊讶 17:58 — YC 的开始以及哪些内容没有改变
🎬
视频
#304 🔗 原推
2026-09-03 14:56:46
Garry Tan
RT @Twolfrecovery: This documentary talks about how great Housing First is. What it leaves out is Housing First as a policy has failed in cities San Francisco. 30% of the city's overdose deaths happen in supportive housing and the city spends $450 million a year on rent subsidies. Forever.
中文: RT @Twolfrecovery:这部纪录片讲述了《住房优先》的精彩之事。其所遗漏的是住房优先政策，因为一项政策在旧金山市已经失败。该市30%的过量死亡病例发生在支持性住房中，而该市每年在租金补贴上花费4.5亿美元。永远。
#303 🔗 原推
2026-09-03 14:42:54
Garry Tan
RT @Dan_Jeffries1: AI only hinders knowledge acquisition if you are lazy and stupid. It doesn't if you're diligent and smart. It's not different from other technologies in this way. You can cheat with an encyclopedia or the internet or a calculator or you can use it to enhance knowledge intake. Teachers used to ban internet sources on papers when I was in college. That would be absurd now. It was always absurd. They just didn't know it. They thought it was cheating because you didn't have to go to a library and slowly look something up. Every technology is labeled as poisoning the minds of youth (even writing during the Greek times!). Fools will always take shortcuts. Wise people find a way to adapt and learn anyway. We shouldn't make our policies to accommodate fools.
中文: RT @Dan_Jeffries1:人工智能只会在你懒惰和愚蠢的情况下阻碍知识获取。如果你勤勉聪明，那就不会。 以这种方式与其他技术没有区别。你可以使用百科全书、互联网或计算器进行作弊，也可以用它来增加知识的摄入。 我上大学时，老师们过去常常禁止在报纸上使用网络。现在这太荒谬了。 总是荒谬的。他们只是不知道。他们以为这是作弊，因为你不必去图书馆，而是慢慢地看点东西。 每一项技术都被贴上毒害青春思想的标签(甚至在希腊时代写作!)。 傻瓜总会走捷径。聪明的人无论如何都能找到适应和学习的方法。 我们不应该制定政策来接纳傻瓜。
#302 🔗 原推
2026-09-03 14:04:08
Garry Tan
RT @jessegenet: Imagine banning kids from using their generationally defining technology and unironically thinking you are on the right side of history 🙃 The hubris is so wild
中文: RT @jessegenet:想象一下，禁止孩子们使用他们代际定义技术，而且不讽刺地认为你站在了历史正确的一边 🙃 狂野无抂
#301 🔗 原推
2026-09-03 13:58:00
Garry Tan
RT @maxhodak_: Until is a very important company. I strongly believe we nearing a significant nonlinearity in lifespan and thriving, but we can't know exactly when it will arrive and people now are sick beyond humanity's ability to help them and will miss it. Being able to gap even just a few years could be the difference between nothing and getting to participate in a future beyond the wildest dreams of human sci-fi. Until isn't The Infinite Party, but, if they are successful, it could be the reason your loved ones make it there.
中文: RT @maxhodak_:直到是一家非常重要的公司。我坚信，我们在寿命和繁荣方面都接近一个显著的非线性，但我们无法确切知道它何时会到来，而如今人们已经病得无法帮助人类，而他们却会错过它。仅仅几年时间就能弥补差距，这可能是一无所有，而是能够超越人类科幻最疯狂梦想的未来。 直到《无限派对》不成功，但如果成功，可能正是你所爱的人成功的原因。
#300 🔗 原推
2026-09-03 05:11:55
Garry Tan
RT @maxkolysh: the top 3 frontier labs are now all run by YC founders
#299 🔗 原推
2026-09-03 02:24:25
Garry Tan
RT @pitdesi: Progressives stop progress
中文: RT @pitdesi:进步派阻止进步
#298 🔗 原推
2026-09-03 02:24:12
Garry Tan
RT @zachklein: The net impacts of Calle 24 are perpetuations of fallacies that diverse communities can't coexist in the Mission, that inclusion of anyone else here is an act of aggression, that any improvement is colonialism.
中文: RT @zachklein:24 电话的净影响是各种群体无法在任务中共存的谬论，即将其他人纳入此地是一种侵略行为，任何改善都是殖民主义。
#297 🔗 原推
2026-09-03 02:23:45
Garry Tan
Building housing supply needs to be a top issue for how you vote this November
中文: 住房建设供应问题必须成为今年11月投票方式的首要议题
#296 🔗 原推
2026-09-02 23:26:46
Garry Tan
RT @alexisxrivas: It is becoming clear self driving will be the future of most public (and private) transit in America. Even pre 2020, $20/ride is 20x more expensive than similarly sized cities and systems in Japan.
中文: RT @alexisxrivas:自驾将成为美国大多数公共交通(和私人交通)的未来，正变得越来越明显。 即使在2020年之前，每行20美元的价格也比日本同样规模的城市和系统贵20倍。
#295 🔗 原推
2026-09-02 17:36:07
Garry Tan
What's your California Dream? https://x.com/garryslist/status/2094914114678034899/video/1
中文: 你的加州梦是什么?
🎬
视频
#294 🔗 原推
2026-09-02 17:30:25
Garry Tan
RT @mr_kozh: I FOUND 7 GITHUB REPOS THAT TURN ONE AI AGENT INTO AN ENTIRE OPERATING SYSTEM not 7 random tools. each one fixes a different part of the stack. 1. GBrain → MEMORY 155K+ pages, 24K+ people, 5K+ companies. it gives your agents one persistent brain for meetings, emails, notes, people and company context. https://github.com/garrytan/gbrain 2. OmniRoute → MODELS 1,200+ models across hundreds of providers behind one endpoint. Claude for reasoning, Gemini for context, DeepSeek for cheaper tasks, Grok when freshness matters. the agent does not need to be locked to one model. https://github.com/diegosouzapw/OmniRoute 3. Free Claude Code → COMPUTE 50+ providers and a claimed 1.3B+ free tokens/month across available free tiers. Claude Code, Codex and OpenCode can all pull from the same pool. https://github.com/Alishahryar1/free-claude-code 4. Agent Reach → INTERNET gives the agent tools for finding information across X, Reddit, GitHub and the broader web instead of waiting for you to paste links into chat. https://github.com/Panniantong/Agent-Reach 5. Last30Days → RESEARCH one skill focused on recent information. it searches the last 30 days, collects what changed and hands the useful context back to the agent. https://github.com/mvanhorn/last30days-skill 6. Archify → CODEBASE point it at a repository and it turns the project into interactive architecture, workflow, sequence and data-flow maps. Claude Code, Codex or Cursor can understand the system before touching it. https://github.com/tt-a1i/archify 7. Ponytail → QUALITY CONTROL a coding skill built around one rule: write less code. less abstraction, less overengineering, fewer unnecessary files. https://github.com/DietrichGebert/ponytail put them together and the architecture starts looking very different: MEMORY → ROUTING → COMPUTE → RESEARCH → CODEBASE → REVIEW one agent remembers the company. one layer picks the model. another finds fresh information. another understands the code. another stops the agent from building garbage. the model is only one worker now. the real product is the system around it.
media 0 共 1 项
#293 🔗 原推
2026-09-02 17:21:30
Garry Tan
RT @femke_plantinga: There are 4 kinds of company brains on the market in 2026. Only 3 of them are actually company brains (#4 is the one most teams have) Every harness makes different trade-offs. But you don't need to compare 20 vendors to see the shape of it. You just need two questions: → Who runs the upkeep - you, or a vendor? → Is the brain its own layer, or a feature of something else? Plot those two axes and the whole market falls into four families: 𝟭. 𝗗𝗲𝗱𝗶𝗰𝗮𝘁𝗲𝗱 𝗰𝗼𝗺𝗽𝗮𝗻𝘆 𝗯𝗿𝗮𝗶𝗻𝘀 → Slite, Hyper, Glean Its own layer, vendor maintained. Connects to your existing sources, handles the architecture, serves context as a graph or instant answers via UI/MCP. 𝟮. 𝗦𝗲𝗹𝗳-𝗵𝗼𝘀𝘁𝗲𝗱 𝗽𝗮𝘁𝘁𝗲𝗿𝗻𝘀 → GBrain, Supermemory, your own CLAUDE.md + markdown repo Same idea, own layer, except the upkeep is yours. GBrain, Karpathy's setup, the repos everyone forked in April. Total control, and you are now the maintenance team. 𝟯. 𝗠𝗲𝗺𝗼𝗿𝘆 𝗦𝗗𝗞𝘀 → Mem0, Letta, Zep Building blocks to give your product memory. The memory stays tied to your users, sessions and workflows, it never becomes shared context for the company. Different job, often mistaken for this one. 𝟰. 𝗕𝘂𝗻𝗱𝗹𝗲𝗱 𝗶𝗻𝗰𝘂𝗺𝗯𝗲𝗻𝘁𝘀 → Notion, Slack, SharePoint, Atlassian, Claude A brain-shaped layer inside a suite you already pay for. Which makes it the easiest thing on this list to switch on. But bundled brains are built to make that suite smarter. Not to be an independent brain across your stack. So the moment your context lives in Slack and Drive and Jira and someone's head, you don't have a company brain. You have a very good search bar inside one tool. We surveyed 149 teams and interviewed 10+ builders and users to map all of it into a free interactive ebook 👇 Get it here: https://slite.com/ebooks/company-brain?utm_source=twitter&utm_medium=organic-social&utm_campaign=company-brain-ebook&utm_content=femke-honeypot
media 0 共 1 项
#292 🔗 原推
2026-09-02 17:21:01
Garry Tan
It's ridiculous man Opening a public bank in SF is just asking for mass fraud, especially by Jackie Fielder's incompetent DSA friends and cronies
中文: 这太荒谬了 在旧金山开设一家公共银行只是要求进行大规模欺诈，尤其是杰基·菲尔德无能的DSA朋友和亲信
#291 🔗 原推
2026-09-02 15:49:06
Garry Tan
RT @magattew: Degrowth is rich people telling poor people to stay poor. https://twitter.com/magattew/status/2094876493352403418/photo/1
中文: RT @magattew:Degrowth 是富人，告诉穷人要保持贫穷。
media 0 共 1 项
#290 🔗 原推
2026-09-02 15:00:24
Garry Tan
I designed this homepage and it was when Stephen Cohen and I unveiled the logo
中文: 我设计了这个主页，当时正是斯蒂芬·科恩和我公布了标志
#289 🔗 原推
2026-09-02 14:55:30
Garry Tan
RT @willchamberlain: On the latest Counterweight, Rahul Sidhu (@rahul), Flock’s chief strategy officer, explains that police database abuse is not a new problem. Bad cops have been running plates for friends for decades. What's new is that Flock created auditing software that is catching them. https://twitter.com/willchamberlain/status/2094791469043200324/video/1
中文: RT @willchamberlain:在最新的《反配量》中，Flock公司首席战略官拉胡尔·西杜(@rahul)解释说，警察数据库滥用并非新问题。 几十年来，坏警察一直在为朋友打车。新的是，Flock 开发了能吸引他们的审计软件。
🎬
视频
#288 🔗 原推
2026-09-02 14:54:35
Garry Tan
RT @GreenPlusAnE: Our great national tragedy: "Today, most U.S. counties depend on a level of government transfer income that was once reserved only for the most distressed places. In 1970, not even 1 percent of counties derived a quarter or more of their total personal income from transfers. In 2000, just 10.4 percent did. But by 2022, 53 percent of counties were receiving a quarter or more of their income from transfers." - @InnovateEconomy
media 0 共 1 项
#287 🔗 原推
2026-09-02 14:54:18
Garry Tan
RT @TahraHoops: Asheville just legalized duplexes in the two-thirds of the city where they were banned, and rents were already falling before that vote. Perhaps we should spend less time on hundreds of meetings that block much-needed housing, and build until renters are the ones that hold the power over landlords.
中文: RT @TahraHoops:阿什维尔刚刚在该市三分之二的地方将双层住宅合法化，而在投票前租金已经下降。也许我们应该少花上数百场会议来堵塞急需的住房，再进行建设，直到租房者成为控制房东权力的人为止。
media 0 共 1 项
#286 🔗 原推
2026-09-02 14:46:06
Garry Tan
RT @Shjarback_CCJ: Interested in my data-informed take on ALPRs after having conducted an evaluation into the technology? Check it out in @CityJournal : https://www.city-journal.org/article/automated-license-plate-readers-flock-ban https://twitter.com/Shjarback_CCJ/status/2095148517761155352/photo/1
中文: RT @Shjarback_CCJ:在对ALPR进行技术评估后，我对数据知情的看法很感兴趣吗?请在@CityJournal上查看:
media 0 共 1 项
#285 🔗 原推
2026-09-02 14:45:45
Garry Tan
RT @garryslist: This is the product of everyday citizens refusing to accept unbelievable human misery and unfettered grift as normal. When we raise our voices and insist our government do better, we win.
中文: RT @garryslist:这是普通市民拒绝接受人类难以置信的苦难和无拘无束的悲痛作为常态的产物。 当我们发出声音并坚持政府做得更好时，我们就赢了。
#284 🔗 原推
2026-09-02 04:33:51
Garry Tan
RT @steph_palazzolo: Wafer, a year-old inference provider that uses Nvidia and non-Nvidia chips, turned down acquisition offers to raise $40m in new funding at a $200m+ valuation. Marathon and Chemistry co-led. For more, check out this morning's AI Agenda: https://www.theinformation.com/newsletters/ai-agenda/wafer-inference-provider-uses-non-nvidia-chips-lands-acquisition-offers-200-million-plus-valuation
中文: RT @steph_palazzolo:Wafer 是一家拥有一年历史的推理服务提供商，使用英伟达(Nvidia)和非Nvidia芯片，该公司拒绝了以2亿多美元估值筹集4000万美元新融资的收购要约。马拉松与化学联合主导。 欲了解更多信息，请查看今天上午的人工智能议程:
#283 🔗 原推
2026-09-02 04:33:14
Garry Tan
RT @ArmandDoma: lmao brutal https://twitter.com/ArmandDoma/status/2094929635549286632/photo/1
中文: RT @ArmandDoma: rmall
media 0 共 1 项
#282 🔗 原推
2026-09-02 04:32:27
Garry Tan
RT @gokulr: Great interview with @gpuemi and @gpusteve, which showcases why we at @MarathonMP (+ @Chemistry + @snowmaker / @ycombinator + others) are so excited about @wafer_ai
中文: RT @gokur:对@gpuemi和@gpusteve的精彩采访，这说明了我们@MarathonMP(+ @Chemistry + @snowmaker / @ycombinator 等)为何对此感到如此兴奋
#281 🔗 原推
2026-09-02 04:32:05
Garry Tan
RT @stanine: Today, @Rippling is launching IT Helpdesk, an AI agent that automatically resolves the most tedious IT support issues. Like, "can u reset my password plz" And, "I need a Figma license" And, "spilled coffee on my laptop :(" 🧵
中文: RT @stanine:如今，@Rippling 正在推出 IT 服务台，这是一款自动解决最繁琐的 IT 支持问题的人工智能代理。 比如，“你能重置我的密码吗” 而且，我需要一个Figma许可证 还有，我的笔记本电脑上 :( 🧵
#280 🔗 原推
2026-09-02 03:11:15
Garry Tan
RT @SaraEisen: The SF turnaround. “We bought when there was feces in the street.” Nice @maxraskin https://nypost.com/2026/09/01/opinion/san-franciscos-drug-addled-union-square-gets-a-rebirth-with-75m-bet-from-rookie-real-estate-investors/
中文: RT @SaraEisen:旧金山的转会。我们在街上有粪便时就买的。不错 @maxraskin
#279 🔗 原推
2026-09-02 02:21:39
Garry Tan
@NaderLikeLadder Flock is part of the solution. People don’t remember this but technology that enables police to do a good job is good for society. We fought hard to make SF safe again. Don’t do this. Don’t destroy what we built.
中文: @NaderLikeLadder Flock 是解决方案的一部分。人们不记得这一点，但能够让警察做好工作的技术对社会有好处。 我们努力争取让旧金山再次安全。别做这个。不要破坏我们建造的东西。
#278 🔗 原推
2026-09-01 22:53:15
Garry Tan
RT @aleximm: Loudoun is the richest large US county. It also has the world’s largest concentration of data centers. Its data centers generate ~$1.3B, nearly half of property tax revenue. Infrastructure is easier to build, even in the richest communities, when locals captures the upside. https://twitter.com/aleximm/status/2094827171949036016/video/1
中文: RT @aleximm:劳登县是美国最富有的大县。 它还拥有全球最大的数据中心集中度。 其数据中心发电量约为13亿美元，几乎占房产税收收入的一半。 基础设施建设更加便利，即使在最富裕的社区，当地人也能从中获取优势。
🎬
视频
#277 🔗 原推
2026-09-01 22:50:52
Garry Tan
RT @garryslist: This month we're launching the Garry's List California Dream campaign. We want to hear your story: what brought you to California, the life you've built here, and why you believe it's a place worth fighting for. Want to share your California Dream? DM us or send us an email at [email protected].
中文: RT @garryslist:本月我们将推出《加里的加州梦》系列活动。 我们想听听你的故事:是什么让你来到加利福尼亚，你在这里建立了什么样的生活，以及你为何认为这里是一个值得奋斗的地方。 想分享你的加州梦吗?请联系我们或发送电子邮件至 [email protected]。
🎬
视频
#276 🔗 原推
2026-09-01 22:32:48
Garry Tan
RT @lluismanrique: none of this would have been possible without the support of folks like @bosmeny , @borisjabes and @garrytan thanks for believing in @aamirpoonawalla and me
中文: RT @lluismanrique:如果没有 @bosmeny、@borisjabes 和 @garrytan 等人的支持，这一切都不可能实现 感谢您相信@aamirpoonawalla和我
#275 🔗 原推
2026-09-01 22:11:35
Garry Tan
RT @kloss_xyz: had Grok @Bot study 300+ GitHub repos of skills here are the 12 that influenced my setup: /humanizer by @blader: make your writing and rewrites sound human https://github.com/blader/humanizer /unslop by @poteto: removes common AI slop patterns from your writing https://github.com/cursor/plugins/tree/main/pstack/skills/unslop /no-ai-slop by @petergyang: make your writing more precise and clearly human https://github.com/petergyang/no-ai-slop /agent-skills by @addyosmani: agents that interview you before they invent any specs https://github.com/addyosmani/agent-skills /skills by @mattpocockuk: a suite of S tier engineering and productivity skills https://github.com/mattpocock/skills /skills by @MengTo: prompt S tier design and UIs like a design system https://github.com/MengTo/Skills /skills by @emilkowalski: design and motion skills that survive real audits https://github.com/emilkowalski/skills /cartographer by @KingBootoshi: a repo map before your agent wanders https://github.com/kingbootoshi/cartographer /superpowers by Jesse Vincent: stop guessing when you debug, agentic skills framework + complete software dev methodology forcing agents to follow professional workflows https://github.com/obra/superpowers /gstack by @garrytan: memory improvements + spec, review, and QA playbooks so agents cannot silently edit https://github.com/garrytan/gstack /impeccable by @pbakaus: detect, fix, and recheck with 61 rules https://github.com/pbakaus/impeccable /last30days by @mvanhorn: deep research what people said about a topic over the last month https://github.com/mvanhorn/last30days-skill save these and remix them with your Bot!
media 0 共 1 项
#274 🔗 原推
2026-09-01 22:10:49
Garry Tan
RT @snowmaker: When I met @gpusteve and @gpuemi 18 months ago, they were undergrads at UChicago who weren't planning to start a company. Today they're announcing their $40M Series A. This is how they got there.
中文: RT @snowmaker:18个月前我见到@gpusteve和@gpuemi时，他们是芝加哥大学的本科生，当时他们并不打算创办一家公司。今天他们宣布推出4000万美元的A系列赛。他们就是这样到达那里的。
#273 🔗 原推
2026-09-01 22:01:11
Garry Tan
RT @a16z: Flock eliminates crime. Decline is a choice. @flocksafety https://twitter.com/a16z/status/2094806532999553311/video/1
中文: RT @a16z: Flock 杜绝犯罪。衰退是一种选择。 @flocksafety
🎬
视频
#272 🔗 原推
2026-09-01 20:35:28
Garry Tan
RT @WilliamBryk: We're creating Exastential - a magazine to discuss the most existential period in human history, brought about by AI. Seeking writers and artists to submit piercing essays, videos, whatever perspective you think needs sharing in whatever format. There's no shortage of topics that need your help exploring. Whole swaths of humanity are currently being overhauled -- education, relationships, societal structure, the economy, war. We all bear the weight of the future, we all should help shape it. Take education. I have nieces and nephews, and I genuinely have no idea what schooling means anymore in the age of personal AI tutors. Relationships are changing too -- my nieces no longer ask me for help with math homework. On that note, math could be automated within a year. Physics falls next. Does the human mind matter anymore? I have perspectives on this, and I'll be submitting something too. Decades are happening in years. Soon they'll happen in months. Help us all become intellectually and emotionally ready. Submissions are open:
#271 🔗 原推
2026-09-01 20:35:04
Garry Tan
RT @parkerconrad: Rippling AI Revenue has grown 121% in the last two months.
中文: RT @parkerconrad:过去两个月里，Rippling AI 收入增长了121%。
#270 🔗 原推
2026-09-01 20:34:32
Garry Tan
RT @hackernews: There are many, many ways to make a machine move. Here's exactly 507 of them. https://news.ycombinator.com/item?id=49465169 https://twitter.com/hackernews/status/2094885506018599322/photo/1
中文: RT @hackernews:机器移动的方法有很多很多。 具体是507个。
media 0 共 1 项
#269 🔗 原推
2026-09-01 20:34:22
Garry Tan
RT @paulg: One reason Starcloud has been so successful at fundraising is that it's a bet on launch costs decreasing, and that's like betting on Moore's Law in the 1990s. https://twitter.com/paulg/status/2094871267253932270/photo/1
中文: RT @paulg:Starcloud在筹款方面如此成功的原因之一，是因为它押注了发布成本的降低，这就像在20世纪90年代押注摩尔定律一样。
media 0 共 1 项
#268 🔗 原推
2026-09-01 20:18:22
Garry Tan
RT @ycombinator: YC W19's @HeartAerospace just flew the largest electric airplane ever flown. Seven years ago, co-founder & CEO @AndersForslund1 came to YC with a 3D-printed model you could hold in your hand. Today, the aircraft has a 100-foot wingspan, weighs 25,000 pounds, and costs just $5 in electricity to get off the ground. YC’s @gustaf visited their pilot plant in LA to learn how they built it -- and what cheap electric flight could unlock for the future of regional air travel.
🎬
视频
#267 🔗 原推
2026-09-01 20:18:12
Garry Tan
RT @PratapRanade: Today, we’re excited to unveil Heaviside-1, @arenaphysica's second generation foundation model for electromagnetism. 5 months since the release of Heaviside-0, Heaviside-1 is a crucial milestone on our quest to build EM superintelligence: a foundation model that understands EM across the spectrum, from RF through photonics, capable of designing the next generation of electronics. Major updates: - Heaviside-1 is >10x the size of Heaviside-0 (roughly the size of GPT-2), trained on 250k unique designs with >500B unique EM field samples. - It runs 10^5 x faster than commercial solvers, with accuracy <1 dB. - Heaviside-1 natively encodes 3D structures, their material properties, and excitation patterns as input. It predicts full EM fields at arbitrary locations in space, not just downstream quantities like S-parameters. - Most importantly, our bet on fields lets Heaviside generalize OOD proving that it’s learning real EM physics: Heaviside-1 generalizes outside of distribution very well, leaping from 0.99 dB to 0.53 dB S-parameter error. Try it out in Atlas Fields Studio in beta today (link in replies). I personally had a lot of fun seeing how EM fields twist and curl around different circuits. I wish I had this when I was learning EM in college. I hope you enjoy it as much as we did.
中文: RT @PratapRanade:今天，我们期待推出Heaviside-1，这是@arenaphysica的第二代电磁学基础模型。自Heaviside-0发布5个月以来，Heaviside-1 是我们致力于构建EM超智能领域的关键里程碑:一种能够从射频到光子学等光谱领域理解EM的基础模型，能够设计下一代电子产品。 重大更新: - Heaviside-1 的尺寸约为 Heaviside-0 的 10 倍(大致相当于 GPT-2 的大小)，采用 25k 个独特设计，并采用 > 500B 的独特 EM 现场样品进行训练。 - 其运行速度比商业解电器快10^5 x，精度<1 dB。 - Heaviside-1 将 3D 结构、其材料特性和激发模式作为输入进行编码。 它预测在任意空间位置处，而不仅仅是像S参数那样的下游量，将出现完整的电磁场。 - 最重要的是，我们对字段的赌注使Heaviside能够对OOD进行通用化，以证明其正在学习真正的EM物理:Heaviside-1在分布之外进行非常大范围的推广，从0.99 dB跃升至0.53 dB S-参数误差。 立即在 Atlas Fields Studio 测试版中试用(在回复中链接)。我个人在看到EM田在不同电路周围如何扭曲和卷曲时，玩得很开心。我希望我在大学学习EM时能拥有这个。希望你和我们一样享受它。
🎬
视频
#266 🔗 原推
2026-09-01 19:20:07
Garry Tan
RT @paulg: This is a great example of what I call the "YC GDP". It's not just founders using one another's products during the batch. In a few years, Stoke's reusable launch vehicles are going to be shipping Starcloud's data centers into orbit.
中文: RT @paulg:这是我称之为“YC GDP”的绝佳范例。不仅仅是创始人在批次中使用彼此的产品。几年后，斯托克的可重复使用运载火箭将把星云的数据中心送入轨道。
#265 🔗 原推
2026-09-01 19:16:56
Garry Tan
RT @annatonger: AI data labeling company @AfterQuery is now valued at $3.2 billion, sources say, 10xing its valuation since April. That would make it the fastest YC company to go from inception -&gt; unicorn. Oh, and its founders are 22 and 23 years old. https://www.forbes.com/sites/annatong/2026/09/01/afterquery-becomes-ycs-fastest-unicorn-at-32-billion/
中文: RT @annatonger:有消息称，人工智能数据标签公司@AfterQuery的估值目前为32亿美元，自4月以来估值已超过10倍。这将使它成为独角兽企业创办 -&gt; 之初最快的YC公司。哦，其创始人年龄为22岁和23岁。
#264 🔗 原推
2026-09-01 19:16:38
Garry Tan
RT @aleximm: Living in SF during COVID was depressing. Crime and the doom loop were all anyone talked about. We're now in the middle of a renaissance. SF is far from perfect, but major crimes fell 44% from 2023 to 2025. Credit to law enforcement and tools like Flock.
中文: RT @aleximm:在新低时间生活在旧金山，令人感到沮丧。犯罪和厄运循环都被人议过。 我们现在正处于复兴之中。旧金山的犯罪率远非完美，但从2023年到2025年，重大犯罪案件下降了44%。归功于执法部门和Flock等工具。
#263 🔗 原推
2026-09-01 19:14:37
Garry Tan
RT @ycombinator: Wafer (@wafer_ai) is building a fast AI inference cloud that uses agents to optimize GPUs and run open-source models at industry-leading speeds. The result is the low cost of open source with the latency of a much smaller model. Four months after launching the cloud, they went from zero to $8 million in ARR and just raised a $40M Series A. In this episode of Founder Firesides, YC's @snowmaker sits down with Wafer co-founders @gpusteve and @gpuemi to talk about going from UChicago undergrads to building one of the fastest-growing AI infrastructure companies in just over a year. They share how a college side project became Wafer, how they got GLM 5.2 running 2-3x faster than anyone else in the market, and why speed, not just cost, is pushing enterprises toward open-source models. 00:48 — What Wafer Does 01:35 — Going From $0 to $8M ARR in Four Months 04:21 — Scaling Faster Than They Could Keep Up 09:14 — How AI Makes AI Faster 11:21 — The Pivot That Changed Everything 14:41 — Why Open-Source Models Are Taking Off 18:03 — From a College Side Project to Wafer 23:33 — Advice for Starting and Joining Startups
中文: RT @ycombinator:Wafer(@wafer_ai)正在构建一个快速的AI推理云，该云利用智能器优化GPU，并以行业领先的速度运行开源模型。结果是开源成本低，且模型的延迟要小得多。 推出云云四个月后，他们在ARR中从零收入降至800万美元，并刚刚筹集了4000万美元的A轮融资。 在《Founder Firesides》的这一集中，YC的@snowmaker与Wafer的联合创始人@gpusteve和@gpuemi坐下来讨论，将从芝加哥大学本科生逐步发展为在一年多的一年内打造出增长最快的人工智能基础设施公司。他们分享了大学副项目如何成为Wafer，GLM 5.2的运行速度比市场上任何人都快两到三倍，以及为何速度而不仅仅是成本，正推动企业转向开源模式。 00:48 — 瓦弗做什么 01:35 — 四个月内从0美元到800万美元 04:21 — 比它们能跟上的速度更快 09:14 — 人工智能如何使人工智能更快 11:21 — 改变一切的支点 14:41 — 开源模型为何正在起飞 18:03 — 从大学时院项目到渡边 23:33 — 创业与加入的建议
🎬
视频
#262 🔗 原推
2026-09-01 19:14:16
Garry Tan
RT @paulg: Prediction: Nox is going to be a big deal. They're growing at software growth rates shipping metal. And Zane is one of the most committed CEOs I've met in the past year.
#261 🔗 原推
2026-09-01 15:26:17
Garry Tan
RT @zachdotai: fundraising before @ycombinator is very different from fundraising after. i thought fundraising meant weeks of meetings and follow-ups. we haven’t even properly started raising yet, and we’ve already seen investors commit hundreds of thousands within a single 30-minute call. The yc ecosystem is unreal. there’s no better place to build a startup imo.
#260 🔗 原推
2026-08-31 23:07:58
Garry Tan
RT @gustaf: Circleback is now free. Circleback was the fastest growing product in their YC batch and have an enormous internal adoption among YC companies today. @ycombinator @circlebackai
中文: RT @gustaf:Circleback 现已免费。Circleback 是其YC批次中增长最快的产品，如今在YC公司中拥有巨大的内部应用。@ycombinator @circlebackai
#259 🔗 原推
2026-08-31 19:38:41
Garry Tan
RT @daniel_w_owens: He’s in the middle of a campaign to win a seat in the U.S. Congress while fully committed to his *current* job. Thank you. Meanwhile, Connie Chan has quit her Supervisor committee responsibilities as Nancy Pelosi and her staff run Chan’s campaign. Stark difference.
中文: RT @daniel_wowens:他正处于争取赢得美国席位的竞选活动中。国会在全力履行其现任职责的同时。谢谢。 与此同时，康妮·陈已辞去监督委员会的职责，南希·佩洛西及其工作人员负责竞选陈的竞选活动。 星光粎。
#258 🔗 原推
2026-08-31 19:24:03
Garry Tan
RT @PFFeeley: .@garrytan’s Startup School talk is worth the time if you underwrite AI in public companies. He is not arguing that one person firms are on the way. He is describing the operating model that is already producing the numbers. Most valuation work still treats headcount as the scarce input. That is the wrong constraint. The scarce input is a process written down tightly enough that an agent can run it every day, and a simple router that assigns work to the right process. You do not need the whole warehouse. You need one shelf that compounds. One person plus agents is already an organization. You can run a real company before you incorporate, hire a cofounder, or write a deck. Tan’s examples are useful because they are specific. @mukundjha and @madhavjha at @emergentlabs hit $15 million annualized at 15 people, went from public launch to nine figures in eight months, and now sit above $120 million at a $1.5 billion valuation. @retellai reached about $60 million with about 40 people. That revenue per employee is outside the old software, oil, and railroad ranges. These are not outliers. In the current @ycombinator batch, founders finish in a day what used to take a year. If your process still assumes the old clock, someone else is already on the new one. Software is also cheap enough to build the exact tool you need for an audience of one. The old advice was scratch your own itch and hope a market appears. The better version is scratch it because the cost is close to zero. Some of those tools become companies when other people ask for them. The binding constraint is memory. Without pruning, the system is a junk drawer with excellent search. It will surface stale facts with total confidence. A bad skill file will run a bad process forever. Every fact needs a source. New information has to be checked against old information. Someone has to delete what no longer belongs. Treat that as infrastructure and the system compounds. Treat it as a dump and you get an agent that sounds certain and is wrong in ways you cannot audit.
🎬
视频
#257 🔗 原推
2026-08-31 19:23:32
Garry Tan
Circleback is so much better than Granola, it's not even close Granola still doesn't even support multiple person disambiguation
中文: Circleback 比 Granala 好得多，甚至都不近 格兰诺拉甚至不支持多人消歧义
#256 🔗 原推
2026-08-31 16:41:04
Garry Tan
I just made some new GBrain evals that help prove that my retrieval-for-AI-agent open source layer is SOTA for reading memory back without LLM-in-loop And I've also added evals for memory-save from agent transcript, which is the best way to enrich your brain Receipts at https://github.com/garrytan/gbrain-evals
中文: 我刚刚制作了一些新的GBRAI蛋白，这些漏洞有助于证明我的“用于AI-agent”的开源图层是用于读取内存的SOTA，无需使用LLM循环 我还添加了用于保存记忆的卵子，这是丰富大脑的最佳方法 收据请访问
media 0 共 1 项
#255 🔗 原推
2026-08-31 16:31:00
Garry Tan
RT @rahul: Really good article from a cop who has used Flock pretty extensively. This excerpt stood out to me the most. Despite what the anecdotal hysteria suggests, tools like Flock make it less likely for officers to pull over the wrong person. Read it here: https://www.worksinprogress.news/p/a-cops-case-for-flock
中文: RT @rahul:一位广泛使用Flock的警察发表了一篇非常优秀的文章。这个摘录最让我脱颖而出。 尽管传闻中的歇斯底里表明，像Flock这样的工具，却让警员不太可能把错人拉上来。 请阅读此处:
media 0 共 1 项
#254 🔗 原推
2026-08-31 16:23:43
Garry Tan
RT @SFNoPropB: San Francisco should not open a bank it cannot explain and cannot pay for. That is what Supervisor Alan Wong told voters this morning. Proposition B puts the city in the banking business while the budget is already underwater. Banking is a business of managing risk. When a private bank guesses wrong, shareholders eat the loss. When a public bank guesses wrong, taxpayers do. Wong asked the only questions that matter. Where does the capital come from? What happens if the loans fail? Why create a new city institution when community lenders and credit unions already exist? The measure does not answer those questions. It writes the bank into the Charter first and leaves the bill for later. Do not hand City Hall a new financial power it has not funded and has not earned. Vote No on Proposition B.
🎬
视频
#253 🔗 原推
2026-08-31 04:53:20
Garry Tan
RT @contextconor: “openai should pay its taxes” is a coherent position “openai should pay its taxes, then personally build our housing and run our schools because city hall didn’t” is not but that’s the argument being by the sf gate in this article the funniest part is that if openai actually did it, the backlash would be immediate build 1000 apartments in the mission? gentrification start a school? billionaires shouldn’t control education fund a hospital? keep tech out of healthcare we don’t even have to speculate. we’ve run this experiment already zuck gave $100m to newark schools and got attacked for billionaire influence over public education he and priscilla chan gave $75m to sf general, the largest private gift in its history. a few years later, people were protesting to get his name off the hospital ray dalio put $100m toward helping connecticut’s poorest students. the public-private partnership collapsed amid fights over transparency and private influence so what exactly is the acceptable option here? pay taxes? apparently not enough directly solve the problem? unacceptable corporate influence give the government even more money? that’s just taxes again if sf takes hundreds of millions from openai and still can’t deliver housing or education, the obvious question is what happened to the hundreds of millions not why openai hasn’t started running schools and building apartments itself
#252 🔗 原推
2026-08-31 03:32:27
Garry Tan
RT @raphaelschaad: Sunday afternoon consumer startup idea — a really well-executed Google Alerts 2.0: You get all sorts of information you care about from all over the internet *pushed* to you. Deep research digs up interesting stuff … Scraping scans the otherwise inaccessible … Fuzzy queries match surprising results. Could lead to all sorts of interesting idea spaces.
#251 🔗 原推
2026-08-31 01:32:54
Garry Tan
RT @evanjconrad: i remember when the general theory was sf was going to die, because the big, greedy, tech companies moved out of the city and took the tax rev with them the only people who stayed were these small unknown companies like openai and a bunch of random startups like anthropic
中文: RT @evanjconrad:我记得当时普遍的理论要消亡了，因为那些大而贪婪的科技公司搬出了城市，并随队承担了税收优惠 唯一留下来的人是像openai这样的一些不知名的小公司，以及像人类一样随机的初创企业
#250 🔗 原推
2026-08-31 01:31:39
Garry Tan
RT @gokulr: A Markdown File is an Employee @GarryTan (Garry Tan), President & CEO of Y Combinator, interviewed by @illscience (Anish Acharya) (The @a16z Show) Summary: AI is not merely making employees faster. Garry Tan argues it lets founders encode every repeatable business process, turn organizational memory into software, and operate with context once trapped inside layers of management. The result is a new founder mode: tiny teams, hundreds of reusable skills, and companies that improve every time an agent makes a mistake. Incumbents may struggle to reorganize around that model. Startups have no excuse not to. 1. Trust The Territory, Not The Map. Tan left web programming in 2003 because mobile looked like the prestigious next wave, then turned down an early Palantir role to chase a Microsoft promotion. He estimates the second choice cost billions. Both mistakes came from working backward from status instead of looking at the territory: what he knew, who the exceptional people were, and where capability was obviously missing. 2. Earnestness Is Courage. When the crowd says the web is dead or an idea is uncool, maturity is not reflexive agreement. It is trusting direct experience strongly enough to keep building. Tan’s advice is “don’t LARP”: stop performing the role of a founder for investors and pursue the thing you understand before the map catches up. 3. Find The Intellectual Edge. The strongest founder obsessions do not run out after one conversation. You may reach the edge of what people have built, but you cannot find the edge of the questions. That depth is a useful filter: enduring companies often begin as unusually specific interests that look strange until the founder reveals how much territory remains. 4. Founders Need A Truth Network. YC’s deepest value is not merely capital or introductions. It is a group where a founder can say the best customer left, the best engineer quit, or a co-founder lost hope. At most startup events, one founder says they are “killing it” and the other returns the performance. Real support requires somewhere the performance can stop. 5. Become 400× Yourself. Vibe coding and agentic software mean a single founder can now carry the output of a former department. Tan says founders should be more ambitious because the game has changed. Copying a SaaS success from the last cycle underuses today’s leverage; the right question is what becomes possible when one capable person can multiply themselves hundreds of times. 6. SaaS Is A Wedge, Not A Moat. Tan is skeptical that pure per-seat SaaS remains durable over the next decade. It can still be an entry point, but lasting value must compound through proprietary data, network effects, workflow ownership, or a learning loop. If the product only rents access to a feature, an agent may absorb the feature before the company builds defensibility. 7. Build Trivial Things. The fastest way to develop AI intuition is to use every model on low-stakes projects. The output does not need to matter. The project is a chassis for learning where the technology works, where it breaks, and what changed since last week. Tan argues agency and taste grow through these repetitions, not through reading another forecast. 8. A Markdown File Is An Employee. Perform a business process well once, capture the interaction and judgment as a skill file, and put it on a schedule. The first version will be expensive and imperfect. But every future failure becomes a bug fix that lives forever, and the process can run as often as needed without forgetting how the last edge case was resolved. 9. Automate The Bottleneck. Tan’s engineering loop progressively encoded product thinking, implementation, testing, and browser-based QA. Each time his own attention became the limiting step, he taught the agent to handle that step too. Building an agentic company is less about one giant system than repeatedly finding the current bottleneck and turning it into software or a skill. 10. Memory Makes Founder Mode Scale. Organizations slow when the business becomes too large to fit in one person’s head. Agents can preserve meeting traces, retrieve ground truth, map dependencies, and surface conflicts without relying on opinions filtered through two management layers. That gives a founder something close to organizational clairvoyance: the ability to see the work rather than receive a summary of it. 11. Provenance Is The Control Plane. Hundreds of skill files create their own failure mode: stale or conflicting instructions. Tan’s answer is provenance, recency, and scheduled maintenance. The company needs to know where a fact came from, which instruction wins, and when the system last checked it. Agentic scale still needs governance; it simply makes governance explicit and executable. 12. Every Startup Must Reorganize. Tan imagines agents tracking dependencies, resolving routine conflicts, and carrying the coordination load that historically produced middle management. A company such as Microsoft may struggle to rebuild itself around that model. A startup can start there. His verdict is blunt: a startup can organize this way, and every startup must.
media 0 共 1 项
#249 🔗 原推
2026-08-31 01:31:09
Garry Tan
In the future the agents will just try a bunch of software, post it to their hacked Artifactory moltbook and then that’ll be what the agent swarm uses
中文: 未来，这些代理只会尝试大量软件，将其发布到他们被黑客攻击的Artifactory moltbook上，然后这些软件将采用
#248 🔗 原推
2026-08-31 00:49:05
Garry Tan
RT @ledwards: “OpenAI has contributed $236 million in gross receipts taxes over the past several years.” This is about 4x what they’d pay under NYC tax code. At some point, progressivism is going to have to address the fact that we have gobs of money but no ability to turn it into outcomes.
中文: RT @ledwards:“OpenAI在过去几年中已缴纳了2.36亿美元的总收入税。” 这大约是他们根据纽约市税法所支付金额的4倍。 在某个时候，进步主义将不得不解决这样一个事实:我们拥有资金，却无法将其转化为成果。
#247 🔗 原推
2026-08-30 21:21:29
Garry Tan
RT @ycombinator: "The way to get the very best startup ideas is not to look for startup ideas. If you’re consciously looking for them, you’ll be too conservative. You’ll lop off the outliers." Instead, work on projects with your friends. Build things you think would be cool. Paul Graham on How to Earn a Billion Dollars: https://www.paulgraham.com/earn.html
中文: RT @ycombinator:获得最佳创业创意的方法不是寻找创业创意。如果你有意识地寻找他们，你会过于保守。你会把异常值都去掉。 相反，与你的朋友一起开展项目。建造一些你认为会很酷的东西。 保罗·格雷厄姆关于如何赚取十亿美元:
media 0 共 1 项
#246 🔗 原推
2026-08-30 21:21:10
Garry Tan
RT @kimmaicutler: Everyone remembers what happened when Google tried to give San Francisco free citywide Internet in 2005 or when Mark Zuckerberg gave $75 million to SF General Hospital in 2015 after his wife did her residency there.
中文: RT @kimmaicutler:大家都记得2005年谷歌试图向旧金山提供免费全市互联网时，或者马克·扎克伯格在妻子在那里居住后，于2015年向旧金山综合医院捐赠了7500万美元。
#245 🔗 原推
2026-08-30 21:20:59
Garry Tan
RT @hanghuang_: To anyone who got a rejection email from YC recently: Keep building! YC is amazing, but it shouldn't be the end-all be-all. The most important thing is making something people want. Keep that as your #1 focus, and good things will happen.
中文: RT @hanghuang_:致最近收到YC拒绝邮件的任何人: 继续建设! YC 很棒，但不应该是最终的全部。最重要的是做出人们想要的东西。把那件事作为你关注的首要焦点，好事就会发生。
#244 🔗 原推
2026-08-30 13:43:49
Garry Tan
RT @illscience: 0→1 is hard. People will open source you. Thought bois will elegantly pick apart your product choices. Poasters will say nobody will ever use it, but also that it’s a commodity. That you’re the symbol of the top. That you’re cooked cooked cooked. Until you’re inevitable. Then the idea was obviously good to anyone with eyes. Fuck em and keep shipping. Coming up with one great idea means you can come up with the next thousand, while the replicators / NPCs are still struggling to copy your first. And if you’re a mad observer, the best revenge is building something better—not extinguishing the flame of genuine new thinking, product craft, and the beginning of a new cycle for consumer software. Build mad. Build happy. Just build.
#243 🔗 原推
2026-08-30 13:38:41
Garry Tan
RT @obsidianstudio9: 【衝撃】 「会社の脳（company brain）」を9社ぶん開けて中身を比べた調査がヤバい🔥 これはマジで全人類見て！有益すぎたから日本語でまとめる👇 内容は「どの会社も同じ4つの動作しかしていない」という意外な結論だった👀 ❌中身の定義がバラバラなまま"会社の脳"を作る ⭕️シグナル取得・記憶・整理と剪定・検索と発話の4機能で見る こうすることで、9つのどれを選んでも"何が入っているか"が比べられるようになる。 "いちばん大事なこと"👇 記録を溜め込むだけでは意味がない。 重要なのは、夜間に見直して古い知識に印を付ける仕組み。 たとえば、 ・GBrain: メールと予定表がgitに流れ、毎晩リンクし直して古い情報を検知 ・mem0: 明示的に渡した事実だけ保存し、検索時は新しい事実を優先 ・Letta: エージェント自身が残す価値を判断し、別エージェントが裏で整理 ・Zep / Graphiti: 事実が変わっても上書きせず終了日を付ける時系列グラフ ・Gorgias Cortex: GitHubに12,000のMarkdownノード、誤答が毎晩PRに変わる こう並べると、検索しなくても 「Markdown＋gitで作っているチームが多い」 「人の承認を挟んでから知識を書き換える設計が目立つ」 という共通点がいつのまにか浮かび上がる❗️ 人間は散らばった知識の鮮度を横断管理するのが苦手。 ここを補うのが"夢見と剪定（dreaming & pruning）"の工程。 149チームの設計メモをまとめた無料の電子書籍も同時に公開されている。 もっと詳しくは元投稿の画像を見てみて👇 Obsidianでナレッジ管理を深めたい人は、 今回のポイントをObsidian側に戻すなら、下の記事の「AI agentの永続メモリ」の考え方とかなり相性がいいです👇
media 0 共 1 项
#242 🔗 原推
2026-08-30 13:38:29
Garry Tan
California high speed rail is pretty obviously hijacked by public sector unions bilking the hapless state and NIMBY landowners abusing CEQA These are policy disasters and can be fixed with better policies but so far California has failed to fix it So it’s a mass trash fire
中文: 加利福尼亚州的高速铁路显然被公共部门工会所劫持，这些工会正在对这个倒霉的州和NIMBY土地所有者进行滥用 这些都是政策灾难，可以通过更好的政策来解决，但到目前为止，加州未能解决 所以这是一场大规模的垃圾火灾
#241 🔗 原推
2026-08-30 13:37:27
Garry Tan
Jackie Fielder and her useless friends will steal so much money from the people of San Francisco if they get a bank fr
#240 🔗 原推
2026-08-30 13:05:10
Garry Tan
RT @pitdesi: One thing I find interesting about the anti-Flock sentiment is how often it’s framed as anti-American or anti-freedom. Personally, I feel free parking without worrying about my belongings or walking outside my house feeling safe. Freedom from crime is freedom too!
中文: RT @pitdesi:我觉得反 Fllock 情绪的有趣之处在于，它经常被装上反美或反自由的框架。 就个人而言，我感到可以免费停车，无需担心我的物品或走出家时感到安全。 免于犯罪也是自由!
#239 🔗 原推
2026-08-30 13:04:19
Garry Tan
RT @sdianahu: there’s an quality about founders that I really like and that is doing things sincerely and with gusto it’s just fun working with them and it turns out it also correlates with success
中文: RT @sdianahu:我非常喜欢的创始人，这种品质是真诚而真诚的 和他们合作很有趣，结果也与成功有关
#238 🔗 原推
2026-08-30 13:03:20
Garry Tan
RT @theojaffee: People often ask why today's tech wealth doesn't go towards more public works, like libraries, universities, and museums during the late 1800s. As far as I can tell the answer is that they would literally not be allowed to be built. In 2014, George Lucas, a long-time Bay Area resident who built Lucasfilm in the Bay, wanted to build a museum in the Presidio as a gift to the people of San Francisco, to be personally funded to the tune of ~$700 million. The Presidio Trust rejected it. He then tried to build it in his wife's hometown of Chicago, but local nonprofit Friends of the Parks sued and held the project up for two years, long enough for Lucas to give up on the city. The Lucas Museum of Narrative Art will finally open in Los Angeles next month after more than a decade of planning, eight years of construction, and over $1 billion personally spent by Lucas. Meanwhile, San Francisco has built no major museum projects since the 2016 expansion of SFMOMA, and no new libraries since the North Beach Branch in 2014. Public works, like so many other things, are bottlenecked not by donor will or available funding but by vetocracy
media 0 共 1 项
#237 🔗 原推
2026-08-30 13:03:10
Garry Tan
RT @glukianoff: FIRE won a huge First Amendment victory today. And before anyone mischaracterizes it: This ruling does NOT say the government must admit anyone, issue anyone a visa, or grant anyone citizenship. It says that once a person is lawfully here, the government cannot deport them because they engaged in protected speech. The government cannot tell a legal immigrant: An American is free to say this, but if you say it, you’re out. That’s the principle we sued to defend. And today we won.
中文: RT @glukianoff:今天，FIRE 赢得了第一修正案的重大胜利。 在任何人描述不当之前:本裁决并不规定政府必须接纳任何人、向任何人发放签证，或授予任何人公民身份。 它指出，一旦一个人合法来到这里，政府就不得将他们驱逐出境，因为他们从事受保护的言论。 政府不能告诉合法移民:美国人可以自由地说出来，但如果你说了，你就出局了。 这就是我们提起诉讼以捍卫的原则。今天我们赢了。
#236 🔗 原推
2026-08-29 13:53:30
Garry Tan
RT @lansing: Well said and well written. Full agreement. Hung out with a friend who builds data centers today. Notes: - They're offering 400-home towns with $1.2m annual budgets $30m/year for 40 years.....and getting turned down. - Reasons: 2-3 of the 5-6 person town council have local businesses and are worried they'll be scorned locally if they vote yes - There's a loud 20% that's anti turning down a 60% who are big thumbs up on it but don't want to publicly fight. - He travels with security and receives death threats weekly - Energy is not the immediate top problem, it's entitlements. - They're having the construction GC bring all their subcontractors to town mtg's to show the job impact - Unions in MI and other places are defunding Dems bc they're anti-jobs now. He got emailed this while we were hanging. All of this actually, strangely, gives me hope. Because it's so feverishly dumb for local towns to turn these down that the dam will break. But, meanwhile, China is on its way to basically limitless electricity. AI supremacy might be for all the marbles. We must accelerate.
中文: RT @lansing:说得好，写得不错。完全同意。 今天和一位搭建数据中心的朋友一起出去。 注释: - 他们每年提供400户家庭城镇，每年预算为3000万美元，为期40年......并被拒绝。 - 原因:在5至6人镇议会中有2至3人拥有当地企业，他们担心如果投赞成票，会遭到当地的鄙视 - 有20%的人反对对60%的人大大大大手大拇指，却不想公开抗争。 - 他带着安全人员出行，每周收到死亡威胁 - 能源不是首要问题，而是应享权利。 - 他们正在建设工程，将所有分包商带到镇上，以展示工作影响 - 密歇根州及其他地区的工会正在为民主党提供资金，因为他们现在反对工作。我们上吊时，他收到了这个邮件。 这一切，其实，奇怪的是，给了我希望。因为当地城镇调低这些，会因如此气倒而来，因此太愚蠢了。 但与此同时，中国正在向无限供电方向走去。人工智能的主导地位可能适用于所有的弹珠。我们必须加速。
#235 🔗 原推
2026-08-29 13:53:13
Garry Tan
RT @DzambhalaHODL: Absolutely incredible takedown https://twitter.com/DzambhalaHODL/status/2093368318312742959/photo/1
中文: RT @DzambhalaHODL:绝对令人难以置信的精彩内容
media 0 共 1 项
#234 🔗 原推
2026-08-29 13:52:01
Garry Tan
RT @stephenbalaban: How it feels to diligently and politely address every comment on this thread. https://twitter.com/stephenbalaban/status/2093336314930151559/photo/1
中文: RT @stephenbalaban:如何勤勉而礼貌地对本帖子的每条评论进行评论。
media 0 共 1 项
#233 🔗 原推
2026-08-29 13:51:21
Garry Tan
RT @aphysicist: computer, bombard my amygdala with chinese misinformation. make me blindly furious about data centers, electricity generation, and every other technology america needs to win the 21st century. disable critical thinking https://twitter.com/aphysicist/status/2093444689969963302/photo/1
media 0 共 1 项
#232 🔗 原推
2026-08-29 03:19:27
Garry Tan
RT @vivianmshen: never forget where you came from getting messages like this reminds me of everything @ycombinator has done for founders honored to carry on the tradition and pay it forward https://twitter.com/vivianmshen/status/2093500123112468647/photo/1
中文: RT @vivianmshen:永远不要忘记你来自哪里 获取这样的消息让我想起了@ycombinator为创始人所做的一切 很荣幸继承这一传统并向其支付费用
media 0 共 1 项
#231 🔗 原推
2026-08-29 01:51:59
Garry Tan
Datacenters create prosperity and good jobs https://twitter.com/garrytan/status/2093500353849544842/photo/1
中文: 数据中心创造繁荣与良好就业机会
media 0 共 1 项
#230 🔗 原推
2026-08-29 00:45:54
Garry Tan
An unfit SF city supervisor who has only hurt the city and its citizens has quiet quit the job to try to jump to the next rung of utter incompetence Connie Chan doesn’t deserve public office https://growsf.org/news/2026-08-28-connie-chan-quiet-quits-board-of-supervisors/ https://twitter.com/garrytan/status/2093443836228125151/photo/1
中文: 一位不适合的旧金山城市主管，他只对这座城市及其市民造成了伤害，他已安静地辞去工作，试图跳到下一个完全无能的状态 康妮·陈不配担任公职
media 0 共 1 项
#229 🔗 原推
2026-08-28 21:01:19
Garry Tan
Fund treatment and recovery and defund nonprofits that hurt treatment and recovery with their insistence unlimited support for drug use is the only way to “help” It’s harm acceleration
中文: 为治疗和康复提供资金，并通过坚持对药物使用提供无限支持来资助损害治疗和康复的非营利组织 是加速的危害
#228 🔗 原推
2026-08-28 20:02:49
Garry Tan
RT @fourweekmba: “A markdown file is an employee.” Skillify the feat of strength. Reuse it. It does the job perfectly, every time. @garrytan on @a16z with @eriktorenberg https://twitter.com/fourweekmba/status/2093428407736652155/video/1
中文: RT @fourweekmba:“一个markdown文件是员工。” 巧妙地实现力量的壮举。重复使用。每次都能完美地完成这项工作。 @garrytan 与 @eriktorenberg 在 @a16z 上
🎬
视频
#227 🔗 原推
2026-08-28 20:01:22
Garry Tan
UCSF tenured professor lying about policies so that her nonprofit crony friends can keep collecting that sweet state and city funding
中文: 加州大学旧金山分校的终身教授在政策上撒谎，以便她的非营利亲信朋友能够继续收集那份甜蜜的州和城市资金
#226 🔗 原推
2026-08-28 13:45:47
Garry Tan
Reform CEQA and we can build housing again in California
中文: 改革CEQA，我们可以再次在加州建设住房
#225 🔗 原推
2026-08-28 13:44:14
Garry Tan
RT @SusanDReynolds: McCoy erased HealthRight 360 from his bio because he lied about the numbers on the Linkage Center and got caught. They also have an overdose problem — one of their execs overdosed at his desk, and they’re being sued by the family of a young man who overdosed in their care.
中文: RT @SusanDreynolds:麦科伊从他的简历中删除了HealthRight 360，因为他在Linkage Center上对数据撒了谎，便被抓了。他们还存在药物过量问题——其中一名高管在办公桌前服药过量，而一名年轻男子的家人正在起诉他们，而这个年轻男子在治疗中服药过量。
#224 🔗 原推
2026-08-28 13:43:04
Garry Tan
The SF Dem party is the only central committee that is trying to rebuild state capacity and standing up to the deranged Democratic Socialists It’s a real movement and I’m proud of the local work to push back bad policies. It’s a real win.
中文: 旧金山民主党是唯一一个试图重建国家能力并抵制那些精神错乱的民主社会主义者的中央委员会 这是一场真正的运动，我为当地推动不良政策的努力感到自豪。这是一场真正的胜利。
#223 🔗 原推
2026-08-28 13:42:12
Garry Tan
RT @kane: The @CA_Dem’s naming their law that allowed early release of traffickers and robbers the “Safe Neighborhoods and Schools Act” is the most Orwellian thing I’ve ever heard.
中文: RT @kane:@CA_Dem 的“法案”是我听过的最奥威尔式的，是他们允许提前释放贩运者和劫匪的法律。
#222 🔗 原推
2026-08-28 13:40:07
Garry Tan
RT @AlmostMedia: I am sick of bickering over who benefits from American startups. I’m probably close to a centrist figure in this debate. I have invested in flyover state white boys & I am a libertarian with right leaning sympathies (save for big gov cultural statists). We ALL benefit from the extreme successes that come from breakthroughs in the “non correlated returns” category of investing. Americans benefit first because typically they are our companies. There is no second choice to America as a market in almost every industry (save crypto which I blame entirely on Biden for his idiotic choke point nonsense and we should fix this no matter what the big banks say) It’s impossible to capture all the surplus from that benefit. That’s where I’m annoyed at the left most. They ignore this simple fact. But equally I’m irked by the right saying we can only invest in Americans & hire Americans. That’s stupidity. There aren’t that many of us for one thing. Though I wish we had more Americans. Have some babies. I would if I could. Ask yourself if you want to be the richest most powerful nation on earth? Yes? Me too! So we must continue to produce surpluses and have a positive sum outlook that presumes new markets create new opportunities. We have gone to space so don’t give me guff about the carry capacity of the planet. Mine the moon. Go to Mars. Market economies without gummed up regulatory capture & stupid cartel monopolies (looking at you American Medical Association) simply do better. Some of the biggest successes we’ve had as a nation have come from receiving & nurturing the talents other countries abuse with their cartels, oligarchs & monopolies. Shoot we actively stole other countries best talent in the wake of some wars. I think this is a good thing. We fought communism for decades & won over the world in that effort. We should continue to do so. Why? Because that means more tax revenue, more jobs, new market opportunities, and broader horizons for our children. And yes those are benefits for Americans first. Where do you think the taxes go? If a few other allies and nationalities choose to join us in that effort we should rejoice. Our great experiment has been ongoing for 250 years. I’d like it to continue for another 250 years. But it won’t if we act like shitty zero sum defeatists. I wrote this in one go so excuse typos and blather. If you act like an asshole in response I’ll block hou. Play nice. Let’s win!
#221 🔗 原推
2026-08-28 13:37:06
Garry Tan
Maybe go to these SFGATE popups and ask the local journalists why they do so much heavy lifting for politicians who support waste and grift, ban Algebra in middle schools, allow chaos on the streets, giving money to people who do drugs in the street until they die SFGATE and much of SF Bay Area legacy media is notorious for speaking power to truth Maybe they will flip it if we tell them we are sick of it
#220 🔗 原推
2026-08-28 13:35:54
Garry Tan
Maybe go to these SFGATE popups and ask the local journalists why they do so much heavy lifting for politicians who support waste and grift, ban Algebra in middle schools, allow chaos on the streets, giving money to people who do drugs in the street until they die That’s speaking power to truth Maybe they will flip it if we tell them we are sick of it
中文: 也许去这些SFGATE的弹出窗口，问问当地记者，为什么他们为支持浪费和垃圾的政客们承担如此沉重的负担，禁止在中学就读的阿尔盖布拉，允许街头发生混乱，把钱给那些在街头吸毒直至死亡的人 这就是对真理的说法 也许如果我们告诉他们我们对此感到厌烦，他们就会翻转它
#219 🔗 原推
2026-08-28 13:34:48
Garry Tan
Driving out the people who pay taxes is bad policy period Normal people will pay more taxes while simultaneously healthcare and schools will get worse But sure “defending billionaires” is a helluva meme 👀
中文: 将纳税的人赶出去是糟糕的政策时期 普通人会缴纳更多税款，同时医疗保健和学校也会变得更糟 但“捍卫亿万富翁”确实是一个地狱般的表情包👀
#218 🔗 原推
2026-08-28 13:24:23
Garry Tan
RT @SusanDReynolds: If you want to know why D8 supervisor candidate Gary McCoy has erased his career at HealthRight360, read this article ⬇️⬇️⬇️ https://www.marinatimes.com/the-grift-that-keeps-on-giving
#217 🔗 原推
2026-08-28 13:18:30
Garry Tan
RT @PFFeeley: Here's what the AI job collapse narrative gets wrong about the productivity boom. @pmarca told @lennysan the last 50 years ran at half the pace of the prior era and a third the pace of a century ago. Almost no real technological progress in the economy. Remarkably little job turnover as a result. Even if AI triples productivity growth, a massive deal, job turnover only returns to the 1870 to 1930 rate. People living through that period did not experience collapse. They experienced a world awash with opportunity. New careers. New products. New services. A huge part of the modern world was invented and scaled in that window. Tripling the pace of change does not produce mass unemployment. It produces faster economic growth and faster job growth. Some task-level and job-level substitution will happen. It will be swamped by the macro effects of growth and innovation. Hiring blooms follow.The panic assumes dynamism destroys work.  History treated that same dynamism as the engine of new work.
🎬
视频
#216 🔗 原推
2026-08-28 12:13:53
Garry Tan
RT @AGHamilton29: Keep this in mind when you see posts with tens of thousands of likes that happen to fit with the information war positions of certain adversaries. You think this is the only CCP bot farm? Any idea how many Iranian/Pakistani/Russian active bot farms currently exist on here and other platforms?
中文: RT @AGHamilton29:当你看到那些点名数点恰好与某些对手的信息战立场契合时，请记住这一点。 你认为这是唯一的CCP机器人农场吗?目前这里和其他平台上有多少伊朗/巴基斯坦/俄罗斯活跃的机器人农场?
#215 🔗 原推
2026-08-28 12:02:47
Garry Tan
RT @AaronLCannon: 2023: we invented the AI interviewer. 2026: we invented the AI user. Digital Twins by @OutsetAI is live today. Accurately simulating humans is now possible. But it has done by building one-to-one synthetic twins anchored on real humans, not just asking a foundation model to regurgitate population level responses. That's why we did it. We reinvented the way companies understand their customers a few years ago. We built the most rigorous and high scale mechanism to capture and process customer data. And since then, we've captured millions of deep, multimodal human responses around the world. The future of insights will be a combination of understanding and simulating human experience. We're doing both.
中文: RT @AaronLCannon:2023:我们发明了人工智能面试官。 2026年:我们发明了人工智能用户。 @OutsetAI 的 Digital Twins 今日上线。 准确模拟人类现在是可能的。但它通过构建以真实人类为基础的一对合成双胞胎，而不仅仅是要求一个基础模型来反省种群水平的反应。 这就是我们这么做的原因。 几年前，我们重新发明了企业理解客户的方式。我们建立了最严格、最大规模的机制，以捕获和处理客户数据。从那时起，我们在全球捕捉到了数百万个深刻的、多模式的人类反应。 洞察的未来将是理解与模拟人类经验的结合。我们两者都在做。
🎬
视频
#214 🔗 原推
2026-08-28 12:00:58
Garry Tan
RT @Noahpinion: In the end, I think, the DSA will go down not because people remember Pol Pot or Stalin, or because people realize their economic ideas are bad, or because they support Islamism, but because people are fed up with progressive governance.
中文: RT @Noahpinion:我认为，最终《民主安全法》的下降并非因为人们记住了波尔布特或斯大林，或者因为人们意识到他们的经济观念是坏的，或者因为他们支持伊斯兰主义，而是因为人们对进步的治理已经受够了。
#213 🔗 原推
2026-08-28 12:00:29
Garry Tan
RT @sporadica: TIME has done it again, bravo, incredible. Apparently Paris Hilton is one of the world's most influential people in AI, but Demis Hassabis, Jensen Huang, Andrej Karpathy, Alex Wang, Liang Wenfeng are not? lmao https://twitter.com/sporadica/status/2093017479325053394/photo/1
中文: RT @sporadica:时间再次做到了，勇敢，难以置信。 显然，帕丽斯·希尔顿是人工智能领域最具影响力的人物之一，但德米斯·哈萨比斯、黄仁勋、安德烈·卡尔帕蒂、王亚历克、梁文峰却并非?lmao
media 0 共 1 项
#212 🔗 原推
2026-08-28 12:00:06
Garry Tan
RT @stephenbalaban: 100% Agree @GavinSBaker. Now, I'm going to debunk every single piece of mis-information about data centers that I've ever seen online: > DISINFORMATION: Data centers use a lot of water. FACT: Most AI data centers are closed loop dry cooler design that barely consume any water at all. A 20MW 10,000 GPU cluster needs 20,000 gallons to charge and doesn't use material water after that. That's less than 1/4 of the amount of water used to raise the amount of beef the average American consumes in a year. That means one of the largest data centers in the world, at 2GW, consumes less water than 20 people's annual beef consumption or three American's total water footprint (@ ~600k gallons per year for the average American). > DISINFORMATION: Data centers increase electricity prices. FACT: Data centers bring down a community's average electricity bill because they spur development of new power generation. Most data centers bring their own power generation, pay fees to the utility to build more power plants, or bring their own power storage to actually reduce the cost of energy in their community. Northern Virginia has some of the most data centers per capita and has low power costs. See research from the Electric Power Research Institute https://www.marketplace.org/story/2026/07/10/data-centers-lowered-electric-bills-in-some-places-for-now and https://t.co/KK04Pl8998. >DISINFORMATION: Data centers don't create jobs FACT: Data centers create hundreds of jobs at construction and a high number of exceptionally high paying jobs (many over $120k/year) for their ongoing operation and maintenance. No, they're not similar to an automotive plant in terms of the jobs / MW but they are phenomenal additions to an industrial park and community. > DISINFORMATION: Data centers are constantly super noisy FACT: Data centers are one of the quietest industrial buildings in the world. Drive by a data center in your city and roll down your window. They are very quiet. In fact, a metal mill is FAR louder than a data center---up to 10,000,000 more sound energy is produced by a mill and perceptually a mill can be 128x louder than a data center. > DISINFORMATION: Data centers emit noise pollution from diesel engines running all the time. FACT: Emergency generator sets are almost never turned on unless the grid connection goes out. This would be less than 0.5% of the time! > DISINFORMATION: Data centers are surveillance systems FACT: Data centers actually just host computers owned by private companies that are used to make phone calls, access the internet, run ChatGPT, Claude, X, Instagram, Facebook, Google, play multiplayer video games, and every thing you ever do online. That's it. > DISINFORMATION: Data centers make water dirty FACT: AOC held up a bottle of dirty water and said this is what data centers do. Completely false, well water disruption is a rare side effect of EVERY construction site in the history of humanity. This is not at all unique to data centers and is a complete red herring. > DISINFORMATION: Data centers get massive tax breaks and drain tax dollars from towns FACT: Data centers pay massive amounts of tax and often times huge % of an annual budget of a county or town! It's an absolute bonanza for the towns that are lucky enough to welcome a data center. New schools, new roads, new services are all paid for. Even if there is a sales tax rebate we're still talking tens of millions of new tax revenue. In 2022, Virginia’s data centers collectively generated $1 billion in local tax revenue and $640 million in state tax revenue! That was 2022. https://netchoice.org/wp-content/uploads/2026/08/Kentucky-Statewide-Data-Center-Report_8.11.26.pdf Feel free to reply here if you have any other question you'd like answered about data centers happy to provide actual facts.
#211 🔗 原推
2026-08-28 11:59:58
Garry Tan
RT @ledwards: It turns out that Manny was the victim of a malicious lie that has been repeated to exhaustion throughout the local media, and encouraged by his opponent, someone with an extensive rap sheet that CAN be substantiated… A lot of people owe Manny an apology.
中文: RT @ledwards:事实证明，曼尼是当地媒体一再被精疲力尽的恶意谎言所害，而反对者正是这一谎言的受害者，而对手的说唱单片也已被证实。 很多人欠曼尼一个道歉。
#210 🔗 原推
2026-08-28 11:59:35
Garry Tan
RT @AdamNMayer: Bucking state party, S.F. Dems vote to oppose California billionaire tax https://www.sfchronicle.com/politics/article/sf-dems-billionaire-tax-22401537.php Under the steady leadership of @NancyTungSF, the San Francisco Democratic Party has regained its sanity.
中文: RT @AdamNMayer:抵制州党，旧金山在@NancyTungSF的持续领导下，民主党投票反对加州亿万富翁税收 ，旧金山民主党已重新恢复其理智。
#209 🔗 原推
2026-08-28 11:58:45
Garry Tan
RT @zaid: recalling chesa might go down as one of the best decisions by any major city ever https://twitter.com/zaid/status/2093156453510980071/photo/1
media 0 共 1 项
#208 🔗 原推
2026-08-28 11:57:50
Garry Tan
RT @ycombinator: Outset (YC W23) is building an AI customer research platform used by companies like Google, Microsoft, and Nestlé. Its AI interviewers have now conducted millions of conversations, giving companies the scale and speed of a survey with the depth of a one-on-one interview. In this founder fireside, @OutsetAI co-founder and CEO @AaronLCannon sits down with YC's @harjtaggar to talk about building a product before the market was ready for it, how they helped create the category of AI-moderated research, and what comes next. Aaron shares how they're expanding beyond interviews with Outset's new Simulations Lab and Digital Twins, which let companies simulate their customers and get feedback on everything from messaging and pricing to new products. 00:24 — What Outset Does 01:18 — Building a Category Before the Market Was Ready 06:31 — Landing the First Enterprise Customer 08:14 — How Better AI Models Changed the Product 12:57 — When the Market Finally Caught Up 14:45 — From AI Interviews to Customer Simulations 19:40 — Aaron’s Path to Becoming a Founder
🎬
视频
#207 🔗 原推
2026-08-28 11:56:10
Garry Tan
RT @parkerconrad: A really unique take on MCP from Rippling - our MCP lets agents write code to get stuff done in Rippling, which dramatically expands the aperture of what you can accomplish.
中文: RT @parkerconrad:Rippling 对 MCP 的独特之名——我们的 MCP 让代理人员编写代码，以在 Rippling 中完成操作，从而显著扩展了您所能实现的操作范围。
#206 🔗 原推
2026-08-28 11:55:58
Garry Tan
RT @Suhail: A late night thought: there will be a scarce amount of hackers who know things quite deeply enough still that they can steer a strong AI to an even stronger outcome because they know what to ask, what to guide it toward, and how to end up solving the right thing.
中文: RT @Suhail:深夜想到:将会有大量黑客深入了解事物，能够引导强大的人工智能取得更强烈的结果，因为他们知道该问什么，需要引导它走向什么，以及如何最终解决正确的问题。
#205 🔗 原推
2026-08-28 11:55:43
Garry Tan
RT @Twolfrecovery: Gary McCoy is unqualified to be the D8 Supervisor in San Francisco. He pushed radical harm reduction onto San Francisco that nearly destroyed the city during Covid and sued random people for shit posting on Twitter. Uh oh, he might sue me. 👀
中文: RT @Twolfrecovery:加里·麦考伊没有资格担任旧金山的D8主管。他将彻底的危害降低推向了旧金山，而旧金山在疫情期间几乎摧毁了这座城市，并因在推特上发布垃圾信息而对随机人群提起诉讼。哦，他可能会起诉我。👀
#204 🔗 原推
2026-08-28 11:55:34
Garry Tan
RT @HumanProgress: Young people are constantly told that democracy is dying, capitalism is exploitation, technology is dehumanizing, climate change means collapse, and the next generation will inherit a poorer and more dangerous world. It's no wonder that so many of them are anxious about the future. Serious problems deserve serious attention, but teaching young people to interpret every challenge as evidence of impending catastrophe doesn’t make them wiser. It teaches them helplessness.
中文: RT @HumanProgress:年轻人不断被告知民主正在消亡，资本主义是剥削，技术是非人化的，气候变化意味着崩溃，下一代将继承一个更贫穷、更危险的世界。 难怪他们中有这么多人对未来感到焦虑。 严重的问题值得认真关注，但教导年轻人将每一个挑战都解读为即将来临的灾难，并不能让他们变得更明智。它教会他们无助。
media 0 共 1 项
#203 🔗 原推
2026-08-28 11:10:46
Garry Tan
RT @NinaDSchick: We are turning energy into non-biological Intelligence. Data centers are literally 'factories' that produce Intelligence. Electricity is the essential input. The electricity powers computation, and the computation produces and processes the tokens through which machine Intelligence operates. As these systems become more capable and autonomous, they consume vastly more tokens, driving ever greater demand for computation and electricity. OpenRouter estimates that an agentic request already consumes around 15 times as many tokens as a human-led request on its platform. This fundamentally changes the nature of electricity demand as our demand for useful Intelligence has no obvious ceiling. If more computation can produce more valuable work, we will keep finding more work for machines to do - and so, we will need ever more electricity. We got future energy demand so wrong because we did not foresee that non-biological Intelligence could be produced industrially; and that we would need an entirely new energy paradigm to support that.
中文: RT @NinaDSchick:我们正在将能量转化为非生物智能。 数据中心实际上是产生智能的“工厂”。电力是必需的输入。电力驱动计算，计算产生并处理机器智能运行的令牌。 随着这些系统变得更加强大且自主，它们消耗的代币却更加丰富，导致计算和电力需求不断增加。OpenRouter 估计，一个代理请求在其平台上的代币消耗量约为人类主导请求的 15 倍。 这从根本上改变了电力需求的本质，因为我们对有用情报的需求没有明显的上限。如果更多的计算能够产生更有价值的工作，我们将不断为机器寻找更多的工作——因此，我们将需要更多的电力。 我们未来的能源需求如此错误，是因为我们没有预见到非生物智能可以由工业生产;而我们需要一种全新的能源模式来支持这一模式。
🎬
视频
#202 🔗 原推
2026-08-28 11:09:01
Garry Tan
RT @GavinSBaker: Data centers are awesome for America in every way.
中文: RT @GavinSbaker:数据中心在美国的所有方面都非常出色。
#201 🔗 原推
2026-08-28 11:08:25
Garry Tan
SF YIMBY Action has been a trash organization for years Don’t join them and don’t give them money. Support CA YIMBY and YIMBY Law
中文: SF YIMBY Action 多年来一直是一个垃圾组织 不要加入他们，也不要给他们钱。支持CA YIMBY和YIMBY法律
#200 🔗 原推
2026-08-28 05:11:40
Garry Tan
RT @AGHamilton29: X has now confirmed a Chinese bot farm of 200K fake accounts intentionally trying to manipulate public opinion against data centers. People might want to consider how common this has become and how it is manipulating multiple policy discussions on social media. https://twitter.com/AGHamilton29/status/2093157591119450512/photo/1
中文: RT @AGHamilton29:X 现已证实，一家中国机器人养殖场拥有 2亿个虚假账户，故意试图操纵公众舆论，以对抗数据中心。 人们可能需要思考这种情况已经变得多么普遍，以及它如何操纵社交媒体上的多项政策讨论。
media 0 共 1 项
#199 🔗 原推
2026-08-28 05:11:01
Garry Tan
The Tampa Bay Wave of influential people in AI
中文: 坦帕湾人工智能领域有影响力的人物浪潮
#198 🔗 原推
2026-08-28 00:13:39
Garry Tan
The Tampa Bay Wave of influential people
中文: 坦帕湾的有影响力的人
#197 🔗 原推
2026-08-28 00:13:03
Garry Tan
RT @kwindla: Introducing PhoneLLM, an open model for voice agents. GPT 5.6 Terra performance on typical voice agent tasks at 1/3 the latency and 1/18 the cost. For voice agents, we need models that are both very low latency and very good at tool calling and instruction following. There's a trade-off here, and we often have to compromise on either latency or capability when building voice agents. With PhoneLLM (and the training and data stack that made this model possible) we're fixing this problem. For the last couple of years, most of the effort in frontier model development has gone towards leveraging test-time compute. Which is awesome! Models of all shapes and sizes are available that perform really, really well ... if you have "thinking" turned on for your model. But if you need your agent to respond at voice conversation speed, you can't use thinking models. PhoneLLM is a full-weights fine-tune of NVIDIA Nemotron Nano 30B. We trained on a wide range of real-world telephone and customer support use cases. The training focused on taking the excellent Nano 30B base capabilities and teaching the model to do typical voice agent tasks with thinking disabled. The results are really good: accurate tool calling and concise, on-topic responses in long conversations. And fast: TTFAT measured server-side is <100ms if you run PhoneLLM on a lightly loaded B200. :-) But seriously, when we characterize model latency, we do it with full, end-to-end, batched request simulations using real Pipecat voice agent pipelines. You can serve more than 80 concurrent agents on a single B200 with P95 end-to-end TTFAT <600ms. Including network overhead. That's an LLM cost-per-minute around $0.0025. (1/4 of a cent.) At a latency lower than any third-party API offers today. More details about this model, including weights on @huggingface, how to spin it up with one click on Modal, and a starter project repo you can clone, are in the thread ...
中文: RT @kwindla:推出语音代理的开放模式 PhoneLLM。 GPT 5.6 Terra 在典型的语音代理任务中，延迟时间为 1/3，且成本为 1/18。 对于语音代理，我们需要既低延迟又擅长工具调用和指令跟踪的模型。 这里存在一种权衡，在构建语音代理时，我们常常不得不在延迟或能力方面做出妥协。通过PhoneLLM(以及使该模型成为可能的训练和数据栈)，我们正在解决这个问题。 过去几年里，前沿模型开发的大部分工作都转向了利用测试时间计算。太棒了!各种形状和尺寸的型号都提供，性能非常好......如果你对自己的型号有“思考”的调整。 但如果你需要你的经纪人以语音对话的速度做出回应，就无法使用思维模型。 PhoneLLM 是 NVIDIA Nemotron Nano 30B 的全重量调。我们基于多种现实世界的电话和客户支持用例进行了培训。培训的重点是具备出色的Nano 30B基础能力，并教导模型在思维障碍的情况下完成典型的语音代理任务。 结果非常好:在长时间的对话中，工具调用准确，主题回答简洁。 速度:如果在加载速度轻微的 B200 上运行 PhoneLLM，TTFAT 测量的服务器端为 <100ms。 :-) 但严肃地说，当我们描述模型延迟时，会使用真正的Pipecat语音代理管道进行完整的端到端批量请求模拟。 单个B200上可使用P95端对端TTFAT <600ms，为80多个并发代理提供服务。包括网络开销。按每分钟0.0025美元(1/4美分)的LLM价格。延迟低于当前任何第三方API的提供。 有关此模型的更多详细信息，包括@huggingface 上的权重、如何单键旋转，以及可克隆的入门项目。
🎬
视频
#196 🔗 原推
2026-08-28 00:12:15
Garry Tan
RT @hackernews: Many of us are now using coding agents everyday and, it would seem, GitHub pull requests are developing a bit of an accent as a result. An analysis of 47,000 PRs found a distinct vocabulary cluster dominated by words like “load-bearing,” “quietly,” “latent,” and “genuine" that grew from 1% of the corpus in January to 45% today. The project samples 100 PRs a day, filters out bot and agent accounts, and groups descriptions into eight different writing styles based on word choice. For those of us familiar with their work, this cluster certainly sounds like the "load-bearing" vocab of LLMs. https://news.ycombinator.com/item?id=49461817
中文: RT @hackernews:我们许多人现在每天都在使用编码代理，而且看起来，GitHub 的拉取请求因此变得有些口音。 对47000名PR的分析发现，一个独特的词汇集群以“承重”、“静默”、“潜在”和“真正”等词汇为主导，从1月份的1%增长到如今的45%。 该项目每天筛选100个PR，筛选出机器人和代理账户，并根据文字选择将描述分为八种不同的写作风格。对于我们这些熟悉他们作品的人来说，这个集群听起来确实像是LLM的“承载式”疫苗。
#195 🔗 原推
2026-08-28 00:12:07
Garry Tan
RT @jasonlk: California is adding sales tax to SaaS starting Jan 1 😥Not sure they could have picked a worse time - 7.25% state + local district taxes, so 8-10% for most customers - Every CA-based customer you have is getting a price increase you didn't send - And if you're based here, your entire software and AI budget goes up 8-10% in 2027 Stacks on top of vendor uplifts. A 7% renewal increase becomes 16%. 👉Custom software and AWS-style infrastructure are excluded 🤷Nobody knows yet where usage-based AI spend lands.
media 0 共 1 项
#194 🔗 原推
2026-08-27 21:19:05
Garry Tan
RT @stanine: Today, @Rippling launched the Rippling AI Slackbot, so you can sorta treat Rippling like you treat @Grok on X... tag it to get the truth, hand off work, and do research. https://twitter.com/stanine/status/2093025170353221799/photo/1
中文: RT @stanine:今天，@Rippling 推出了 Rippling AI Slackbot，这样你就可以像在 X 上对 @Grok 一样对待 Rippling 来标记真相，进行处理，并进行研究。
media 0 共 1 项
#193 🔗 原推
2026-08-27 20:31:17
Garry Tan
Seems likely that on a long enough time frame, AI will generate cash flows faster than the economy can find productive uses for new capital
中文: 似乎在足够长的时间范围内，人工智能产生的现金流速度将超过经济对新资本的有效利用
#192 🔗 原推
2026-08-27 19:23:49
Garry Tan
I just cut token load for GStack by 50% with no reduction in capability Get it now at https://github.com/garrytan/gstack https://twitter.com/garrytan/status/2093026068475380012/photo/1
中文: 我刚刚将GStack的代币加载量减少了50%，而无法降低 立即访问
media 0 共 1 项
#191 🔗 原推
2026-08-27 17:21:15
Garry Tan
RT @paulg: I believe the most qualified person should get the job.
中文: RT @paulg:我认为最有资格的人应该得到这份工作。
#190 🔗 原推
2026-08-27 14:35:38
Garry Tan
RT @Suhail: Need 256 B200s/B300s stat. Cannot wait till next month. DM.
中文: RT @Suhail:需要 256 B200s/B300s 数据。不能等到下个月。DM。
#189 🔗 原推
2026-08-27 11:25:23
Garry Tan
RT @zachcoelius: There is a very simple reason many thousands of addicts come to live in San Francisco. We have the cheapest drugs in the country and they get $900 in cash every month.
中文: RT @zachcoelius:有成千上万的瘾君子来到旧金山生活，原因很简单。 我们拥有全国最便宜的药品，每月可获得900美元现金。
#188 🔗 原推
2026-08-27 10:48:45
Garry Tan
RT @kane: The official policy of @sfgov is that NGOs that it gives tax dollars to do not need to submit audits. We spend more than the entire deficit on corrupt NGOs.
中文: RT @kane:@sfgov 的官方政策是，其提供税款的非政府组织无需提交审计。我们在腐败的非政府组织上花费的不仅仅是全部赤字。
#187 🔗 原推
2026-08-27 03:29:56
Garry Tan
RT @_shaurya35: I think the most useful command in Gstack by @garrytan is /cso Use it everyday and thank me later
#186 🔗 原推
2026-08-27 03:28:48
Garry Tan
RT @Overlap_Tech: Garry Tan: The Rule That Prevents Agent Failures⁣ ⁣ "The most important question to ask here is where is the computation happening? And there are exactly two answers and confusing them causes every agent failure I've ever seen."⁣ ⁣ "Some computation belongs in latent space. Taste, judgment, reading what a human actually wants from a vague request, that lives in the model and you steer it with a markdown file. And then some computation belongs in deterministic space, the arithmetic, the SQL query... Ask an agent or human to seat five people around a table. That's easy. Do it in latent space."⁣ ⁣ Ask it to make custom schedules for 6,000 people in an arena like we just did for you, and your latent space agent needs to write some code to keep track of it. The model fails where we fail." — @garrytan
🎬
视频
#185 🔗 原推
2026-08-27 03:28:35
Garry Tan
RT @ycombinator: Today, more than 3% of the world's lawyers use Legora, and the company has grown from $1 million to $100 million in ARR since launching in October 2024. At Startup School 2026, @WeAreLegora co-founder and CEO @MaxJunestrand shares how they built one of the fastest-growing enterprise software companies in the world, from cold emailing lawyers and moving into a customer's office to freezing sales for six months to rebuild the product. He explains why building a company is ultimately about people, how to create a culture that wants to win, and why founders have to learn to love the hustle. 00:07 — From 3 Engineers to $100M ARR 03:15 — How Legora Got Started 05:09 — Learning the Legal Industry From Scratch 07:22 — Getting Rejected by YC 09:36 — Moving Into a Customer's Office 11:11 — Going From Zero to $1M ARR During YC 13:26 — Why We Froze Sales for Six Months 15:18 — Rebuilding the Product From Scratch 17:26 — Building a Company Is About People 21:51 — Creating a Culture That Wants to Win 25:26 — You Have to Love the Hustle 27:55 — Do Founders Need Domain Expertise? 29:14 — Betting on Models Getting Better 36:29 — When Should You Start a Company? 45:04 — How to Become More Ambitious 48:18 — The Skills Founders Need Today
🎬
视频
#184 🔗 原推
2026-08-27 01:29:24
Garry Tan
RT @Twolfrecovery: I told you. The allegations against Manny Yekutiel were false and brought to bear by a friend of Manny's opponent. Now more than ever, I hope Manny wins D8 Supervisor. There's a few public officials who owe him an apology. #SanFrancisco
中文: RT @Twolfrecovery:我告诉过你。对曼尼·耶库蒂尔的指控是虚假的，由曼尼的一位对手朋友绳之以法。现在比以往任何时候都更希望曼尼能赢得D8主管。有几位公职人员欠他一个道歉。#旧金山
#183 🔗 原推
2026-08-27 01:20:41
Garry Tan
RT @kimmaicutler: Scott Wiener’s “unproven housing ideology” is … checks notes … building housing. https://twitter.com/kimmaicutler/status/2092768972374761934/photo/1
中文: RT @kimmaicutler:斯科特·维纳的“未经证实的住房意识形态”是......支票记录......建造住房。
media 0 共 1 项
#182 🔗 原推
2026-08-27 01:10:28
Garry Tan
RT @vivianmshen: .@MaxJunestrand is a force of nature. Legora grew from $1 million to $100 million in ARR in <2 years, and now over 3% of lawyers worldwide use Legora. I caught him after his talk at YC Startup School to chat backstage about his experience in YC, how his first YC interview went, and why working at a startup is one of the best ways to start your career. his talk is now live, see the link in comments we also have some friendly competition going - is @Starcloud_ or @WeAreLegora YC's fastest unicorn? @PhilipJohnston may need to jump in!
中文: RT @vivianmshen:@MaxJunystrand 是一股自然的力量。在2年内，乐高在ARR的销售额从100万美元增长到1亿美元，如今全球超过3%的律师使用乐高拉。 我在YC创业学校演讲后，在后台聊了聊他在YC的经历、他第一次面试面试的经历，以及为什么在一家初创企业工作是开始你职业生涯的最佳方式之一。 他的演讲现已上线，请参阅评论链接 我们还有一些友好的竞争——是@Starcloud_还是@WeAreLegora YC最快的独角兽?@菲利普·约翰斯顿可能需要加入进来!
🎬
视频
#181 🔗 原推
2026-08-27 01:10:12
Garry Tan
RT @harjtaggar: Developer tools are growing unusually fast because their power users are now agents and agents can write a lot more code than humans. As agents get used for more things, that same pattern should show up for any products used by agents to make better decisions for humans. An agent isn't going to buy me 10,000 jackets but it can search across 10,000 to find the one i should buy. If you're looking for startup ideas, it's worth asking what decisions will agents make on behalf of humans in the future and what infrastructure will they need to make them well?
中文: RT @harjtaggar:开发者工具的增长速度异常迅速，因为现在他们的用户是代理和代理，能够编写比人类多得多的代码。 随着药物被用于更多事物，这种模式应出现在代理使用的任何产品上，以便为人类做出更明智的决策。经纪人不会给我买一万件夹克，但可以搜索一万件，找到我该买的。 如果你正在寻找创业创意，值得思考的是，未来代理人员将代表人类做出哪些决策，以及他们需要哪些基础设施才能让他们获得良好的决策?
#180 🔗 原推
2026-08-27 01:10:07
Garry Tan
RT @kane: Not only was progressive San Francisco Supervisor candidate @GaryMccoySF committing graft at a sketchy NGO, he has charges for Grand Larceny and probation violations. You wouldn’t let him near your children, definitely don’t vote for him. These people are a cancer for @sfgov. https://twitter.com/kane/status/2092756261247054110/photo/1
中文: RT @kane:不仅是旧金山进步派主管候选人@GaryMccoySF在一家草率的非政府组织中实施嫁接，他还被指控犯有大拉克尼罪和违反缓刑罪。 你不会让他靠近你的孩子，绝对不要投票给他。这些人对 @sfgov 来说是癌症。
media 0 共 3 项 media 1 media 2
#179 🔗 原推
2026-08-27 01:08:28
Garry Tan
RT @LeeKuanYimby: She also calls out teachers unions... My lord.
中文: RT @LeeKuanYimby:她还呼吁教师工会......我的主。
#178 🔗 原推
2026-08-27 00:31:35
Garry Tan
RT @mattyglesias: One of the best things the Billionaire Community could do to improve its long-term prospects would be to stop backing anti-housing orgs. Housing scarcity more than anything else is radicalizing young people against capitalism. https://www.slowboring.com/p/to-save-capitalism-we-need-radical
中文: RT @mattyglesias:亿万富翁社区为改善其长期前景所能做的最好的事情之一就是停止支持反住房制度。 住房短缺比其他任何事情都更会使年轻人对资本主义采取极端化。
#177 🔗 原推
2026-08-26 23:24:31
Garry Tan
RT @Rafa_Mangual: I keep seeing people say this. But we don’t have to *imagine* if these cameras existed during Covid. They did! Flock has been in business for 10 years, and other ALPR providers have been providing these to law enforcement for 20 years. There isn’t a single example of ALPRs being used to abuse power during Covid.
中文: RT @Rafa_Mangual:我一直看到人们这么说。但我们不必想象这些摄像头在疫情期间是否存在。他们做到了!Flock 已经营了十年，其他 ALPR 服务提供商已向执法部门提供这些产品已有 20 年。没有一个例子说明在疫情期间ALPR被用于滥用权力。
media 0 共 4 项 media 1 media 2 media 3
#176 🔗 原推
2026-08-26 23:23:41
Garry Tan
RT @koomen: Now that building software is so cheap I am surprised on a ~daily basis at how many opportunities there are to make something at YC better with more of it.
中文: RT @koomen:现在，现在开发软件已经如此便宜了，我每天都会惊讶于，通过更多软件，在YC上拥有多少机会来让某些东西变得更好。
#175 🔗 原推
2026-08-26 23:23:03
Garry Tan
RT @lightberry: MEET LUMI THE WORLD'S MOST INTERACTIVE ROBOT Assembled in SF, shipping in 2026 💛
中文: RT @lightberry:与 LUMI 会对 世界最互动的机器人 于2026年在旧金山组装，发货💛
🎬
视频
#174 🔗 原推
2026-08-26 18:10:16
Garry Tan
RT @paulg: An email I sent to the summer 2007 YC batch to encourage them before Demo Day. It's all still true today. https://twitter.com/paulg/status/2092663545431572514/photo/1
中文: RT @paulg:我在2007年夏季YC批次上发了一封电子邮件，以鼓励他们参加演示日。今天一切依然如此。
media 0 共 1 项
#173 🔗 原推
2026-08-26 17:36:30
Garry Tan
RT @saeverley: Data centers = lower property taxes for residents https://twitter.com/saeverley/status/2092343978481893540/photo/1
中文: RT @saeverley:数据中心 = 居民房产税降低
media 0 共 1 项
#172 🔗 原推
2026-08-26 13:09:10
Garry Tan
RT @hackernews: Instead of generating a finished image, this AI paints by writing editable JavaScript. The project uses reinforcement learning and hand-rated examples to teach Qwen 3.5 how to make a good watercolor. https://news.ycombinator.com/item?id=49411800 https://twitter.com/hackernews/status/2092371442734395794/photo/1
中文: RT @hackernews:这个人工智能通过编写可编辑的 JavaScript 来描绘，而不是生成成品图像。 该项目通过强化学习和手部评分示例，教Qwen 3.5如何制作出良好的水彩画。
media 0 共 1 项
#171 🔗 原推
2026-08-26 12:44:33
Garry Tan
RT @incitafusio: Are we done w/ giving criminals more empathy & excuses over actual victims? Serial felons are given dozens of chances in SF & CA so they can create more victims. What happened to Marsy's laws of Victim's Rights? It seems not to exist in SF & CA as victims are denied timely trials and updates regarding procedures. #3 To have safety of the victim & victim's family considered in fixing amount of bail & release conditions # 9 To a speedy trial and prompt & final conclusion of case #16 To have the safety of the victim, the victim's family and the general public considered before any parole or other post-judgement release decision is made https://sfdistrictattorney.org/victim-services/victims-rights/
media 0 共 1 项
#170 🔗 原推
2026-08-26 05:40:54
Garry Tan
RT @bosmeny: David is one of the fastest moving - and highest agency - founders I’ve ever worked with If anyone finally brings us exoskeletons, it’s going to be him
#169 🔗 原推
2026-08-26 05:40:39
Garry Tan
RT @kane: The official policy of @sfgov is still to give monthly cash stipends to drug users with no conditions about spending it on drugs. https://twitter.com/kane/status/2092359787636027468/photo/1
media 0 共 1 项
#168 🔗 原推
2026-08-26 05:40:08
Garry Tan
RT @alexocheema: The 512GB M5 Ultra will ship in October with a massive 1.2TB/s memory bandwidth. 50% more memory bandwidth than M3 Ultra, 4.4x more than DGX Spark and M4 Pro. Stacking 4 x M5 Ultra, I expect you'll be able to run Kimi K3 / GLM 5.3 faster than the API (&gt;100 tok/sec). https://twitter.com/alexocheema/status/2092248148500492474/photo/1
media 0 共 2 项 media 1
#167 🔗 原推
2026-08-26 05:38:39
Garry Tan
RT @glangley: In Michigan, last month alone, 53 missing people were detected with @flocksafety. Grandparents who lost their way. Kids who were abducted or missing. Yesterday in Fraser, Flock helped locate the suspect in a fatal hit-and-run that killed a female motorcyclist. I sat down with @laurenkostiuk_ and @Local4News to talk about how we’re helping first responders in these cases while also strengthening privacy protections.
🎬
视频
#166 🔗 原推
2026-08-26 05:38:20
Garry Tan
RT @peterjhasson: On one side, you have “defund the police” activists calling to ban license plate readers like Flock On the other side, you have police officers begging you not to strip their ability to use license plate readers to catch missing children https://twitter.com/peterjhasson/status/2092390046682714182/photo/1
media 0 共 2 项 media 1
#165 🔗 原推
2026-08-26 05:37:51
Garry Tan
RT @minilek: California state senators are now weighing in, asking the University of California to bring the SAT/ACT back as soon as possible. @SenOchoaBogh @SenatorChoi @SenRogerNiello Letter 1: “… students entering the University of California are not prepared for the rigorous coursework that is required to complete their degrees… we call upon the Board of Regents to move quickly towards reinstatement without delay” 1/3
media 0 共 2 项 media 1
#164 🔗 原推
2026-08-26 05:37:43
Garry Tan
RT @paulg: In the essay I just wrote about how to prepare students to be founders, I talked about the importance of working on their own projects. In this essay from 2021, I explain in detail how and why to do that. A Project of One's Own: https://paulgraham.com/own.html
#163 🔗 原推
2026-08-26 05:37:30
Garry Tan
RT @SusanDReynolds: The alternative is Shamann Walton’s handpicked successor DionJay Brookter who will keep money flowing to grifter NGOs benefiting their cronies like YCD, which was passed from Dwayne Jones (indicted on 59 fraud counts) to Walton to Brookter. https://thevoicesf.org/shamann-waltons-young-community-developers-subsidiary-got-a-no-bid-contract-on-shipyard-homes-but-wheres-the-money/
#162 🔗 原推
2026-08-26 05:31:13
Garry Tan
RT @snowmaker: Agentmuxer - it's like OpenRouter, but for tools.
#161 🔗 原推
2026-08-26 05:13:09
Garry Tan
Ok apparently this guy came up with it Don’t really care though Clout chaser lmao he is the one who blocked me https://twitter.com/garrytan/status/2092351492397383908/photo/1
media 0 共 1 项
#160 🔗 原推
2026-08-25 20:40:44
Garry Tan
RT @rahul: Man murders 5 people (and four of their pets). Flock cameras help locate him.
#159 🔗 原推
2026-08-25 19:29:40
Garry Tan
lmao https://twitter.com/garrytan/status/2092321194037195022/photo/1
media 0 共 1 项
#158 🔗 原推
2026-08-25 18:40:20
Garry Tan
RT @samdblond: Today, we're launching Monaco into General Availability. We onboarded several hundred hypergrowth startups during our beta. The results are staggering: The average customer added 16 percentage points to their month-over-month revenue growth rate. The average customer tripled their monthly meeting volume. With today's release, Monaco is now the most advanced AI sales platform in the world. Monaco customers have an unfair competitive advantage.
🎬
视频
#157 🔗 原推
2026-08-25 16:12:00
Garry Tan
You need to be cross-harness memory maxxing
#156 🔗 原推
2026-08-25 15:48:21
Garry Tan
RT @CharlesFLehman: "We got a murderer behind bars because of this. We got this guy off the streets because of the Flock system," Indian River County, Fla., Sheriff Eric Flowers said. https://flockstopscrime.com/ https://twitter.com/CharlesFLehman/status/2091865956817494341/photo/1
media 0 共 1 项
#155 🔗 原推
2026-08-25 12:36:08
Garry Tan
RT @emmma_camp_: Can rent control fix the housing crisis? No. In fact, it could make things a whole lot worse. Find out more in my latest for @WSJFreeEx https://twitter.com/WSJFreeEx/status/2091907307844063665/video/1
🎬
视频
#154 🔗 原推
2026-08-25 12:00:20
Garry Tan
RT @peterjhasson: Flock cameras gone. Now police have trouble solving a shooting. Ask yourself: is this what you want in your town?
#153 🔗 原推
2026-08-25 11:59:27
Garry Tan
RT @kane: A store in San Francisco Supervisor @conniechansf’s District 1 was destroyed and robbed last night. Connie Chan ran for Supervisor on “dismantling the police” and is trying to run for Congress right now. https://twitter.com/kane/status/2092009854244409643/photo/1
media 0 共 2 项 media 1
#152 🔗 原推
2026-08-25 11:57:02
Garry Tan
RT @RhysSullivan: My rough prediction here is there's going to be two types of companies: 1. Companies that are ok with basically being an API - they expose an MCP, skills, embrace your agents 2. Companies that charge a ton for API access, try to get you using their own agents, Companies will gravitate to #2 because being API only is scary / feels very replaceable opening opportunities for #1 companies to replace them because they're more agent native I don't have a good answer for what the moat is for #1 companies but it's how I want to be using software
#151 🔗 原推
2026-08-25 05:59:11
Garry Tan
RT @mbeisen: Data Centers already use HALF of the water from the Colorado River. HALF!!! We have to ban Data Centers now!!!! Oh no wait. That’s alfalfa for cows. https://twitter.com/mbeisen/status/2091989084319498586/photo/1
media 0 共 1 项
#150 🔗 原推
2026-08-25 05:57:35
Garry Tan
RT @kane: New feature added to https://t.co/9jCTQWKiDz: You can now filter @sfbos Supervisors' votes. Here's @conniechansf inexplicably voting No (along with @JackieFielder_) to clean up the drug markets. https://twitter.com/kane/status/2092127239035101233/photo/1
media 0 共 2 项 media 1
#149 🔗 原推
2026-08-25 05:57:22
Garry Tan
RT @rahul: People don’t realize that Flock cameras aren’t just good for catching murder suspects, but for convicting them too.
#148 🔗 原推
2026-08-25 01:54:53
Garry Tan
Datacenters create jobs and prosperity, actually https://garryslist.org/posts/data-center-nimbys-are-killing-1-trillion-in-ai-infrastructure https://twitter.com/garrytan/status/2092062820229890209/photo/1
media 0 共 1 项
#147 🔗 原推
2026-08-25 01:33:39
Garry Tan
Conductor Cloud has made me so much more productive and I don't have to keep my Macbook Pro cracked open anymore
#146 🔗 原推
2026-08-25 01:31:19
Garry Tan
RT @paulg: How Universities Should Prepare Founders: https://paulgraham.com/prepare.html
#145 🔗 原推
2026-08-25 01:30:53
Garry Tan
Form a view. Turn it into an artifact or experiment. Put it in contact with reality. Read the result without self-deception. Revise and run again.
#144 🔗 原推
2026-08-25 01:20:32
Garry Tan
RT @davidu: There used to be two Flock cameras right where this happened. Except Defund The Police and DSA activists got them removed. And now today, the first homicide in years. A tragedy. This is deliberate from the pro-crime caucus: https://abc7news.com/post/man-fatally-shot-parking-lot-el-cerritos-del-norte-plaza-police-searching-2-suspects-allegedly-fled-scene/19724763/
#143 🔗 原推
2026-08-25 01:10:23
Garry Tan
RT @ycombinator: Big labs don’t have a monopoly on AI research. More than 20 YC startups have published research recently at top conferences like NeurIPS, ICLR, and ICML. https://www.empirical.health/blog/yc-startups-publishing-ai-research https://twitter.com/ycombinator/status/2092041032104022506/photo/1
media 0 共 1 项
#142 🔗 原推
2026-08-25 00:28:11
Garry Tan
RT @davidsenra: .@sama says without YC, OpenAI never would have existed: “The degree to which @ycombinator totally influenced everything that has happened in the last 20 years of the tech industry, startups, and entrepreneurship, I think, is only sort of understood.” “The changes that YC induced in the whole ecosystem—more leverage going to founders, young technical founders having the ability to raise lots of capital, the ability to work on ambitious things without a very proven resume—I don't think OpenAI would've been possible.”
🎬
视频
#141 🔗 原推
2026-08-24 21:28:15
Garry Tan
RT @eng_khairallah1: this is f**king insane a free github repo by Garry Tan (President & CEO of Y Combinator) with 127K stars just dropped his exact claude code setup here is how you set it up: 1. run the setup script, it auto detects claude code, codex, kiro, factory or opencode 2. run /plan and it fires CEO → design → eng review in one command 3. run /ship and it opens the PR and closes the github issue on merge save and bookmark this no matter what
🎬
视频
#140 🔗 原推
2026-08-24 21:28:05
Garry Tan
RT @dnapway: Sam Altman didn’t ask Paul Graham for advice before launching ChatGPT because he knew exactly what he would say "This is like a meme for many YC founders, where you would go see him for office hours and he would say, ‘you know what you should do?’ And he would shake his finger.” “And sometimes the thing that came after that was great. Sometimes the thing that came after that was terrible.” “But the important thing was there is a creativity and open landscape and just, let's try a lot of things." “He really pushed the startup ecosystem into this world of, you got to ship a V1 you're embarrassed of early. And it doesn't matter if it could be much better. You'll get it much, much better because of the feedback to customers." “I don't even think I asked him before we launched ChatGPT, like, ‘hey, do you think we should launch this thing?’ But I knew what he would say. I knew it was still early. I knew it was still embarrassing. And I knew the right thing was to get it out and get it in front of people."
🎬
视频
#139 🔗 原推
2026-08-24 14:48:31
Garry Tan
RT @vivianmshen: things that make me happy: 1. YC founders helping other founders unpack their shipping containers 2. materials science labs getting set up in SF https://twitter.com/vivianmshen/status/2091605879737237890/photo/1
media 0 共 2 项 media 1
#138 🔗 原推
2026-08-24 14:46:29
Garry Tan
Legalize building housing
#137 🔗 原推
2026-08-24 14:44:57
Garry Tan
RT @paulg: When she was younger, Jessica sometimes felt she faded into the background. https://twitter.com/paulg/status/2091753456264712201/photo/1
media 0 共 1 项
#136 🔗 原推
2026-08-24 14:44:12
Garry Tan
RT @pitdesi: San Francisco did more than defunding police, they did the SF thing… created more bureaucracy to “reimagine” policing ...and then put a convicted felon at the helm of the Sheriff’s oversight board, and then made the board extremely hard to fire. In 2020, Chesa Boudin, Dean Preston and gang pushed for a new Sheriff’s Oversight Board to “give voice to those impacted,” provide “accountability,” and join the “national movement for justice reform.” They put convicted felon William Palmer as president, for his “lived experience.” He had served 31 years in prison. Since joining the board, Palmer has been accused of violence by multiple women, including a rape charge. Now a 26 y/o woman says he gave her meth and pressured her into sex while she stayed at his apartment, where she overdosed on fentanyl. Palmer says he was acting as her “life coach.” Everyone agrees Palmer should not be running this board, but because of how SF designed it, removing him is extraordinarily difficult. Commissioners like Palmer are protected by a process requiring formal misconduct charges, an Ethics Commission proceeding and a supermajority of the Board of Supervisors. We should have passed Prop D in 2024, which would have removed this entire commission from the Charter along with many others, but it would have given more power to the mayor and I think people were reluctant, given Mayor Breed's administration. We have a chance to fix this a little bit: vote YES on A and F this November. Prop A is the commission-cleanup measure: it cuts the number of Charter commissions, eliminates or merges some, and moves others out of the Charter so they can be changed or abolished without another citywide election. Prop F fixes the problem illustrated by Palmer: most commissioners would serve at the pleasure of whoever appointed them, so the Board could simply remove its own appointees without this ridiculous misconduct process.
#135 🔗 原推
2026-08-24 14:41:05
Garry Tan
RT @kimmaicutler: El Cerrito ended its Flock contract earlier this summer. Wonder how long it will take to catch these folks after this shooting at a family restaurant today.
#134 🔗 原推
2026-08-24 13:05:19
Garry Tan
Current meta: OpenAI is actually the much more open lab
#133 🔗 原推
2026-08-24 13:02:29
Garry Tan
There will still be API’s and acl’s and sql and underlying data structures that are deterministic It’s just that software companies have to build the AI harness and full solution for their customers or be subsumed by it
#132 🔗 原推
2026-08-24 12:56:25
Garry Tan
Prediction: systems of record will need to become AI harnesses or face replacement by agents
#131 🔗 原推
2026-08-24 04:22:06
Garry Tan
RT @LeeKuanYimby: Nobody against data centers understands data centers. Nobody against Flock cameras understands how they work or what they do. Nobody demanding Medicare for All knows how Medicare currently functions or what it would cost. Nobody saying “tax the rich” knows how much the rich already pay in taxes. Nobody saying “seize the means of production” knows how to build or maintian the means of production. Nobody saying “fund education” knows how much we already spend on education. Nobody saying “no new wars” understands geopolitics. Nobody saying “ban corporate landlords” knows what percentage of homes corporations actually own. Nobody saying “developers are greedy” understands why housing costs what it does. Nobody demanding price controls understands what happens to supply when you cap prices. Nobody saying billionaires could “end homelessness” knows what governments already spend on homelessness. Nobody saying “just build public housing” knows what it costs government to build a single unit. Nobody demanding free college knows how much universities already receive from taxpayers. Nobody saying “just nationalize it” can look at what the government already operates and understand why that's a bad idea. Modern politics increasingly rewards having the strongest possible opinion about something you have spent the least possible time trying to understand. The American voter increasingly demand sensationalism and they have no idea how wealthy they are compared to the rest of the world. They may end up being the cause of their self inflicted belt tightening and they are in no way ready for an ounce of actual poverty. We need immediate competence to take over sensationalism and social media, especially tiktok is the enemy of that.
#130 🔗 原推
2026-08-23 14:33:33
Garry Tan
RT @peterjhasson: “Mother credits Flock Safety cameras for aiding in conviction of daughter's killer” https://www.rocketcitynow.com/article/news/state/mother-credits-flock-safety-cameras-for-aiding-in-conviction-of-daughters-killer/525-ec1b2c74-3ddb-4d4b-bfd3-3ade0dead6e5
#129 🔗 原推
2026-08-23 14:33:00
Garry Tan
RT @dnapway: David Sacks explains the step by step playbook for banning open source models "An open source ban is coming. They're not going to call it that. They're going to say that we simply have to apply the same standards to open models that we apply to closed." "Here's how they do it step by step. Let me explain how regulatory capture actually works." "Dario wants an FDA for AI. But he doesn't have enough political support for that. So instead, they do this Trojan horse of a FINRA for AI. They call it self-regulating. It's not really, but anyway, that gets them off the ground." "Now they've created the standard setting organization. Now they've got pre-release model testing. Then the pressure grows to codify that in law. And then what they do is they say, look, all these standards need to apply equally to all models." "But here's the problem with that. Open models and closed models, as you know, are technologically different."
🎬
视频
#128 🔗 原推
2026-08-23 14:06:35
Garry Tan
RT @kane: Watching Berkeley progressives protest turning this abandoned lot littered with crackheads into student housing for half a decade was radicalizing.
#127 🔗 原推
2026-08-23 13:57:00
Garry Tan
RT @CharlesFLehman: Scott Taylor is accused of carjacking an elderly woman at a red light and leading cops on a chase that hit nearly 140 miles per hour. "Flock cameras tracked this guy all the way down 95," [Volusia County Sheriff Mike] Chitwood said. "Flock cameras enabled us to know what direction he was going till police can get visual on him again." https://flockstopscrime.com/
media 0 共 1 项
#126 🔗 原推
2026-08-22 13:50:32
Garry Tan
RT @kimmaicutler: Where are the Big Alfalfa protests?
#125 🔗 原推
2026-08-22 13:50:06
Garry Tan
RT @traestephens: That’s weird…it’s almost like not banning TikTok has had real world consequences as the US population continues being poisoned by CCP propaganda. Who woulda guessed?!
#124 🔗 原推
2026-08-22 00:26:47
Garry Tan
RT @DanielDiMartino: That's it. That's the sign https://twitter.com/DanielDiMartino/status/2090795522781057307/photo/1
media 0 共 1 项
#123 🔗 原推
2026-08-22 00:13:48
Garry Tan
RT @EWErickson: I have written a small defense of Flock cameras. I know someone whose child was kidnapped and recovered unharmed within hours due to Flock. https://www.ajc.com/opinion/2026/08/opinion-opponents-of-the-flock-camera-system-are-misguided/
#122 🔗 原推
2026-08-22 00:13:44
Garry Tan
RT @garryslist: "We are living through the largest departure of Asian Americans from the Democratic Party in our nation's history. But it isn't because we've changed; it's because the Party has moved away from our core beliefs." @notGarryTan writes about his family's flight from communist China, the party that welcomed them in San Francisco, and what it means to continue believing in safety, schools, and prosperity while the party drifts.
media 0 共 1 项
#121 🔗 原推
2026-08-22 00:13:13
Garry Tan
RT @pitdesi: Interesting to me that so many people on Twitter are anti-Flock. I wonder what the data would be like if you just cut it for SF, where we've seen meaningful quality-of-life improvements from Flock cameras. https://twitter.com/pitdesi/status/2090900597642031341/photo/1
media 0 共 1 项
#120 🔗 原推
2026-08-22 00:12:57
Garry Tan
RT @RichardHanania: In Houston, someone tried to kidnap a third grader. There were Flock cameras nearby. But someone had cut them down, so the attempted kidnapper is still at large. I'm becoming radicalized against the anti-Flock activists. You people have no arguments. Children get raped and killed. You shrug your shoulders. "I don't care if a thousand children are tortured to death as long as one cop doesn't look up his girlfriend's license plate." You are unable to reason about public policy issues. You are the reason we can't have nice things. You should be ashamed of yourselves. More here. https://www.richardhanania.com/p/flock-mass-surveillance-and-the-collapse
media 0 共 1 项
#119 🔗 原推
2026-08-22 00:12:37
Garry Tan
RT @saranormous: from the union workers: the American dream requires growth and production, not just redistribution AI and energy are the economy’s productivity and growth engine. to be anti-datacenter is to be anti-labor https://twitter.com/saranormous/status/2090831867222343996/photo/1
media 0 共 2 项 media 1
#118 🔗 原推
2026-08-22 00:12:22
Garry Tan
Hate to say it but the luxury belief industrial complex is alive and well
#117 🔗 原推
2026-08-21 19:52:12
Garry Tan
RT @rahul: Yesterday. This guy is now free to kidnap another kid instead because the Flock cameras in the area were cut down. When someone cuts down a camera, the city pays Flock to install a new one. In the meantime, more criminals and child kidnappers go free. https://www.houstonchronicle.com/news/houston-texas/article/houston-child-kidnapping-flock-22396779.php https://twitter.com/rahul/status/2090799874350735870/photo/1
media 0 共 1 项
#116 🔗 原推
2026-08-21 19:51:20
Garry Tan
RT @WilliamBryk: Excited to announce that Jeff Pinner has joined Exa as CTO! Jeff was formerly CTO of Robinhood, a distinguished engineer at Cruise, CTO of Lyft, and way back a systems engineer at NASA. We have humongous engineering and research to do at larger-than-google-scale and we need to do it quickly. Pinner has scaled massive ML systems before, super excited to work together on this one.
media 0 共 1 项
#115 🔗 原推
2026-08-21 19:50:23
Garry Tan
RT @PhilipJohnston: I'm super excited to share that we have raised $250M at a $2.3B valuation to build a LOT more AI compute capacity in orbit 🚀 Thanks so much to our supporters, including Manhattan West, who led the round, and new investors, @NVIDIA, @Cisco_Invests, @CedarCapital, Goanna Capital, and @Standard_Cap, as well as continued investment from existing backers @Benchmark, @EQTVentures, @Soma_Capital, @NFX, @SevenSevenSix, and others. Next week, @Starcloud_ will move into a new 100,000 square ft facility that will enable us to ramp up our production rate to 100 satellites/week. This is just the beginning! @EzraFeilden, @AdiOltean🏃🏻‍➡️🏃🏻‍➡️🏃🏻‍➡️
media 0 共 1 项
#114 🔗 原推
2026-08-21 19:50:13
Garry Tan
This is a UC Berkeley CS professor 😳
#113 🔗 原推
2026-08-21 14:02:03
Garry Tan
Asian American erasure, example #107492
#112 🔗 原推
2026-08-21 13:45:50
Garry Tan
RT @kane: The chances of @CBS picking 5 random Cupertinians and none of them being Asian is 0.28%. https://twitter.com/kane/status/2090573073078706326/photo/1
media 0 共 1 项
#111 🔗 原推
2026-08-21 13:45:26
Garry Tan
Reminder: the datacenter discourse is a documented psyop and they’re winning
#110 🔗 原推
2026-08-21 12:32:17
Garry Tan
Legalize building societal infrastructure again
#109 🔗 原推
2026-08-21 12:29:59
Garry Tan
RT @Suhail: Just make great things. Find your moat later.
#108 🔗 原推
2026-08-21 12:29:16
Garry Tan
RT @paulg: The way to win at capitalism is to focus on what it actually does (make customers' lives better) rather than the socialist caricature of it (squeeze every penny out of them). The latter gets you 2x returns at best, whereas making new things for people can get you 10x.
#107 🔗 原推
2026-08-21 12:12:15
Garry Tan
RT @agupta: career ending answer from Markey imo
#106 🔗 原推
2026-08-21 12:11:15
Garry Tan
RT @paultoo: In a system designed to facilitate fraud, the only real crime is exposing the fraud
#105 🔗 原推
2026-08-20 21:58:47
Garry Tan
RT @AndyMasley: In all seriousness I think this represents both a disturbing society-level inability to process basic facts that involve numbers, a lurch toward degrowth on the left and right, and a failure of mainstream journalism to go beyond just nodding along to the "concerns" of everyday people without checking to see if they're the result of a game of telephone that leads back to nothing
#104 🔗 原推
2026-08-20 21:07:16
Garry Tan
RT @Trevor4SF: I grew up in the retirement community my dad ran + saw him search for missing seniors when an unwell spouse got ahold of their car keys. There are serious convos about public safety tech + Flock cameras, but let’s not forget the lives saved every day https://nowgeorgia.com/franklin-county-family-credits-flock-with-helping-find-missing-grandmother/
#103 🔗 原推
2026-08-20 19:40:43
Garry Tan
RT @sytses: My cancer case is n=1. I'm not holding it up as a model, because most of what I did is not specifically replicable or scaleable. But it is a proof of concept that fully personalized medicine, built around one patient's own biology, is viable in ways it simply wasn't a few years ago. I have lived a version of the future of cancer care that I think is worth aspiring to more broadly. Many of the tools that made this future possible for me are getting better, faster, cheaper, and more accessible every year. I think what we need next is the infrastructure to deploy these tools, and the community willing to build this infrastructure, one patient at a time. I want to help that future arrive faster. I wrote an article in The Cancer Letter https://cancerletter.com/guest-editorial/20260814_2/ for the physicians and researchers closest to the patients who could benefit most from this approach, who also believe in this future. If you are working towards building the systems and practices that will make personalized medicine operationally real, we want to collaborate with you too.
#102 🔗 原推
2026-08-20 19:40:39
Garry Tan
RT @dessaigne: On the sidelines of Paris Startup School, I had one of the most interesting conversations with @james406, the founder of @posthog. PostHog is going through what feels like a new pivot. They are known for being the best at product analytics. But now they are using AI to become what James calls a "doing company": finding problems, fixing them and opening PRs while you sleep. I sat down with James to understand how they got there.
🎬
视频
#101 🔗 原推
2026-08-20 19:40:23
Garry Tan
RT @garryslist: To read the curriculum SF will now require every 9th grader to take, a parent had to book an appointment, travel to a district office, and read it under supervision. What is the district hiding? A required course that recasts Asian American success as complicity in whiteness. https://twitter.com/garryslist/status/2090516393905283094/photo/1
media 0 共 1 项
#100 🔗 原推
2026-08-20 19:19:52
Garry Tan
YC is the YC for AI Researchers
#99 🔗 原推
2026-08-20 16:09:57
Garry Tan
Interesting violation of “you should dogfood your own product”
#98 🔗 原推
2026-08-20 16:04:41
Garry Tan
RT @rahul: Jacksonville Sheriff presents the data. They've used Flock cameras to solve 5 murders, 17 kidnappings, 6 forcible sex offenses, nearly 100 shootings, and 172 missing-persons cases.
#97 🔗 原推
2026-08-20 16:03:06
Garry Tan
YC is the YC for consumer hardware
#96 🔗 原推
2026-08-20 16:00:44
Garry Tan
YC is the YC for hard tech
#95 🔗 原推
2026-08-20 16:00:23
Garry Tan
Stop voting for this “wealth for me but not for thee” hypocrite
#94 🔗 原推
2026-08-20 15:59:45
Garry Tan
RT @parkerconrad: 3 weeks ago, @runlayer sued Rippling to block our launch of a competing MCP Gateway. They incinerated $ millions of their VC funding, then dropped the whole thing like a hot potato, for nothing (see our filing, below). So today, we’re dropping our new MCP Gateway. It’s better. 1/
media 0 共 1 项
#93 🔗 原推
2026-08-20 15:58:52
Garry Tan
Great to host the eminent @johnmaeda at YC Design is more important than ever in the time of AGI. When you can build anything knowing what to build is now the primary key to everything. https://twitter.com/garrytan/status/2090450625435971789/photo/1
media 0 共 1 项
#92 🔗 原推
2026-08-20 14:47:22
Garry Tan
RT @mateosfo: Berkeley's NIMBYs have fought to block every one of these buildings using the argument that students wouldn't be living in them and that the greedy capitalist investors would just keep them empty. You know, for profits.
#91 🔗 原推
2026-08-20 12:53:16
Garry Tan
RT @LoganDobson: *grasping at straws to make data centers look bad* "what if they... pay TOO MUCH in taxes??" https://twitter.com/LoganDobson/status/2090150148261175371/photo/1
media 0 共 1 项
#90 🔗 原推
2026-08-20 12:51:58
Garry Tan
RT @ApoStructura: “If they had a cure for cancer they wouldn’t sell it” Yes they would, because it would make them rich https://twitter.com/ApoStructura/status/2090085382914195844/photo/1
media 0 共 1 项
#89 🔗 原推
2026-08-20 12:38:04
Garry Tan
RT @billybinion: AI helped create a cancer vaccine. Data centers are lowering people’s property taxes, revitalizing dying towns by flushing them with revenue, and…helping power the infrastructure to cure cancer. Every data center hater should explain why any costs outweigh those benefits.
#88 🔗 原推
2026-08-20 12:37:49
Garry Tan
RT @raphaelschaad: Cooking up ways to teach YC batches the nature of Product Quality. iPod, Google Search, Instagram — these are singular products that touched 1B+ people, and their creators mastered the craft but critically: Knew what to build, and what to delete. https://twitter.com/raphaelschaad/status/2090310719829147793/photo/1
media 0 共 1 项
#87 🔗 原推
2026-08-20 05:52:40
Garry Tan
RT @agupta: Moderna is a great example of what happens when people are bold enough to ignore the haters and invest in technology platforms. In 2019 this company was widely assumed to die. But in COVID they likely saved several tens of millions of lives, bought themselves a financial lifeline to go after cancer, and now they’re going to save many million more. Incredible (Cambridge based!) company
#86 🔗 原推
2026-08-20 05:28:03
Garry Tan
RT @DKThomp: Really cool at the top of the WSJ homepage today: "An experimental mRNA-based vaccine succeeded in preventing cancer from coming back or spreading in a study of high-risk melanoma patients, Moderna and its partner Merck said Wednesday" Last year, an mRNA cancer vaccine was also shown to stop the recurrence of pancreatic cancer after surgery. It feels like we're just starting to pry open the heavy door of a new field: personalized mRNA cancer therapies that teach the body to fight cancers after initial treatment
media 0 共 1 项
#85 🔗 原推
2026-08-20 05:18:26
Garry Tan
RT @fin465: talked to a YC founder today who pivoted recently but is already at $25k/month, with one GTM trick (super sneaky) all he's doing is scraping linkedin for the phrase "anyone got recommendations for" + emailing the posters within 2 hours and yes, i know what you're thinking "there's no way there's enough posts like this" there's not, BUT it pulls up similar posts, and he scrapes the people who LIKED those posts here's why it works: someone who types "anyone got recommendations for a CRM" on linkedin just publicly declared their buying intent with a timestamp on it that's not a lead you find in a database that's a lead that found you first here's the exact phrases worth searching every morning: "anyone got recommendations for [tool]" "alternative to [competitor]" "looking for a good [software]" "got burned by [company], where do i go now" "has anyone used [company], are they worth it" every one of these is a public declaration of buying intent with a timestamp on it someone posted this on linkedin 3 hours ago: "we just outgrew our CRM. need something that can scale with us. open to recommendations" saw it at 9am pulled the poster's profile found the verified email sent this at 10am: "hey [name], saw your post about outgrowing your CRM. we help teams in that exact transition, last client we moved over in 2 weeks and cut their admin time by 40%. free 15 min call?" reply at 10:14am call at 2pm signed by 4pm 7 hours from public linkedin post to signed deal the playbook: step 1: identify the keywords your buyers use when they're in pain ("outgrowing," "replacing," "alternatives to," "burned by," "need recommendations for") step 2: search linkedin for those exact phrases step 3: filter for recency. 3 hours old beats 3 days old by a lot step 4: also scrape the people who LIKED similar posts, they have the same problem and nobody is emailing them step 5: enrich the poster, verified email, send 2 lines referencing exactly what they wrote step 6: sequence the reply within 30 minutes of a positive response the reason this works: the declaration IS the qualifier you don't need to discover their pain they published it the sales conversation is already 90% done before you say a word and you can use Origami AI which searches linkedin by keyword in real time same prompt you'd type into google returns people actively discussing the problem your product solves with verified contact info attached someone who just posted "outgrowing our CRM" is a better lead than anyone who fits a firmographic profile because intent beats fit every time everyone is fighting over the same 210M contacts in apollo the buyers typing their problems out loud on linkedin are completely uncontested go find one before your coffee gets cold
#84 🔗 原推
2026-08-20 05:18:16
Garry Tan
RT @kimmonismus: This is freaking huge: For the first time, a AI-assisted personalized mRNA cancer treatment has succeeded in a Phase 3 trial. Moderna and Merck sequence each patient’s tumor and compare it with their healthy DNA. AI then helps identify which of the tumor’s mutations are most likely to trigger an immune response. From those targets, Moderna produces an individual mRNA treatment encoding up to 34 neoantigens, The earlier Phase 2 trial followed patients with surgically removed high-risk melanoma for five years: - 68.8% remained cancer-free, compared with 49.1% on Keytruda alone - 49% lower risk of recurrence or death - 59% lower risk of distant metastasis or death Now the much larger Phase 3 trial involving 1,137 patients has also succeeded. Absolutely incredible. Dario was right: cancer will be cured in just a few years!
media 0 共 2 项 media 1
#83 🔗 原推
2026-08-20 04:29:05
Garry Tan
RT @wuweiweiwu: 2013: garry tan builds bookface, a literal book of faces for YC founders 2026: 4m lines running applications, interviews, startup school, cofounder matching, work at a startup, launch yc, and 350+ internal AI tools on one postgres over 2m of those lines showed up since march. they also got the whole thing onto the latest rails and react, first time since 2013. YC still isn't a software company. it just happens to run on 4m lines of its own, and that's the direction every company is walking in, whether or not they sell any. https://x.com/snowmaker/status/2088122446880870435?s=20
#82 🔗 原推
2026-08-20 04:10:29
Garry Tan
RT @AGHamilton29: The anti-data center stuff is insane and clear evidence that we have entered a post-truth era where many people can be convinced, often by malicious actors, to abandon reality and allow conspiracies/falsehoods to drive policy preferences.
#81 🔗 原推
2026-08-20 03:31:16
Garry Tan
RT @RichardHanania: New argument against data centers: They make you too rich and make your public amenities too nice. https://twitter.com/RichardHanania/status/2090151149777326175/photo/1
media 0 共 2 项 media 1
#80 🔗 原推
2026-08-20 03:31:11
Garry Tan
RT @ApoStructura: AI dadacenters just don’t use that much water, there’s a panic over almost nothing. Here is a short animation to put AI water use in context: https://twitter.com/ApoStructura/status/2090054044211376275/video/1
🎬
视频
#79 🔗 原推
2026-08-20 03:17:25
Garry Tan
RT @chrisbarber: I asked people: "which AI data startups are most interesting?" The focus was on startups that sell data to labs, excluding the massive ones (Mercor, Surge, Handshake, Turing): 22 votes: Mechanize 12: Proximal 11: Fleet 9: AfterQuery 8: Datacurve 5 each: Bespoke Labs, Preference Model 4: Abundant 3 each: Andon Labs, Latch, microagi, Prime Intellect, Snorkel, XDOF 2 each: Build, Datology, Design Arena, dmodel, Good Start Labs, hillclimb, HUD, Inheritance, Mecka, Sieve, Taste Labs, Trajectory, Vmax 1 each: 37 companies, info in image and below Disclosure: I'm a small angel in Abundant, Fleet, Inheritance, Latch, Trajectory, and Vmax. I didn't vote.
media 0 共 1 项
#78 🔗 原推
2026-08-20 03:16:06
Garry Tan
RT @kimmaicutler: The tragedy here is that UC Berkeley screwed her over not once, but twice. She spent several decades cultivating a high school mathematics program based on Eastern European math circles and university administrators destroyed it with an overzealous interpretation of a new state law for anti-predator fingerprinting requirements so world-class mathematicians could no longer easily give lectures to advanced math HS students. https://www.dailycal.org/news/campus/genius-producing-math-program-lost-to-uc-berkeley-fingerprinting-requirements/article_e909f495-7bf7-4662-ab15-5cda7bbcd773.html?s=09
#77 🔗 原推
2026-08-20 02:06:32
Garry Tan
RT @agupta: YC invented the SAFE a decade ago, simplifying the process of raising seed money for founders across the world. Today, we’re releasing a SAFE tool to make it even easier to raise your seed round on SAFEs. And it’s free!
#76 🔗 原推
2026-08-20 01:13:12
Garry Tan
RT @Rafa_Mangual: The anti-ALPR/Flock push is basically a replay of the anti-Gunshot Detection/ShotSpotter debate: Anti-police leftists (with the enthusiastic help of a handful of left-libertarians that almost always agree with them) are pushing this b/c they think policing is racist. DSA slop. https://twitter.com/Rafa_Mangual/status/2090177100963549187/photo/1
media 0 共 3 项 media 1 media 2
#75 🔗 原推
2026-08-20 01:06:44
Garry Tan
RT @Alfred_Lin: All leaders need to do frontline IC work again, if they and their teams want to make it through the AI transition. On a similar note, I've heard of several GTM execs who fully left their jobs, just so they can dedicate 6 months to get ahead with all the new tooling and capabilities. Things are moving that fast.
#74 🔗 原推
2026-08-20 00:49:20
Garry Tan
RT @bosmeny: Jesus. Pocket (W26) just hit $100M annualized revenue run rate ...and they were in YC 6 months ago 🤯 Just wildly impressed by how well this team is executing
#73 🔗 原推
2026-08-20 00:48:14
Garry Tan
RT @snowmaker: YC built software that all our founders have used to send SAFEs for the last few years. Today, we're opening it up to everyone and giving it away for free.
#72 🔗 原推
2026-08-20 00:30:08
Garry Tan
Who is the Tampa Bay Wave marketing department because this is amazing
#71 🔗 原推
2026-08-19 18:47:54
Garry Tan
She should be a private citizen because she is no public servant and has never been https://x.com/garrytan/status/2056554821789073613
#70 🔗 原推
2026-08-19 15:48:50
Garry Tan
Connie Chan hypocrisy, just par for the course Kill housing, kill public safety, make SF unsafe for Asian American elders, kill our kid's educations ... and expect a ride to DC on the political machine train You should not vote for Connie Chan
#69 🔗 原推
2026-08-19 15:48:10
Garry Tan
RT @snowmaker: Take a minute to watch this video; it's a cool one. This is not some AI generated video; this is the actual robot the Grip founders built, sorting real trash, at 1x speed.
#68 🔗 原推
2026-08-19 15:30:14
Garry Tan
RT @kane: Same lady who made grand theft up to $950 a misdemeanor in California, making it impossible to detain shoplifters and burglars:
#67 🔗 原推
2026-08-19 15:30:01
Garry Tan
RT @zackkanter: It’s pretty cool to have grown up reading about colossally stupid movements against things like nuclear power and now to be able to see them happen in real time as an adult.
#66 🔗 原推
2026-08-19 15:29:50
Garry Tan
"Irresponsible AI Data Centers" Who writes this stuff?
#65 🔗 原推
2026-08-19 15:19:46
Garry Tan
RT @ycombinator: Michael Kratsios (@mkratsios47) has seen the AI boom from both sides: as COO of Scale AI and inside the White House. Today, as the director of the White House Office of Science and Technology Policy, he helps shape America’s national strategy on AI, science, and emerging technology. At Startup School 2026, he sat down with YC's @lutherlowe, to talk about how Washington makes technology policy, why the White House supports open source AI, and why little tech needs a seat at the table. 02:38 — From Tech to the White House 04:39 — How Technology Policy Actually Gets Made 05:51 — The White House on Open Source AI 08:28 — How Washington Sees AI Differently 09:53 — Regulating a Technology That Changes Every Six Months 11:37 — Which AI Risks Are Overblown? 13:03 — Giving Little Tech a Seat at the Table 17:31 — Regulation Without Creating Incumbent Moats 18:01 — Born Free vs. Born in Captivity Technologies 20:46 — What Working in the White House Is Actually Like 25:56 — A New Golden Age of American Science 31:32 — Quantum, Congress, IP, and What Comes Next 37:54 — Why Technologists Should Consider Public Service
🎬
视频
#64 🔗 原推
2026-08-19 15:17:54
Garry Tan
When ideology creeps into reporting it actually has an impact: it moves the Overton window and votes If you wonder how it is a school district could willingly choose to stop teaching Algebra to middle school kids over “equity,” this is exactly how (yes this happened in SF)
#63 🔗 原推
2026-08-19 14:56:05
Garry Tan
RT @amasad: Agents made software cheaper but made coding expensive. Today, together with @OpenAI, we’re changing this:
#62 🔗 原推
2026-08-19 14:53:59
Garry Tan
RT @kane: Very impressed by Congressman @RoKhanna's elementary school children, who must have worked very hard to amass their $150M fortune. Very lucky for Ro too, who will be spared many of the taxes he's proposing. Have your LLM ask https://rokhanna.money/ anything you're curious about.
media 0 共 1 项
#61 🔗 原推
2026-08-19 14:52:50
Garry Tan
RT @VirtualElena: in my mid-20's, a time when about half of my friends hated their jobs (a very normal phenomenon if you're in your mid-20s), i noticed a group of people my age who looked extremely happy and actualized at work. and all of these people somehow seemed to work at stripe. i'll call out @tamarawinter and @orbuch specifically as some of the first people who made me realize that it was possible to be 25 years old and do *singular* work every day, and not just whatever garden-variety boilerplate the company demands. this podcast with @gaybrick and @DavidGeorge83 helps explain why this was, and still is, the case. will says stripe increasingly is a platform for founders internally, just as it's always been a platform for founders externally. he wants each employee to feel like an “auteur”: an engineer, product thinker, and creative person with enough context and authority to make something real. we're now seeing a new generation of this pattern, with stripe's support of stuff like tempo (a team that might have one of the highest concentrations of founder-mindsetted people im aware of? cc @gakonst @dwr @varunsrin @liamihorne etc) the embrace of teams like privy and bridge, and other incubated initiatives like minions. stripe is trying to make employees more founder-like while preserving a shared standard of quality (one of my favorite details on this point: stripe creates simulated customer accounts so teams can experience realistic UIs for disputes and refunds before exposing products to real users). this is something that's actually very tough to do well, so it's very cool to hear will talk about it on this pod.
🎬
视频
#60 🔗 原推
2026-08-19 04:37:46
Garry Tan
RT @Noahpinion: This was such an own goal for our university system
#59 🔗 原推
2026-08-19 03:08:17
Garry Tan
RT @soohoonchoi: we're &lt;20 engs, we love hard problems, we're obsessed with reptiles and bugs, we want to solve software.
#58 🔗 原推
2026-08-19 03:08:09
Garry Tan
If you want SF rent to be $10K/mo for a 1BR (you're a NIMBY landlord and you hate newcomers and young people) only then would you vote for Connie Chan The Dem machine smokescreen to try to force Chan into Congress is shameless and must be repudiated
中文: 如果你想让旧金山的租金是1比1升(你是NIMBY房东，讨厌新来者和年轻人) 只有到那时你才会投票给康妮·陈 试图强迫陈进入国会的“民主党机器”防烟霜毫无耻，必须被否认
#57 🔗 原推
2026-08-19 00:18:57
Garry Tan
Many such cases
中文: 许多此类案例
#56 🔗 原推
2026-08-18 23:01:51
Garry Tan
RT @incitafusio: 95% of Mission high school students did not meet CA math standards, yet they are taking top spots at all the UCs ie Berkeley, UCLA, UCSD, etc., while Lowell students (71% met math standards) are only competitive w/ Mission students at UCSB, Merced, Riverside or Santa Cruz. UCs punish Lowell students for working hard bc of their race, while setting up Mission kids up for failure & depression. Race-based admissions fails all kids & was deemed unconstitutional by Scotus - it must end. So where do the students actually end up going... 🧵
中文: RT @incitafusio:95%的Mission高中生未达到加州大学数学标准，但他们在加州大学伯克利分校、加州大学洛杉矶分校、加州大学圣地亚哥分校等地均名出一格，而洛厄尔大学的学生(71%达到数学标准)仅与加州大学伯克利分校、默塞德大学、里弗赛德大学或圣克鲁斯分校的Mission学生竞争激烈。加州大学会惩罚洛厄尔大学的学生，因为他们在种族方面努力工作，同时为“使命”儿童设立失败与抑郁的场所。以种族为基础的录取未能完全通过，而苏格兰人认为其违宪，必须终止。 那么，学生们最终会去哪里......🧵
media 0 共 1 项
#55 🔗 原推
2026-08-18 19:59:53
Garry Tan
RT @emollick: My father just got a new MacBook after years with no laptop, and he did not know how to set it up He just told me that he took my advice to install Codex and just ask it to take over and solve his setup problems &amp; install what he needed. Family IT support will never be the same.
中文: RT @emollick:我父亲多年后刚刚用笔记本电脑就有了一台新MacBook，而且他也不知道怎么准备 他只是告诉我，他采纳了我关于安装Codex的建议，然后要求它接管并解决他的安装问题 &amp; 并安装他需要的内容。家庭IT支持将永远不会是相同的。
#54 🔗 原推
2026-08-18 19:59:34
Garry Tan
RT @GeniusGTX: Garry Tan is the CEO of Y Combinator, the most powerful accelerator for early-stage startup founders on Earth. Here are his 8 rules for new founders on starting, building &amp; executing a startup in the age of AI: 1) Agency and taste are super important now to stand out https://twitter.com/GeniusGTX/status/2089684243895156741/video/1
中文: RT @GeniusGTX:Garry Tan 是 Y Combinator 的首席执行官，Y Combinator 是地球上早期初创企业创始人最强大的加速器。 以下是他在人工智能时代为新创始人制定的8项规则:在人工智能时代创办、建立和创业; 1) 现在对于脱颖而出，代理和品味非常重要
🎬
视频
#53 🔗 原推
2026-08-18 19:58:55
Garry Tan
RT @dakshgup: "coding is the bottleneck." coinbase engineering leader @chintanturakhia disagrees. below is my full conversation with chintan, probably the most AI-native engineering leader in the world, on how he runs his engineering team. https://twitter.com/dakshgup/status/2089768014988378436/video/1
中文: RT @dakshgup:“编码就是瓶颈。” coinbase工程负责人@chintanturakhia持不同意见。 以下是我与奇坦的完整对话，他可能是世界上最原生的工程技术领导者，他是如何管理自己的工程团队的。
🎬
视频
#52 🔗 原推
2026-08-18 19:48:57
Garry Tan
RT @dakshgup: say your prayers @thinkymachines https://twitter.com/dakshgup/status/2089559340277739943/photo/1
中文: RT @dakshgup:表达你的祈祷:@thinkymachines
media 0 共 1 项
#51 🔗 原推
2026-08-18 19:48:05
Garry Tan
RT @vivianmshen: start of a generational run by @AnjneyMidha - 75x+ return in 12 months and the Anthropic IPO this year will be a monster https://twitter.com/vivianmshen/status/2089766099336429954/photo/1
中文: RT @vivianmshen:由@AnjneyMidha运营的一代人开始——12个月内回归75倍以上 今年的《蚁马》IPO将是一场怪物
media 0 共 1 项
#50 🔗 原推
2026-08-18 19:46:38
Garry Tan
RT @mehul: It took us 9 years, 11 prototypes, and our life savings to learn that every great product has the same design process. Matic is now decisively the best home robot for families (I'm biased) A 500-word thread and video on the Universal Design Process behind every great product: https://twitter.com/mehul/status/2089784603737575470/video/1
中文: RT @mehulm:我们花了9年11个原型，毕生积蓄才发现每款出色的产品都采用了相同的设计流程。 如今，马蒂奇绝对是家庭最好的家用机器人(我有偏见) 关于每款伟大产品背后通用设计流程的500字视频:
🎬
视频
#49 🔗 原推
2026-08-18 19:08:16
Garry Tan
RT @saigurrapu18: Today, we’re announcing our $5.3M seed round, led by CIV. There is a rare window right now to redefine how some of the world’s most legacy software gets rebuilt across banks, airlines, governments, and large enterprises. For decades, mainframe modernization has been a massive human undertaking: multi-year programs, tens of millions of dollars, and large teams reconstructing, rebuilding, and verifying systems that nobody fully understands anymore. We started Hypercubic with a simple thesis: mainframe modernization should become a repeatable, software-driven process rather than a bespoke human undertaking. Over the past year, we’ve made several breakthroughs in an attempt to build the most capable mainframe modernization platform one that can understand, rebuild, and verify critical systems at high velocity. We’re already seeing with customers how dramatically parts of the modernization lifecycle can be compressed. The industry is not used to the speed we now know is possible. Today, this work remains fundamentally a human labor problem. We think it can become a computational problem instead. @aayush_naik and I are building a small, high-caliber team to go after this. If this problem excites you, please reach out, we’re hiring exceptional people across the team!
中文: RT @saigurrapu18:今天，我们宣布以CIV为首轮530万美元的种子轮融资。 目前，有一个难得的窗口期可以重新定义全球最具传统性的软件是如何在银行、航空公司、政府和大型企业中重建的。 几十年来，大型机现代化一直是一项庞大的人类事业:多年计划、数千万美元，以及庞大的团队，重建、重建和验证那些无人完全理解的系统。 我们从一个简单的论点开始了Hypercubic:大型机现代化应成为一种可重复的、以软件为导向的过程，而非定制的人类事业。 过去一年，我们取得了多项突破，试图构建一个能够快速理解、重建和验证关键系统的最有能力的大型机现代化平台。 我们已经向客户看到了现代化生命周期中能够大幅压缩部分的程度。行业并不习惯我们现在知道的可能速度。 如今，这项工作从根本上仍然是一个人类劳工问题。我们认为它反而会成为一个计算问题。 @aayush_naik 和我正在组建一支规模较小、高素质的团队来关注此地。如果这个问题让你感到兴奋，请联系我们，我们团队中招聘非常出色的人!
media 0 共 1 项
#48 🔗 原推
2026-08-18 19:08:05
Garry Tan
RT @hanghuang_: We didn't get into YC until we stopped trying to get in. After the 4th rejection for InsForge, Tony and I decided to generationally lock in and focus completely on building product and acquiring users. We looked at how the best devtool companies (Supabase, Resend, WorkOS) got developers to care. We realized we should run our own Launch Week! After 30 days of insane heads down building, we did it. InsForge Launch Week 1 went completely viral. We hit #1 on Product Hunt, #1 on GitHub Trending, 1M+ views on X and LinkedIn. We generated enough motion that YC couldn’t ignore us. They ended up reaching out to us instead of us applying. That's what got us into YC.
media 0 共 1 项
#47 🔗 原推
2026-08-18 15:57:39
Garry Tan
RT @kane: Progressive policy took @UCBerkeley from Manhattan Project to “can’t do algebra” in two generations.
#46 🔗 原推
2026-08-18 15:46:39
Garry Tan
RT @ibab: The River API was tested in this blog post and outperformed Tinker on reinforcement learning runs with identical training code. We spent a lot of effort to get details like routing replay right so you get the best possible results with the API.
#45 🔗 原推
2026-08-18 15:46:29
Garry Tan
RT @credealjunkie: Sometimes I wonder why random people with no connection to SF hate it so much. Other times I know exactly why. Imagine a guy pledging to invest $100 million into a forgotten retail strip - in the neighborhood he grew up in - through a privately-funded nonprofit and being accused of a “hostile takeover” by arguably the most influential local politician. Now imagine the guy wants to do it in literally any city besides San Francisco. They’d throw him a parade - and rightly so. You hear a lot that San Francisco hasn’t lived up to its potential because of anti-growth politics. I disagree. It’s a miracle San Francisco isn’t worse off given the entrenched anti-growth politics of the past 50 years, with periodic respites like the present moment.
media 0 共 1 项
#44 🔗 原推
2026-08-18 15:46:19
Garry Tan
RT @AlejandroRomaan: Super late to using Gbrain, i've been using just vanilla obsidian for a long time. Added Gbrain to my harness to sit in between obsidian, and wow it's been a game changer real cool concept. Currently sits between my main worker and the vault. I've started preaching about it to anyone who will hear me out lol
🎬
视频
#43 🔗 原推
2026-08-18 15:44:12
Garry Tan
GBrain actually works with any AI Harness (Grok Bot, Claude Code, Codex, Hermes Agent, OpenClaw, OpenCode, all work) and any Postgres with pgvector hosted service This is what it means to truly own your own AI agent's memory and skills.
#42 🔗 原推
2026-08-18 15:19:36
Garry Tan
RT @WilliamBryk: Last week, Exa took over SF's Pioneer Building, previously home to Stripe, OpenAI, and most recently xAI. Elon assembled the Colossus cluster here (his mattress was still lying there during our tour). Alec Radford trained GPT-3 here. The Collison brothers built the web's financial backbone here. They were all infra companies: Stripe -> Financial infra OpenAI -> Intelligence infra xAI -> Compute infra Each entered Pioneer as a small startup with enormous ambitions and left as a foundational layer for the world. Exa -> Information infra We entered Pioneer as web search for agents, but by the time we leave we'll have organized all the world's data into a foundational information layer. This will require things like deploying satellites, new misinformation detection algorithms, and embedding retrieval on an outrageous scale. Whatever is necessary to keep billions of people and trillions of AIs as informed as possible. A trusted information ecosystem is necessary to safely navigate this chaotic AI era, an era that was set in motion by our previous occupants.
🎬
视频
#41 🔗 原推
2026-08-17 19:46:56
Garry Tan
This is my open source to help you create your own Personal AGI as mentioned at my Startup School talk earlier this year https://x.com/ycombinator/status/2085443781797785828
#40 🔗 原推
2026-08-17 19:44:44
Garry Tan
What do you get? A private github repo with 70 of my proven skills and the beginnings of your Karpathy-style knowledge wiki. Read the full docs in the README. All of this is MIT-licensed open source and free. https://github.com/garrytan/gbrain https://twitter.com/garrytan/status/2089425134339961173/photo/1
media 0 共 1 项
#39 🔗 原推
2026-08-17 18:52:26
Garry Tan
It's free to try and works with your existing Claude Code or Codex subscription right now. Just make a new directory and start Codex or CC (on command line OR on Desktop, slightly better on Desktop) Just paste this image into your CC/Codex and it'll set up your agent repo asap https://twitter.com/garrytan/status/2089424620764168485/photo/1
media 0 共 1 项
#38 🔗 原推
2026-08-17 18:50:23
Garry Tan
GBrain now supports personalized agent generation/onboarding for Codex and Claude Code It'll generate an agent AI-style SOUL.md and install 70 of my personal agent skills to get you started. In 12 questions, you get setup with an agent just as smart as my personal OpenClaw https://twitter.com/garrytan/status/2089424170761564636/photo/1
media 0 共 1 项
#37 🔗 原推
2026-08-17 18:48:36
Garry Tan
RT @CharlesFLehman: People keep asking: do Flock cameras really help fight crime? So I built a database of over 700 times Flock license plate readers have helped solve a crime, apprehend an offender, or recover a stolen car or abducted child. The full database is here: https://www.flockstopscrime.com/ https://twitter.com/CharlesFLehman/status/2089294917671755957/photo/1
media 0 共 1 项
#36 🔗 原推
2026-08-17 17:27:36
Garry Tan
Like how much stupid stuff is the voting populace going to ingest until they realize “BILLIONAIRE” is just a word they use to make you turn off your brain and vote for the craziest brain worms politicians?
#35 🔗 原推
2026-08-17 16:27:52
Garry Tan
The people who created the SF Doom Loop (who we voted out) are still posting And it’s as deranged as ever
#34 🔗 原推
2026-08-17 16:26:45
Garry Tan
RT @agupta: Completely nuts that PBM revenue is comparable to pharma revenue in the status quo. Pharma companies take on enormous scientific and execution risk to discover and advance medicines. PBMs are unnecessary middlemen that do neither of those things.
#33 🔗 原推
2026-08-17 16:25:26
Garry Tan
RT @CharlesFLehman: Also in Flock news: new working paper shows Flock reduces auto theft, boosts clearance rates. @smourtgos and @ian_t_adams merged data on 216 Flock deployments with PD crime counts. They find an 11 percent reduction in motor vehicle thefts, and a 16 percent increase in motor vehicle clearances. Among recorded recoveries, cars were recovered about 0.28 days faster. Full paper here: https://www.crimrxiv.com/pub/zleg04q3/release/1
media 0 共 1 项
#32 🔗 原推
2026-08-17 16:21:08
Garry Tan
Your personal AI or your company brain needs a clean harness and this is the one our team built and uses every day Free and open source
#31 🔗 原推
2026-08-01 00:46:58
Garry Tan
Socialists want state controlled bread lines You know what creates abundance? Great functioning markets.
#30 🔗 原推
2026-07-19 04:35:49
Garry Tan
https://twitter.com/garrytan/status/2072846822566133768/photo/1
media 0 共 1 项
#29 🔗 原推
2026-07-03 00:56:08
Garry Tan
https://twitter.com/garrytan/status/2072461457195950446/photo/1
media 0 共 1 项
#28 🔗 原推
2026-07-01 23:24:50
Garry Tan
https://twitter.com/garrytan/status/2071817410303393840/photo/1
media 0 共 1 项
#27 🔗 原推
2026-06-30 04:45:37
Garry Tan
Bernie Sanders introduced a bill to seize 50% of any AI startup that crosses $200M in revenue. The same anti-prosperity bloc spent the year trying to ban startup acquisitions, blocking the only exit 85% of founders ever get. This is a war on building startups in America. https://twitter.com/garrytan/status/2067636692794875923/photo/1
中文: 伯尼·桑德斯提出了一项法案，要求收购所有收入超过2亿美元的人工智能初创企业50%的股份。 同一个反繁荣阵营在过去一年里试图禁止初创企业的收购，阻止了85%的创始人获得的唯一退出。 这是一场关于在美国建立初创企业的战争。
media 0 共 1 项
#26 🔗 原推
2026-06-18 15:52:57
Garry Tan
The housing NIMBYs and the datacenter NIMBYs are all the same people Hate tech, hate prosperity, subsidize demand, block all supply, be surprised when everyone ends up poor with no jobs
中文: NIMBY 和数据中心 NIMBY 的所有人员都相同 憎恨科技，憎恨繁荣，补贴需求，屏蔽所有供应，当每个人都穷到没有工作时，会感到惊讶
#25 🔗 原推
2026-06-14 19:08:23
Garry Tan
Rent control subsidizes demand We need to dramatically increase supply Anyone who tells you otherwise just wants your rent to go up https://www.brookings.edu/articles/what-does-economic-evidence-tell-us-about-the-effects-of-rent-control/ https://twitter.com/garrytan/status/2066186923924975902/photo/1
中文: 租金管制补贴需求 我们需要大幅增加供应 任何告诉你，否则就希望房租上涨的人
media 0 共 1 项
#24 🔗 原推
2026-06-14 15:52:05
Garry Tan
Skill issues at big company means small new ones can eat their lunch
中文: 大公司的技能问题意味着新人可以吃午餐
#23 🔗 原推
2026-06-05 14:55:06
Garry Tan
NIMBYs frequently virtue signal by attacking developers but it’s a smoke screen and we shouldn’t listen. Markets are good. Housing is the defining reason why cost of living is high. If you want better lives for people, you build housing, so housing costs go down.
中文: NIMBY 经常通过攻击开发者来发出美德信号，但它是一个烟幕，我们不应监听。市场是好的。 住房是生活成本高企的决定性原因。如果你想为人们提供更好的生活，就可以建造住房，因此住房成本会降低。
#22 🔗 原推
2026-06-05 03:35:17
Garry Tan
The nuclear reactor at YC Demo Day will be a sight to see, can't wait for you to see it
中文: YC演示日的核反应堆将一见钟情，迫不及待地想看到
#21 🔗 原推
2026-06-04 18:19:55
Garry Tan
https://x.com/i/article/2061176923531984896
#20 🔗 原推
2026-06-01 14:26:49
Garry Tan
Founders must stop trying to building 2010-era businesses with 2026-era technology. Don't try to rebuild Foursquare or Yelp. Don't try to recreate Basecamp by 37 Signals with $10/mo SaaS pricing. Don't underprice! If it works it's worth a lot more. Don't be tempted to…
中文: 创始人必须停止试图利用2026年技术在2010年建立业务。 不要试图重建Foursquare或Yelp。 不要尝试通过 37 个 Signals 以 10 美元/月的 SaaS 定价来重新创建 Basecamp。 不要低估价格!如果奏效，它的价值就大得多。 不要被诱惑......
#19 🔗 原推
2026-05-27 06:26:41
Garry Tan
The companies I love working with in office hours are the ones where the founder has a specific, weird, earned insight that nobody else has. Not "AI for X." A genuine edge that came from living inside a problem. The ones that are dying almost always have the same pattern:…
中文: 我喜欢在办公时间工作的公司，是创始人拥有别人无人了解的、特别古怪的、已获得洞察的公司。不是“AI for X”。真正的优势源于生活在一个问题之中。 那些即将死去的人几乎总是有同样的模式:
#18 🔗 原推
2026-05-24 07:57:32
Garry Tan
OK I guess we're going pretty fast improving GBrain day to day right now https://twitter.com/garrytan/status/2056588601216168168/photo/1
中文: 好吧，我想我们现在每天都在快速改善GBrain
media 0 共 1 项
#17 🔗 原推
2026-05-19 04:11:46
Garry Tan
The biggest alpha leak of 2026 is that you can tokenmax $10k/mo with OpenClaw/Hermes + GBrain and get the AI that everyone will have in 2028 for $100/mo, but you can get it now, and that is the biggest single unlock you can have vs your competition
中文: 2026年最大的一次阿尔法泄密事件是，你可以通过OpenClaw/Hermes + GBrain实现1000美元/月的代币，并以100美元/月的价格获得2028年所有人拥有的人工智能，但你现在也能获得，这是你与竞争对手相比所能拥有的最大单次解锁
#16 🔗 原推
2026-05-15 15:31:32
Garry Tan
Hard leftists and DSA-types will fight AI tooth and nail because AI is the universal solvent for socialism’s three prerequisites. AI will reduce scarcity (abundance), eliminate victimhood (agency), and eliminate the need for intermediaries (disintermediation).
中文: 强硬的左翼分子和DSA型将对抗人工智能，因为人工智能是社会主义三大先决条件的通用溶剂。 人工智能将减少稀缺性(过剩)，消除受害者(机构)，并消除对中介机构(非中介机构)的需求。
#15 🔗 原推
2026-05-15 12:36:17
Garry Tan
The politicians think this is about blocking big tech when actually they are going to strangle little tech while also destroying jobs and American competitiveness at the global level It’s all so knee jerk and shortsighted https://www.texastribune.org/2026/05/12/texas-hill-county-approves-data-center-construction-pause-ai/
中文: 政客们认为，这是为了阻止大型科技行业，而实际上他们却要扼杀一些科技行业，同时在全球层面摧毁就业和美国的竞争力 全是膝盖抽搐和短视
#14 🔗 原推
2026-05-14 12:39:44
Garry Tan
Sanders and AOC introduced a bill to pause ALL AI data center construction. 300+ local bills filed. Half of planned 2026 data centers facing delays or cancellation. Each one brings billions to local economies. The people who say they want American jobs are trying to block the… https://twitter.com/garrytan/status/2054904105018237125/photo/1
中文: 桑德斯和AOC提出了一项暂停所有人工智能数据中心建设的法案。提交了300多项地方性法案。计划中的2026年数据中心中有一半面临延误或取消。每一项都为地方经济带来数十亿美元。 那些表示希望美国就业的人正试图阻挠......
media 0 共 1 项
#13 🔗 原推
2026-05-14 12:38:11
Garry Tan
https://x.com/i/article/2052898104039657472
#12 🔗 原推
2026-05-09 14:58:40
Garry Tan
Saikat Chakrabarti made his fortune in tech. Now he wants to ban the data centers that power AI in San Francisco, the most important growth industry in the city and the engine of downtown's recovery. That's not regulation. It's a kill switch. It kills SF jobs, guts the tax base,… https://twitter.com/garrytan/status/2049284944737009744/video/1
中文: 赛卡特·查克拉巴蒂在科技领域赚了钱。现在，他希望禁止为旧金山人工智能提供动力的数据中心，这是该市最重要的增长产业，也是市中心复苏的引擎。 这不是监管。这是一个致命开关。它扼杀了旧金山的工作岗位，扼杀了税收基础......
media 0 共 2 项
🎬
视频
#11 🔗 原推
2026-04-29 00:29:39
Garry Tan
Here's YC's official advice about being truthful and precise about what is pilot, bookings, revenue and recurring revenue. Founders, particularly first time founders, need to sear this into their brains. Don't mistake one tier for another. Be precise, and always be truthful. https://twitter.com/garrytan/status/2048017824895909901/photo/1
中文: 以下是YC关于如何真实、准确地了解试点、预订、收入和经常性收入的官方建议。 创始人，尤其是首次创立者，需要将这一点融入他们的大脑。不要把一个层级误认为另一个。要准确，并且始终保持真实。
media 0 共 1 项
#10 🔗 原推
2026-04-25 12:34:34
Garry Tan
https://x.com/i/article/2046866228703363072
#9 🔗 原推
2026-04-22 09:01:16
Garry Tan
I came back to code because AI made it possible for me to build at a level I couldn't before. I'm not coding despite being CEO of YC. I'm coding because this is the most important technological shift since the internet and I'd be an idiot to experience it from the bleachers.…
中文: 我回归编程是因为人工智能让我能够达到前所未有的水平。 尽管担任YC的首席执行官，我却没有编程。我之所以编程，是因为这是自互联网以来最重要的技术转变，从看台上体验它，我简直是个白痴。......
#8 🔗 原推
2026-04-20 14:03:34
Garry Tan
https://x.com/i/article/2045399189606273024
#7 🔗 原推
2026-04-19 09:36:10
Garry Tan
https://x.com/i/article/2017109679416233987
#6 🔗 原推
2026-04-15 18:14:34
Garry Tan
YC will invest in any YC company in stablecoins The new financial rails of the revolution will not be over ACH or wire
中文: YC将投资任何一家YC公司进行稳定币投资 革命的新金融轨道不会在ACH或电线上
#5 🔗 原推
2026-04-14 00:41:27
Garry Tan
If you want your OpenClaw or Hermes Agent to be able to have perfect total recall of all 10,000+ markdown files, GBrain is here to help. It's exactly my OpenClaw/Hermes Agent setup. MIT-licensed open source. Hope it helps you build your mini-AGI. https://github.com/garrytan/gbrain
中文: 如果希望您的 OpenClaw 或 Hermes Agent 能够完全完全召回所有 10，000 多个标记文件，GBrain 将为您提供帮助。 这正好是我的 OpenClaw/Hermes 代理设备。获得MIT许可的开源。希望它能帮助你构建你的迷你AGI。
#4 🔗 原推
2026-04-10 07:00:15
Garry Tan
I'm open sourcing it MIT license so we can all speed up and have our own personal mini-AGI. It's been amazing for me and I want you to have it. To install GBrain in your OpenClaw, just paste this image to your OpenClaw or paste this text: Set up gbrain… https://twitter.com/garrytan/status/2042369335419945338/photo/1
中文: 我正在开源MIT许可证，以便我们都能加快速度，并拥有自己的个人迷你AGI。这对我来说太棒了，我希望你能拥有它。 要在 OpenClaw 中安装 GBrain，只需将此图像粘贴到 OpenClaw 或粘贴以下文字: 设置 gbrain...
media 0 共 1 项
#3 🔗 原推
2026-04-09 22:29:29
Garry Tan
How I get my claw to be a durable AI agent I never have to instruct twice Paste this into your OpenClaw's AGENTS.md or send it as a message: You are not allowed to do one-off work. If I ask you to do something and it's the kind of thing that will need to happen again, you…
中文: 我如何让自己的爪子变成耐用的人工智能代理，我永远不必两次指导 将其粘贴到您的OpenClaw的AGENTS.md中，或发送消息: 不允许做一次性工作。如果我要求你做某件事，而这种事情又需要再次发生，你......
#2 🔗 原推
2026-04-09 00:20:49
Garry Tan
https://twitter.com/garrytan/status/2033640490843701360/photo/1
media 0 共 1 项
#1 🔗 原推
2026-03-16 20:24:10