AK
Scaling Trajectories for Complex Tasks through Recursive Self-Rewrite paper: https://huggingface.co/papers/2610.02826 https://twitter.com/_akhaliq/status/2106960577071296651/photo/1
中文: 通过反复自我改写来扩展复杂任务的轨迹 论文:
media 0 共 1 项
#211 🔗 原推
2026-10-05 04:12:01
AK
Octrees as an Explicit 3D Language paper: https://huggingface.co/papers/2610.02388 https://twitter.com/_akhaliq/status/2106959958973759807/video/1
🎬
视频
#210 🔗 原推
2026-10-05 04:09:34
AK
EditHero A Benchmark for Long-Horizon Part-Level 3D Editing and Vibe Modeling paper: https://huggingface.co/papers/2610.02298 https://twitter.com/_akhaliq/status/2106959328385077426/video/1
中文: 编辑英雄 长地平线半层3D编辑与振动建模的基准 论文:
🎬
视频
#209 🔗 原推
2026-10-05 04:07:03
AK
RT @Aleph__Alpha: https://huggingface.co/Aleph-Alpha/kolibri-1
#208 🔗 原推
2026-10-04 00:15:46
AK
Adaptive Reward Routing Dynamic Multi-Reward Optimization for Joint Audio-Video Diffusion via Forward-Process RL paper: https://huggingface.co/papers/2609.37200 https://twitter.com/_akhaliq/status/2106080325386305893/photo/1
中文: 自适应奖励路由 通过前向过程RL实现关节音频视频扩散的动态多回报优化 论文:
media 0 共 1 项
#207 🔗 原推
2026-10-02 17:54:13
AK
EgoTools Towards Tool-Centric Reasoning in Real-World Egocentric Videos paper: https://huggingface.co/papers/2609.39378 https://twitter.com/_akhaliq/status/2106063986664075749/video/1
中文: EgoTools 面向现实中以工具为中心的推理 以Egocentric视频 论文:
🎬
视频
#206 🔗 原推
2026-10-02 16:49:17
AK
RT @HuggingPapers: AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks A 27B agent that turns more test-time rounds into better solutions, trained on verifiable ML and algorithmic tasks. Achieves 81.8 on MLE-bench Lite, 70.7 on Frontier-CS, and transfers to deep research. https://twitter.com/HuggingPapers/status/2105755914296852789/photo/1
media 0 共 1 项
#205 🔗 原推
2026-10-02 01:08:18
AK
RT @semiDL: Excited to announce Volantis's $88M Series A. We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory. By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot. Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours. More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more. Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more. We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more: https://volantissemi.ai/news-insights/our-88m-series-a-demolishing-the-memory-wall-with-photonics-post
media 0 共 1 项
#204 🔗 原推
2026-10-02 00:51:21
AK
RT @HuggingPapers: The Teacher Is a Direction, Not a Destination New on-policy distillation method that extrapolates RL-induced representation residuals, letting students match or exceed the teacher across four model pairs. Code and checkpoints planned for release. https://twitter.com/HuggingPapers/status/2105572799104299172/photo/1
中文: RT @HuggingPapers:老师是方向，而不是目的地 新的政策蒸馏方法，可推断RL诱导的表示残差，使学生在四个模型对上匹配或超过教师。计划发布代码和检查点。
media 0 共 1 项
#203 🔗 原推
2026-10-01 17:34:14
AK
RT @tavus: Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM). https://twitter.com/tavus/status/2105704169009246248/video/1
中文: RT @tavus:介绍格里芬，这是首个通过图灵测试视频的模型。 与它交谈的人中，有48%的人认为它是真实的人。以往的系统通过率为 <3%。在英伟达全复式AI视频的基准测试中排名第一。 这是首个人机交互模型(HIM)。
🎬
视频
#202 🔗 原推
2026-10-01 17:23:05
AK
RT @HuggingPapers: UniEvo-VL A self-evolving framework where a single multimodal model acts as both teacher and student, learning from its own constructive feedback to boost image generation without external supervision. https://twitter.com/HuggingPapers/status/2105633095743398062/photo/1
media 0 共 1 项
#201 🔗 原推
2026-10-01 14:16:01
AK
RT @jakeserval: https://x.com/i/article/2105490472856903680
#200 🔗 原推
2026-10-01 14:07:47
AK
LongLive-Plug Once-for-All Distillation for Video Generation paper: https://huggingface.co/papers/2609.38154 https://twitter.com/_akhaliq/status/2105432314633490445/video/1
中文: 长生不平 视频生成用万有蒸馏 论文:
🎬
视频
#199 🔗 原推
2026-09-30 22:59:15
AK
Omni-IO Skills Harnessing Your Agent Omni-Native paper: https://huggingface.co/papers/2609.31847 https://twitter.com/_akhaliq/status/2105431737262350346/video/1
中文: 全能技能 将您的代理人全本性进行 论文:
🎬
视频
#198 🔗 原推
2026-09-30 22:56:57
AK
LEGO-Anything Coding Agents for 3D Scene Reconstruction paper: https://huggingface.co/papers/2609.36380 https://twitter.com/_akhaliq/status/2105430425657364973/video/1
中文: 乐高——一切 用于3D场景重建的编码代理 论文:
🎬
视频
#197 🔗 原推
2026-09-30 22:51:45
AK
I will be at the Hugging Face Open Together event at the Midway in SF on October 16th sign up here: https://luma.com/OpenTogether https://twitter.com/_akhaliq/status/2105380761495093276/video/1
中文: 我将于10月16日在旧金山中途岛举行拥抱面对面共同活动 注册网址:
🎬
视频
#196 🔗 原推
2026-09-30 19:34:24
AK
SpatialClaw Rethinking Action Interface for Agentic Spatial Reasoning paper: https://huggingface.co/papers/2606.13673 https://twitter.com/_akhaliq/status/2105203437680168986/photo/1
中文: 空间爪 用于空间推理的重新思考操作界面 论文:
media 0 共 1 项
#195 🔗 原推
2026-09-30 07:49:46
AK
Audio8 ASR Infinite streaming speech recognition model the native streaming architecture decodes 12.5 times per second a rolling KV Cache keeps both memory and latency constant, even in 24/7 operation one text token per clock step (12.5 / 8.3 / 6.25 decisions per second), balancing perception granularity and resource cost ML intern in huggingchat setup a gradio workflow to try it out: https://huggingface.co/spaces/akhaliq/audio8-asr-workflow huggingchat: https://huggingface.co/chat/ model: https://huggingface.co/Edge0/Audio8-ASR-Infinite
🎬
视频
#194 🔗 原推
2026-09-30 03:32:55
AK
RT @CompleteSkeptic: Links (3/3): 4 WorkflowEvals: https://huggingface.co/collections/typesafe/workflowevals Repo for reprodcution: https://github.com/typesafe-ai/WorkflowEvals Bonus dataset: https://huggingface.co/datasets/typesafe/evalsafe-onet Blog on anti-benchmaxxing: https://www.completeskeptic.com/p/lies-damned-lies-and-benchmarks
中文: RT @CompleteSkeptic:链接(3/3): 4 工作流程: 转载: 奖金数据集: 关于反benchmaxxing的博客:
#193 🔗 原推
2026-09-29 21:11:16
AK
RT @HuggingPapers: Post-Training Leaves Behavioral Shadows on Unrelated Decisions With 5,664 single-word teacher answers, a student gains +5.34pp on HumanEval+ — without ever seeing code. https://twitter.com/HuggingPapers/status/2104967837525668148/photo/1
中文: RT @HuggingPapers:培训后会因无关决策而留下行为阴影 拥有5664个单字教师答案，一名学生在HumanEval+上获得+5.34pp，但从未看到过代码。
media 0 共 1 项
#192 🔗 原推
2026-09-29 17:06:39
AK
RT @ClementDelangue: getting acquired by @nvidia = hugging face can now hire people we couldn't as a small startup and give them a decade to make open-source AI win! if you're one of them, my dms are open
中文: RT @ClementDelangue:被@nvidia收购，现在可以聘请我们作为小型初创企业无法聘用的人，并给予他们十年时间，让开源人工智能赢得胜利! 如果你是其中之一，我的dms会开放
#191 🔗 原推
2026-09-29 17:06:28
AK
RT @jakeserval: I was on @CNBC Squawk Box with one of our customers, @JackGretz, the CEO of @SeatGeek to talk about how they used Serval to eliminate the ticket backlog and redeploy their engineers as embedded business partners. ⚡ 50% of IT requests automated in first 60 days ⚡ 132 automations built in 8 weeks ⚡ Zero layoffs. Hiring this year already ahead of all of last year ⚡ IT time redeployed to solve harder problems When you automate the low-hanging tasks, the support requests and the password resets, you can take the most technical people in your organization and deploy them into work that's more productive than what they were doing before. Full clip below. 👇 @getserval
🎬
视频
#190 🔗 原推
2026-09-29 17:03:45
AK
RT @general_compute: 🚨 BREAKING: Super excited to announce we're deploying the world's fastest inference with @Cerebras. Talk to any developer and they're excited to build with 20x faster AI... the problem is there's almost no compute available. We're here to solve that. Using GPUs for prefill - it's now more affordable than ever too. Thank you to the whole Cerebras team and excited to grow this partnership. First tokens live Q127 🚀
中文: RT @general_compute:🚨 突发:非常激动地宣布，我们将使用 @Cerebras 来部署全球最快的推理技术。 与任何开发者交谈，他们都很兴奋能使用20倍速度更快的人工智能来构建......问题在于几乎没有可用的计算功能。 我们来解决这个问题。 使用GPU进行预填充——现在比以往更实惠了。 感谢整个Cerebras团队，并期待扩大这一合作关系。 首批代币在 Q127 上实时 🚀
#189 🔗 原推
2026-09-29 17:02:44
AK
FuseReg Regularizing Layer Fusion Mitigates the Reconstruction-Generation Gap in Representation Autoencoders paper: https://huggingface.co/papers/2609.31620 https://twitter.com/_akhaliq/status/2104649800654504175/photo/1
中文: FuseReg 正向层融合可缓解表征自动编码器中的重建代级差距 论文:
media 0 共 1 项
#188 🔗 原推
2026-09-28 19:09:49
AK
RT @abidlabs: 5 years ago, we released @Gradio to make it easier to build frontends for AI apps. At that time, that was the main blocker for making AI more accessible to the community. But this is no longer a challenge. So we asked ourselves and other developers: "what are the main challenges that you face now when building and using AI apps" – and we are now working hard to redesign Gradio from scratch to solve these issues!
#187 🔗 原推
2026-09-28 19:07:56
AK
RT @HuggingPapers: InternW0-Δ: a world action model for robots Pretrained on over 20,000 hours of open robot and human data, this model jointly learns predictive dynamics and action generation, outperforming prior methods across benchmarks and real robots. https://twitter.com/HuggingPapers/status/2104606518024786257/video/1
中文: RT @HuggingPapers:国际机器人世界行动模型 在经过超过2万小时开放机器人和人类数据的预训练中，本模型共同学习预测动力学和动作生成，在测试和真实机器人方面优于以往的方法。
🎬
视频
#186 🔗 原推
2026-09-28 16:28:56
AK
RT @datapointai: today, we’re releasing the largest open-source human video preferences dataset, along with doubling our data grant to $2M - 300K+ annotations by real people - 15 SOTA video models ranked (Seedance 2.5, Omni 1.1, Wan 3, Flux Video 3 etc.) - 8 categories (ads, camera movement, physics etc) dataset + benchmark + frontier plot below:
中文: RT @datapointai:今天，我们将发布最大的开源人类视频偏好数据集，同时将数据资助翻倍至200万美元 - 真实人的300K+注释 - 排名的15个SOTA视频模型(Seedance 2.5、Omni 1.1、Wan 3、Flux Video 3等) - 8个类别(广告、相机运动、物理等) 数据集 + 基准图 + 边框图:
🎬
视频
#185 🔗 原推
2026-09-28 16:27:30
AK
RT @t_mux: High quality Image generation can be fast. Viggle Turbo (6-step Qwen-Image-2.1) now runs as a @Gradio Workflow: prompt, up to 6 reference images and settings wired into one node you can see and rewire. Workflow by @_akhaliq 🙌 Try it https://huggingface.co/spaces/Viggle/Qwen-Image-2.1-viggle-turbo-workflow
#184 🔗 原推
2026-09-28 05:49:51
AK
ProgramDistill From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks paper: https://huggingface.co/papers/2609.18805 https://twitter.com/_akhaliq/status/2103952916545761536/video/1
🎬
视频
#183 🔗 原推
2026-09-26 21:00:39
AK
RT @jeffboudier: Uh oh Open Together is 90% full and we're still 3 weeks away... 50 spots left before the waitlist, get yours! Should we bring robots? https://twitter.com/jeffboudier/status/2103581590869955001/video/1
中文: RT @jeffboudier:哦，开在一起，已经满员90%了，我们还有三周时间...... 等待名单前还剩50个点，带上你的! 我们应该带机器人吗?
🎬
视频
#182 🔗 原推
2026-09-25 23:47:25
AK
Qwen-Image-2.1-viggle-turbo v0.2 is out on Hugging Face Text-to-image and image editing in 6 steps about 5× faster than the 40-step Qwen-Image-2.1 model: https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo https://twitter.com/_akhaliq/status/2103578627359268987/video/1
中文: Qwen-Image-2.1-viggle-turbo v0.2 在 Hugging Face 上推出 从文字到图像和图像进行6个步骤 比40步Qwen-Image-2.1快约5倍 型号:
🎬
视频
#181 🔗 原推
2026-09-25 20:13:22
AK
Training Object Permanence in World Models paper: https://huggingface.co/papers/2609.28654 https://twitter.com/_akhaliq/status/2103540737656918351/photo/1
中文: 世界模型中的训练对象持久性 论文:
media 0 共 1 项
#180 🔗 原推
2026-09-25 17:42:48
AK
RT @PrunaAI: The Pruna-Qwen-Image-2.1 Space is now live! You can now test our few-step LoRA adapters for Qwen-Image-2.1 by @Alibaba_Qwen directly in your browser, no setup or code required. - 5 steps for maximum speed - 8 steps for the best balance of speed and quality - Text-to-image and image editing support Put it to the test and show us the results. B𝘦𝘵𝘵𝘦𝘳 𝘷𝘦𝘳𝘴𝘪𝘰𝘯𝘴 𝘰𝘯 𝘵𝘩𝘦 𝘸𝘢𝘺. 🚀 Test the Pruna-Qwen-Image-2.1 Space on @huggingface by @_akhaliq: https://huggingface.co/spaces/PrunaAI/Pruna-Qwen-Image-2.1 Check Pruna-Qwen-Image-2.1: https://huggingface.co/PrunaAI/Pruna-Qwen-Image-2.1
中文: RT @PrunaAI:Pruna-Qwen-Image-2.1 空间现已上线! 现在，您可以直接在浏览器中通过 @Alibaba_Qwen 测试我们的几步 LoRA 适配器，使用 Qwen-Image-2.1，无需设置或代码。 - 最高速度5步 - 8 个步骤，实现最佳速度与质量平衡 - 支持文本到图像编辑 试看并给我们看结果。更好的版本在路上。🚀 在@huggingface上测试Pruna-Qwen-Image-2.1空间:@_akhaliq: 请查看Pruna-Qwen-Image-2.1:
🎬
视频
#179 🔗 原推
2026-09-25 17:25:08
AK
Pruna-Qwen-Image-2.1 a set of a few-step LoRA adapters that make Qwen-Image-2.1 up to 6.3× faster for image generation and editing Generate or edit images in just 5 or 8 steps instead of 40. Choose 5 steps for maximum speed or 8 steps for the best balance HF workflow: https://huggingface.co/spaces/akhaliq/Pruna-Qwen-Image-2.1
中文: 普鲁纳-奎恩-图片-2.1 一组几步LoRA适配器，可使Qwen-Image-2.1在图像生成和编辑方面实现最高6.3倍 生成或编辑图像只需5或8个步骤，而不是40个步骤。选择5个步骤以实现最高速度或8个步骤以实现最佳平衡 高频工作流程:
🎬
视频
#178 🔗 原推
2026-09-24 19:15:15
AK
On the Diffusibility of High-Dimensional Latents paper: https://huggingface.co/papers/2609.28473 https://twitter.com/_akhaliq/status/2103184974107328677/photo/1
中文: 高维的无花法 论文:
media 0 共 1 项
#177 🔗 原推
2026-09-24 18:09:07
AK
Contrastive Language Models HF: https://huggingface.co/Contrastive-LM https://twitter.com/_akhaliq/status/2102938541471015137/video/1
🎬
视频
#176 🔗 原推
2026-09-24 01:49:53
AK
RT @arcee_ai: We’re proud to sponsor Open Together, a free community event hosted by @huggingface on Friday, October 16, to kick off Open Source AI Week in SF. The evening is split into two parts: ➡️ 6PM – 9PM: 36 live community demos, food, drinks, and time to connect with open-source builders 🪩 9PM – 12AM: Full dance floor with live DJ sets Doors open at 6 PM, and the first 500 people through the door get collectible HF swag. 🤗 RSVP: https://luma.com/opentogether
中文: RT @arcee_ai:我们很荣幸赞助由@huggingface于10月16日星期五举办的免费社区活动Open Together，旨在旧金山启动开源人工智能周。 夜晚分为两部分: ➡️ 下午6点至9点:36个实时社区演示、食物、饮料以及与开源构建者建立联系的时间 🪩 晚上9点至12点:带现场DJ套装的全套舞池 晚上6点开门，前500人通过门时，会收到可收藏的高频车。🤗 RSVP:
🎬
视频
#175 🔗 原推
2026-09-23 18:06:11
AK
RT @Gradio: Qwen-Image 2.1 draws its best pictures when a 9B "prompt rewriter" expands your request first. That is 20 GB in bf16 and uses 1700 words as system prompt, and thinks for ~1,600 tokens before answering. We shrank it to 0.8B. It fits on a laptop now. 🧵 https://twitter.com/Gradio/status/2102464987889553848/photo/1
中文: RT @Gradio:Qwen-Image 2.1 首次扩展您的请求时，会画上最合适的图片。在 bf16 中为 20 GB，使用 1700 个单词作为系统提示，在回答之前会考虑约 1600 个令牌。 我们把它缩小到0.8B。现在它安装在笔记本电脑上。🧵
media 0 共 1 项
#174 🔗 原推
2026-09-22 19:53:18
AK
Transferring the Intelligence of VLMs to Robotic Control paper: https://huggingface.co/papers/2609.22966 https://twitter.com/_akhaliq/status/2102474539620090152/photo/1
media 0 共 1 项
#173 🔗 原推
2026-09-22 19:06:06
AK
RT @richardxp888: Thanks for sharing @_akhaliq!
中文: RT @richardxp888:感谢分享 @_akha利克!
#172 🔗 原推
2026-09-22 19:05:27
AK
onPanda Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction paper: https://huggingface.co/papers/2609.24983 https://twitter.com/_akhaliq/status/2102474190335234301/photo/1
中文: 熊猫 通过代币级更正对LLM和代理的政策对位数据进行有效说明 论文:
media 0 共 1 项
#171 🔗 原推
2026-09-22 19:04:43
AK
RT @Kun11664638: 🚀 Test the Qwen-Image-2.1 workflow on our official demo, thanks to @_akhaliq & @huggingface! 🔗 https://huggingface.co/spaces/Qwen/Qwen-Image-2.1-workflow 💡 Pro tip: Default steps are lowered due to free limits. Set to 40 & enable PE thinking to unleash its FULL power! ⚡🔥
中文: RT @Kun11664638:通过@_akhaliq & @huggingface，在我们的官方演示中测试Qwen-Image-2.1工作流程!🔗 💡 专业提示:由于免限，默认步骤会降低。设置为40和40，使PE思维能够释放其全功能!⚡🔥
#170 🔗 原推
2026-09-22 18:25:28
AK
RRSI Regularized Recursive Self-Improvement of Agent Harnesses paper: https://huggingface.co/papers/2609.24972 https://twitter.com/_akhaliq/status/2102452358118871331/photo/1
中文: RRSI 代理性手环的定期自我改进 论文:
media 0 共 1 项
#169 🔗 原推
2026-09-22 17:37:58
AK
RT @HuggingPapers: Tencent ARC Lab just released GameHorizon Suite A unified data and evaluation suite measuring AAA gameplay capabilities across multiple temporal horizons for VLMs, UMMs, GUI, coding, and game agents. https://twitter.com/HuggingPapers/status/2102314047878345185/photo/1
media 0 共 1 项
#168 🔗 原推
2026-09-22 15:30:01
AK
RT @digitalocean: DigitalOcean Managed Agents is now in public preview. Run Claude Code, Codex, or your own LangGraph agent in a runtime environment that pauses when idle. Put its tools behind one governed endpoint, and pick from 75+ open and proprietary models. One cloud, one bill. Prompts to get started available in the blog: https://do.co/4ysh3it
🎬
视频
#167 🔗 原推
2026-09-22 15:27:14
AK
RT @HuggingPapers: RoboDawn: Transferring the Intelligence of VLMs to Robotic Control Tsinghua and Tencent Hunyuan show a frozen VLM can drive robots through simple move, rotate and gripper commands, reaching 73.6% one-shot success on RoboTwin 2.0 C2R. https://twitter.com/HuggingPapers/status/2102371789842071837/photo/1
中文: RT @HuggingPapers:机器人黎明:将VLM的智能传输到机器人控制 清华和腾讯百源展示了一种冷冻的VLM，能够通过简单的移动、旋转和抓握指令驱动机器人，在RoboTwin 2.0 C2R上达到73.6%的一击成功。
media 0 共 1 项
#166 🔗 原推
2026-09-22 15:17:00
AK
RT @PixVerse: Meet PixVerse R2, our new real-time world model. Explore living worlds. Control and edit them with prompts. Shape the story. Meet characters that remember and respond. https://twitter.com/PixVerse/status/2102404266484989983/video/1
🎬
视频
#165 🔗 原推
2026-09-22 15:07:22
AK
WorldCrafter Consistent Video World Model with Implicit 3D-aware Memory paper: https://huggingface.co/papers/2609.24984 https://twitter.com/_akhaliq/status/2102266547108819254/video/1
🎬
视频
#164 🔗 原推
2026-09-22 05:19:37
AK
RT @jeffboudier: Excited to welcome @reflection_ai as a co-host and sponsor of Open Together! 🤗 We’re already at 50% capacity! Grab your spot before registration switches to the waitlist. 📍 San Francisco · October 16 🎟️ Register: https://luma.com/OpenTogether https://twitter.com/jeffboudier/status/2102236502210310613/video/1
🎬
视频
#163 🔗 原推
2026-09-22 03:45:53
AK
CodeMidas Scaling Agentic Coding RL Environments from Code Itself paper: https://huggingface.co/papers/2609.22068 https://twitter.com/_akhaliq/status/2102205326485557643/photo/1
中文: CodeMidas 从代码本身扩展代理编码 RL 环境 论文:
media 0 共 1 项
#162 🔗 原推
2026-09-22 01:16:21
AK
RT @Gradio: Gradio Workflow on Product hunt! Connect nodes to build AI pipelines, powered by Hugging Face https://www.producthunt.com/products/gradio?utm_source=twitter&utm_medium=social
中文: RT @Gradio:产品搜索的Gradio工作流程! 连接节点以构建由 Hugging Face 提供支持的 AI 管道
#161 🔗 原推
2026-09-21 23:08:54
AK
RT @HuggingPapers: Code2Skill A fully automated pipeline that transforms 19,769 open-source repositories into over 1 million verified, reusable procedural skills for AI agents. https://twitter.com/HuggingPapers/status/2102008836865352067/photo/1
中文: RT @HuggingPapers:Code2Skill 一条完全自动化的管道，将19769个开源仓库转化为超过100万个经过验证的可重复使用过程技能，供AI代理使用。
media 0 共 1 项
#160 🔗 原推
2026-09-21 18:53:28
AK
RT @DeemosTech: 🚀 Introducing #HYPER3D Agentic Mode. AGI optimizes your input and chooses how to model it. Generate editable N-GONS #3D model for #CAD, building & industrial product. 📏Adjust dimensions. 🏃‍♀️Add animation. Towards Production-Ready, to Agent-Ready. This is Vibe Modeling👇 https://twitter.com/DeemosTech/status/2102055632887324775/video/1
中文: RT @DeemosTech:🚀 介绍#HYPER3D 代理模式。 AGI优化了你的输入，并选择如何进行建模。 为#CAD、建筑及工业产品生成可编辑的N-GONS #3D模型。 📏 调整尺寸。🏃♀️ 添加动画。 面向生产准备，迈向代理准备。 这是Vibe建模👇
🎬
视频
#159 🔗 原推
2026-09-21 18:22:42
AK
RT @whitecircle: Introducing Halo, the best framework for post-training of open-source models. Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format. Star us on GitHub: https://github.com/whitecircle/halo https://twitter.com/whitecircle/status/2102087563913609534/video/1
🎬
视频
#158 🔗 原推
2026-09-21 17:37:20
AK
RT @HuggingPapers: Alibaba's Qwen team just released RecreationWorld on Hugging Face A five-platform sandbox where hybrid computer-use agents explore real apps, implement them, and visually verify their own builds. https://twitter.com/HuggingPapers/status/2102068609396736180/photo/1
中文: RT @HuggingPapers:阿里巴巴Qwen团队刚刚在Hugging Face上发布了RecreationWorld 一个五平台的沙盒，供混合计算机使用代理探索真实应用程序、实现并直观验证自身构建。
media 0 共 1 项
#157 🔗 原推
2026-09-21 16:17:46
AK
RT @dimakhanarin: Introducing Codos: The first virtual Chief AI Officer. AI is crushing all benchmarks but real companies still struggle to see P&L impact. Codos interviews employees, deploys automations across all functions and gets smarter over time while running on your own servers. Our NASDAQ-listed and PE-backed customers are adding millions to their bottom line months ahead of schedule and we are proud of the first results we deliver. It’s time to turn the 500BN AI-transformation market into software and unlock the impact for the real economy.
中文: RT @dimakhanarin:介绍 Codos:首任虚拟首席人工智能官。 人工智能正在压垮所有基准，但真正的公司仍难以看到与全时制搏斗的影响。 Codos 会面试员工，在所有功能中部署自动化，并在您自己的服务器上运行时，会随着时间的推移变得更加智能。 我们在纳斯达克上市和由PE支持的客户将比计划提前数月内增加数百万的利润，我们为首批业绩结果感到自豪。 是时候将500BN人工智能转型市场转变为软件，并释放对实体经济的影响了。
🎬
视频
#156 🔗 原推
2026-09-21 16:07:17
AK
Qwen-Image-2.1 is out on Hugging Face app: https://huggingface.co/spaces/akhaliq/Qwen-Image-2.1 https://twitter.com/_akhaliq/status/2101679817762734150/photo/1
中文: Qwen-Image-2.1 在 Hugging Face 上推出 应用程序:
media 0 共 1 项
#155 🔗 原推
2026-09-20 14:28:10
AK
RT @abidlabs: Just one of the many updates in @Gradio 6.28: You can now pass in the live state of any Gradio component as an input to your ML functions! https://twitter.com/abidlabs/status/2101084389908328479/video/1
中文: RT @abidlabs:@Gradio 6.28 中的众多更新之一:现在，您可以以任何 Gradiio 组件的实时状态输入，输入您的机器学习功能!
🎬
视频
#154 🔗 原推
2026-09-19 01:02:41
AK
RT @zihaozhang__: Thanks so much for sharing our work, AK! Really appreciate it 🙌
中文: RT @zihaozhang__:非常感谢分享我们的作品，AK!非常感谢🙌
#153 🔗 原推
2026-09-18 21:54:44
AK
JEPA-Anything Learning Predictive Models across Different Worlds paper: https://huggingface.co/papers/2609.20800 https://twitter.com/_akhaliq/status/2101021609113112658/photo/1
media 0 共 1 项
#152 🔗 原推
2026-09-18 18:52:41
AK
An Empirical Study of Harness Design for Coding Agents paper: https://huggingface.co/papers/2609.20804 https://twitter.com/_akhaliq/status/2101020560964866103/photo/1
中文: 编码剂用逞设计的实证研究 论文:
media 0 共 1 项
#151 🔗 原推
2026-09-18 18:48:31
AK
RT @ClementDelangue: We’ll kick off the Open Source AI Week in SF with a meetup, community demos, and a dance party. Let's all push and celebrate more openness and transparency in AI! https://twitter.com/ClementDelangue/status/2100964987720155620/photo/1
中文: RT @ClementDelangue:我们将在旧金山开启开源人工智能周，举办一场聚会、社区演示活动和一场舞会。 让我们共同推动并庆祝人工智能领域更加开放和透明!
media 0 共 1 项
#150 🔗 原推
2026-09-18 16:29:30
AK
RT @HuggingPapers: NVIDIA's SoL-Pi, now on Hugging Face paper pages A token-efficient agent harness built by recursively scaling auto-research loops. It cuts token traffic by 44.7-49.0% and API cost by about one third while preserving performance. https://twitter.com/HuggingPapers/status/2100861356442284375/video/1
🎬
视频
#149 🔗 原推
2026-09-18 08:55:31
AK
RT @XGEN_labs: 🔥 Today, we are truly excited to announce our technical prototype, the Generative World Simulation system, which integrates JING（镜）, an interactive experience model, with DAO（道）, a computable shared-world engine. The coupled model and engine connect first-person experience with a shared world that continues to evolve beyond any individual observer. Conditioned on actions and observation history, JING enables agent navigate, manipulate, and communicate from a first-person perspective in the world. Watch our demo video to see it in action! On the official WBench leaderboard as of September 17, 2026, XGEN-JING ranked #1 on the Full split, and #2 on the Navi split. 🎉 DAO maintains shared world state and rules, computes the consequences of actions, and provides JING with only what the current observer can perceive. It also supports autonomous agent decision-making, enabling agents to act independently within an evolving shared world. Together, DAO and JING move beyond generating the next frame toward simulating the world behind it. This marks a small step towards OASIS: not just a world that responds to you, but a world—and a society—that evolves with and without you. 💪 Explore XGEN Labs～: 🔗 Website: https://xgenlabs.ai/ 🤗 HF: https://huggingface.co/XGENlabs/XGEN-JING 🦊 GitHub: https://github.com/XGEN-Labs/XGEN-JING/
🎬
视频
#148 🔗 原推
2026-09-18 08:52:59
AK
RT @HuggingPapers: Coding agent harnesses, taken apart A new study isolates planning, action space, and context management across 176 settings. Key finding: context management matters most when budgets are tight, and planning flips from accuracy to efficiency as models improve. https://twitter.com/HuggingPapers/status/2100800263644705178/photo/1
中文: RT @HuggingPapers:编码代理线束，被拆开 一项新研究将规划、行动空间和上下文管理隔离在176个环境中。关键发现:当预算紧张时，语境管理最重要，而规划会随着模型的改进而从准确性转向效率。
media 0 共 1 项
#147 🔗 原推
2026-09-18 08:41:17
AK
RT @NetEaseYouDaoAI: Most streaming ASR gives you text fast. R2T2 gives you text you can act on. We’re open-sourcing Confucius4-R2T2, a 1.7B frontier real-time streaming ASR model, built for voice agents. Here is the link: GitHub: https://github.com/netease-youdao/Confucius4-R2T2 Hugging Face: https://huggingface.co/netease-youdao/Confucius4-R2T2
#146 🔗 原推
2026-09-18 08:35:59
AK
RT @yifanzhang_: Thanks for sharing! https://github.com/yifanzhang-pro/Agora
中文: RT @yifanzhang_:谢谢分享!
#145 🔗 原推
2026-09-17 20:42:25
AK
Agora Git as Shared Memory for Collective AutoResearch paper: https://huggingface.co/papers/2609.18094 https://twitter.com/_akhaliq/status/2100676592577937659/photo/1
中文: 阿戈拉 Git 作为 Collective AutoResearch 的共享内存 论文:
media 0 共 1 项
#144 🔗 原推
2026-09-17 20:01:43
AK
LimiX-2 A Contextual Mechanism Network Towards General Structured-Data Intelligence paper: https://huggingface.co/papers/2609.17488 https://twitter.com/_akhaliq/status/2100636566997836191/video/1
中文: LimiX-2 面向通用结构化数据智能的情境机制网络 论文:
🎬
视频
#143 🔗 原推
2026-09-17 17:22:40
AK
RT @jeffboudier: Built a cool HF Space? Bring it to Open Together 🤗 Community demo applications are now open! Oct 16 · The Midway, SF Meet AI Builders. Try demos. Dance. Free registration ➡️ https://luma.com/OpenTogether Apply by Oct 2 ➡️ https://huggingface.co/spaces/OpenTogether/community-demos #OpenTogether #OpenSourceAIWeek https://twitter.com/jeffboudier/status/2100613673970798836/video/1
🎬
视频
#142 🔗 原推
2026-09-17 16:01:53
AK
RT @jeffboudier: The AI revolution will not be centralized. 🤗 📢 Calling all AI builders to San Francisco on Friday, October 16, one month from today! We’ll kick off Open Source AI Week with Open Together: a huge meetup, community demos, and a dance party. Come meet the people building with open models. Everyone should be able to build AI they can control. That’s the gift of open models, and a future worth celebrating together. I’d love to see you there! Who’s coming? Tell me what you’re building in the replies 👇 Register here: https://luma.com/OpenTogether
中文: RT @jeffboudier:人工智能革命不会集中化。🤗 📢 于10月16日星期五，从今天起一个月，致电所有人工智能开发者前往旧金山! 我们将通过“开放携手”开启开源人工智能周:一场大型聚会、社区演示和一场舞会。快来见见人们用开放式模型来建造。 每个人都应该能够构建他们能够控制的人工智能。这就是开放式模型的馈赠，也是值得共同庆祝的未来。我很想在那里见到你! 谁来的?在回复中告诉我你正在构建什么👇 请在此注册:
media 0 共 1 项
#141 🔗 原推
2026-09-16 18:13:04
AK
StepAudio 3 Realtime Technical Report paper: https://huggingface.co/papers/2609.14005 https://twitter.com/_akhaliq/status/2100257483675873481/photo/1
中文: StepAudio 3 实时技术报告 论文:
media 0 共 1 项
#140 🔗 原推
2026-09-16 16:16:19
AK
Continual Learning Mechanisms Compose for Long-Horizon Memorization paper: https://huggingface.co/papers/2609.06986 https://twitter.com/_akhaliq/status/2100256379034615905/photo/1
中文: 连续学习机制为长视记忆而合谋 论文:
media 0 共 1 项
#139 🔗 原推
2026-09-16 16:11:56
AK
RT @victormustar: HuggingChat goes whale 🐋 New default model: DeepSeek-V4.1-Flash. Go try it to get how good open AI is now. All you need is a Hugging Face (free) account. Bonus: it searches and crawls the web for you using Exa https://twitter.com/victormustar/status/2100181580467564641/photo/1
#138 🔗 原推
2026-09-16 14:24:36
AK
RT @HuggingPapers: PhysBrain 1.5: Open-source SOTA for Embodied AI A single 8B model that understands scenes, generates robot actions, and predicts future frames — all as tokens. 72.5 on 28 benchmarks, beating all open models. https://twitter.com/HuggingPapers/status/2099894590677348649/photo/1
中文: RT @HuggingPapers:PhysBrain 1.5:用于实体人工智能的开源SOTA 一个8B模型，用于理解场景、生成机器人动作并预测未来框架——全部作为代币。28 个基准测试区 72.5，超过所有开放型号。
media 0 共 1 项
#137 🔗 原推
2026-09-15 20:05:24
AK
RT @HuggingPapers: NVIDIA just released FoundationPose on Hugging Face A unified foundation model for 6-DoF object pose estimation and tracking that works on novel objects without fine-tuning. https://twitter.com/HuggingPapers/status/2099914720153129087/photo/1
中文: RT @HuggingPapers:英伟达刚刚发布了关于拥抱脸的FoundationPose 6-DoF 对象的统一基础模型，可对新对象进行建模，无需微调即可进行。
media 0 共 1 项
#136 🔗 原推
2026-09-15 18:28:21
AK
RT @CompleteSkeptic: After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution
🎬
视频
#135 🔗 原推
2026-09-15 18:26:00
AK
RT @odysseyml: Today we’re unveiling Odyssey-3, a big step forward for foundation world models. It can control robots, power humanoids, drive cars (on the roads of India!), train AIs, pilot drones, and even play video games. We can’t wait to see what intelligent systems it enables. https://twitter.com/odysseyml/status/2099900067356586276/video/1
🎬
视频
#134 🔗 原推
2026-09-15 16:46:23
AK
Vidu S2 Real-Time Interactive, Editable, and Spatial Video Generation paper: https://huggingface.co/papers/2609.11638 https://twitter.com/_akhaliq/status/2099883001840828920/photo/1
中文: 维杜 S2 实时交互、可编辑和空间视频生成 论文:
media 0 共 1 项
#133 🔗 原推
2026-09-15 15:28:16
AK
RT @HuggingPapers: DataFlex-RL A controlled evaluation platform for RLVR data policies, comparing rollout selection, reweighting, and domain-mixture choices under a shared GRPO recipe. https://twitter.com/HuggingPapers/status/2099531560768970833/photo/1
中文: RT @HuggingPapers:DataFlex-Rl RLVR数据策略的受控评估平台，通过共享GRPO配方对推广选择、调整权重和域混合选择进行比较。
media 0 共 1 项
#132 🔗 原推
2026-09-14 16:24:01
AK
RT @boltdotnew: Introducing Bolt Forge. Free until Oct 14th: - Up to 50x more usage - The new frontier: GLM, DeepSeek, Kimi - Zero usage charges Live now in your model picker on https://bolt.new/ And one more thing... 👇 https://twitter.com/boltdotnew/status/2099529089430229059/video/1
中文: RT @boltdotnew:介绍 Bolt Forge。免费至10月14日: - 最多可使用50倍 - 新领域:GLM、DeepSeek、Kimi - 零使用费 立即登录 登录您的模特采摘机 还有一件事......👇
🎬
视频
#131 🔗 原推
2026-09-14 16:22:11
AK
SAS Simple Attention Sparsification via End-to-End Optimization of Context Ranking paper: https://huggingface.co/papers/2609.13141 https://twitter.com/_akhaliq/status/2099511334987522298/photo/1
media 0 共 1 项
#130 🔗 原推
2026-09-14 14:51:23
AK
RT @abidlabs: Along with everything else, we've been improving Gradio and just released 6.27! In this release, you'll see a bunch of nice bug fixes, especially around audio streaming and workflows! https://github.com/gradio-app/gradio/blob/main/CHANGELOG.md https://twitter.com/abidlabs/status/2098455388383477781/photo/1
中文: RT @abidlabs:除了其他所有内容外，我们一直在改进Gradio，并且刚刚发布了6.27! 在此版本中，您将看到一系列不错的错误修复，尤其是在音频流和工作流程方面!
media 0 共 1 项
#129 🔗 原推
2026-09-11 17:01:10
AK
RT @HuggingPapers: SWE-Bench Pro Verified OpenCompass released a verified version of SWE-Bench Pro that fixes reward hacking and task-quality issues, revealing frontier models score far lower than previously reported. https://twitter.com/HuggingPapers/status/2098142214450782282/photo/1
中文: RT @HuggingPapers:SWE-Bench Pro 已验证 OpenCompass 发布了经过验证的 SWE-Bench Pro 版本，可修复奖励黑客行为和任务质量问题，显示前沿模型的得分远低于此前报道。
media 0 共 1 项
#128 🔗 原推
2026-09-10 21:40:48
AK
RT @hliriani: Yesterday we announced our series A to reimagine CRM as a business world model. I wanted explain what we mean by that, and why the CRM is the practical place to start building something much larger. https://x.com/i/article/2098159001070673920
中文: RT @hliriani:昨天我们发布了A系列，将CRM重新构想为商业世界模式。 我想解释一下我们所说的含义，以及为什么CRM是开始构建更大规模事物的实用场所。
#127 🔗 原推
2026-09-10 21:35:43
AK
RT @victormustar: I can confirm it now, DeepSeek V4.1 Flash is amazing 🥹 https://twitter.com/victormustar/status/2098080125203976632/photo/1
中文: RT @victormustar:我可以确认，DeepSeek V4.1 Flash 太棒了 🥹
media 0 共 1 项
#126 🔗 原推
2026-09-10 17:28:51
AK
RT @Thom_Wolf: Two big updates 1. I published an @FT op-ed on the OpenAI/HF incident & follow-ups 2. We’re starting an Open Alignment team at @huggingface to work on safety & alignment for open models, incl cybersecurity Need 100x more transparency & research on this https://www.ft.com/content/9faf688d-9192-418e-b7d3-c2202526e85e
中文: RT @Thom_Wolf:两项重大更新 1。我发表了一篇关于OpenAI/HF事件及后续动态的@FT评论文章 2。我们将在@huggingface 组建一个开放联盟团队，致力于安全与安普、开放模式的对齐，包括网络安全 需要再提高100倍的透明度;对此进行研究
#125 🔗 原推
2026-09-10 17:28:23
AK
RT @OpenBMB: 🚀MiniCPM5-2B hits #1 on @huggingface Trending! 🏆 Huge thanks to the community for the incredible support!💗 Ranked #1 among open-weight models under 4B parameters worldwide in the @ArtificialAnlys Intelligence Index. 🤖 Built for Agentic AI Tool calling, deep search, code generation, and more—bringing capable Agents to phones, PCs, and vehicles. 🧠 More than model weights We’re opening up training code, Agent SFT/RL data, UltraX, Meshy, and JustRL II, enabling deeper research and easier reproduction. 📱 Ready for the edge Day 0 support for Intel, AMD & Arm, plus mainstream inference and fine-tuning frameworks. Try the model here: 🤗 Hugging Face: https://huggingface.co/openbmb/MiniCPM5-2B 💻 GitHub: https://github.com/OpenBMB/MiniCPM
中文: RT @OpenBMB:在@huggingface Trending上排名第一的迷你CPM5-2B热门歌曲!🏆 非常感谢社区提供的大力支持!💗 在@ArtificialAnlys 智能指数中，全球在4B参数下的开放权重模型中排名第一。 🤖 专为智能AI打造 工具调用、深度搜索、代码生成等——将功能强大的代理功能带入手机、个人电脑和车辆中。 🧠 比模型权重更多 我们正在开放训练代码、Agent SFT/RL 数据、UltraX、Meshy 和 JustRL II，从而实现更深入的研究和更便捷的复制。 📱 为边缘做好准备 支持英特尔、AMD 和 Arm 以及主流推理和微调框架。 试试这里的模型: 🤗 拥抱面容: 💻 GitHub:
media 0 共 1 项
#124 🔗 原推
2026-09-10 17:19:15
AK
RT @cohere: We were founded on advancements in translation. We couldn’t be prouder to continue that legacy. Find the weights on Hugging Face (available in several quants) under a CC BY-NC 4.0 license. https://huggingface.co/CohereLabs/North-Small-Translate-1.0
中文: RT @cohere:我们建立在翻译技术的进步之上。我们为延续这一遗产感到无比自豪。使用CC BY-NC 4.0许可证，在Hugging Face(多种量子版本中可用)上查找权重。
#123 🔗 原推
2026-09-10 17:15:02
AK
RT @HuggingPapers: AgentGrad: Intervention-guided Prompt Optimization Improves multi-agent prompt optimization via sequential intervention and semantic textual gradient abstraction, achieving SOTA on 5 MAS benchmarks and 2.5x faster optimization. https://twitter.com/HuggingPapers/status/2098022487405793367/photo/1
中文: RT @HuggingPapers:代理分级:干预引导的快速优化 通过顺序干预和语义文本梯度抽象改进多智能加速优化，在5个MAS基准测试中实现SOTA，并实现2.5倍更快的优化。
media 0 共 1 项
#122 🔗 原推
2026-09-10 14:41:52
AK
RT @eglyman: Ramp made a real Broadway musical starring Billy Porter, Billy Zane and Jessica Billy Vosk. It’s a show about bills. By Bills. Starring Bills—with original songs, full choreography, and an actual Broadway playwright. In the 1950s and ’60s, America’s biggest companies staged lavish musicals about cars, appliances, and soda. Somewhere along the way, we lost our nerve. We’re bringing back the industrial musical. One night only, September 25, at the Broadhurst Theatre. https://billpaythemusical.com/
中文: RT @eglyman:Ramp 制作了一部真正的百老汇音乐剧，由比利·波特、比利·扎恩和杰西卡·比利·沃斯克主演。 这是一场关于账单的节目。比尔。主演比尔斯——配以原创歌曲、完整编舞以及一位真正的百老汇剧作家。 20世纪50年代和60年代，美国最大的公司举办了关于汽车、家电和汽水的奢华音乐剧。途中，我们失去了神经。 我们正在重新推出这部工业音乐剧。 仅在9月25日，在布罗德赫斯特剧院的一晚。
🎬
视频
#121 🔗 原推
2026-09-10 14:37:50
AK
RT @deepseek_ai: 🌐 Supporting open source. Expanding deployment options. We’ll work closely with the open-source community on V4.1-Flash inference support and explore more deployment options. Planning a large-scale deployment with 2,000 GPUs + a storage cluster? Let’s talk. 🔹 Model: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash 🔹 Paper: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/DeepSeek_V41_Tech_Report.pdf 6/6
中文: RT @deepseek_ai:🌐 支持开源。扩展部署选项。 我们将与开源社区在V4.1-Flash推理支持方面密切合作，并探索更多部署选项。 计划大规模部署，配备 2000 个 GPU 和一个存储集群?让我们聊聊。 🔹 型号: 🔹 论文: 6/6
#120 🔗 原推
2026-09-10 06:46:58
AK
AuK Technical Report An Open-Source Foundational Model for Speech Generation and Editing paper: https://huggingface.co/papers/2609.08936 https://twitter.com/_akhaliq/status/2097857652651167838/photo/1
中文: AuK技术报告 用于语音生成和编辑的开源基础模型 论文:
media 0 共 1 项
#119 🔗 原推
2026-09-10 01:20:15
AK
RT @abidlabs: two ways to do object recognition... https://twitter.com/abidlabs/status/2097754409816170681/photo/1
media 0 共 1 项
#118 🔗 原推
2026-09-09 20:55:13
AK
Marigold V2 Revisiting Diffusion Transformers for Monocular Depth Estimation paper: https://huggingface.co/papers/2609.08084 https://twitter.com/_akhaliq/status/2097726716861186348/video/1
中文: 万寿菊V2 重温扩散变换器，实现单目深度估算 论文:
🎬
视频
#117 🔗 原推
2026-09-09 16:39:57
AK
NeoHorse-1 Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness paper: https://huggingface.co/papers/2609.08183 https://twitter.com/_akhaliq/status/2097726309736874183/photo/1
media 0 共 1 项
#116 🔗 原推
2026-09-09 16:38:20
AK
RT @HuggingPapers: Tencent releases AuK, a unified speech generation and editing model It handles zero-shot TTS, instruction-driven content/acoustic/paralinguistic editing, plus enhancement and separation through one natural-language interface. The distilled AuK-Flash version runs 4.5x faster. https://twitter.com/HuggingPapers/status/2097660279626609134/photo/1
中文: RT @HuggingPapers:腾讯发布AuK，一款统一的语音生成和编辑模式 它处理零分量的TTS、以教学为导向的内容/声学/穿伞式编辑，以及通过一个自然语言界面进行增强和分离。蒸馏版AuK-Flash的运行速度为4.5倍。
media 0 共 1 项
#115 🔗 原推
2026-09-09 16:10:56
AK
RT @lightfld: We’ve raised $47M in Series A funding led by @a16z to reimagine CRM as a world model of a business.
中文: RT @lightfld:我们通过 @a16z 牵头的 A 轮融资筹集了 4700 万美元，用于将 CRM 重新构想为企业的全球模式。
#114 🔗 原推
2026-09-09 16:09:53
AK
Unlocking Lossless Speedups in LLMs via Discrete Diffusion paper: https://huggingface.co/papers/2609.04010 https://twitter.com/_akhaliq/status/2097454804272566527/photo/1
中文: 通过离散扩散解锁无损加速 论文:
media 0 共 1 项
#113 🔗 原推
2026-09-08 22:39:28
AK
RT @abidlabs: "How much do large language models memorize" by @jxmnop et al. is a fantastic paper that received one of the best paper awards at ICML 2026 I took the central thesis of the paper and asked ML Intern to reproduce it. It's really cool to see it - come up with a plan - set a hard budget for gpu compute - run experiments, track the right metrics - reproduce the central claims (at a smaller scale) for <$8 Incredibly useful for reviewers or anyone building on top of the work!
media 0 共 4 项 media 1 media 2 media 3
#112 🔗 原推
2026-09-08 20:43:42
AK
RT @evilpingwin: You shouldn’t need to be an ML expert to have an ML idea. Today we’re launching ML Intern in HuggingChat. Start with a conversation. Finish with deployable artefacts. https://twitter.com/evilpingwin/status/2097359184207511654/video/1
中文: RT @evilpingwin:你不该需要成为机器学习专家才能拥有机器学习的创意。 今天我们将在HuggingChat中推出ML实习生。 从对话开始。使用可部署的人工制品完成。
🎬
视频
#111 🔗 原推
2026-09-08 17:12:06
AK
RT @HuggingPapers: FlowBalance: verifier-grounded self-improvement for reasoning models Improves math reasoning by +2.12 avg over GRPO on Qwen3-8B, with faster training, better stability, and higher solution diversity. https://twitter.com/HuggingPapers/status/2097298711697072142/photo/1
中文: RT @HuggingPapers:FlowBalance:基于验证者的自我提升，适用于推理模型 通过在Qwen3-8B上以+2.12的速度提升数学推理能力，提高数学推理速度，提高训练速度，提高稳定性，实现更高的解决方案多样性。
media 0 共 1 项
#110 🔗 原推
2026-09-08 15:38:20
AK
RT @antiochrobotics: Today, we’re announcing Antioch’s $32 million Series A, led by @GreylockVC with participation from @A_StarVC, @Category_VC, @BoxGroup, @IcehouseVenture, and angels. While AI has drastically accelerated software development, physical autonomy has been constrained by slow, expensive hardware-based development. Antioch enables physical AI teams to build, test, and validate systems at the speed of software. We’re proud to be working with leading teams including @amazon @ring, @NVIDIARobotics, and @nebiusai as we make scaled, high-fidelity simulation the standard for physical AI development. Our sincere thanks to the legion of investors, customers, partners, and advisors who are making this step change possible. We’re scaling rapidly and hiring across simulation, infrastructure, machine learning, 3D graphics, go-to-market, and operations.
中文: RT @antiochrobotics:今天，我们宣布安提阿推出的A轮融资，由@GreylockVC领衔，由@A_StarVC、@Category_VC、@BoxGroup、@IcehouseVenture和Angels领衔。 尽管人工智能极大地加速了软件开发，但物理自主性却受到基于硬件的缓慢且昂贵的开发的制约。安提阿让物理人工智能团队能够以软件的速度构建、测试和验证系统。 我们很荣幸能与包括 @amazon @ring、@NVIDIArobotics 和 @nebiusai 在内的领先团队合作，将规模化、高保真度仿真作为物理人工智能开发的标准。 我们衷心感谢让这一步骤改变成为可能的投资者、客户、合作伙伴和顾问。 我们正在快速扩展业务，并在仿真、基础设施、机器学习、3D图形、上市和运营方面进行招聘。
🎬
视频
#109 🔗 原推
2026-09-08 15:10:39
AK
RT @mishig25: Monaco F1 circuit replica made with astra https://huggingface.co/spaces/mishig/monaco-circuit-atlas https://twitter.com/mishig25/status/2097091805200220350/video/1
中文: RT @mishig25:使用 astra 制作的摩纳哥F1电路复刻版
🎬
视频
#108 🔗 原推
2026-09-08 03:26:15
AK
RT @julien_c: I’ve been overwhelmed and humbled by the fact that 99% of the reactions to our intent to join forces have been positive. ❤️ Prominent tech figures, including some you could see as having a competitive relationship with NVIDIA, have been vocally supportive: • @satyanadella: “this ecosystem with open models continuing to flourish and grow” • @sundarpichai: “this will strengthen the open model ecosystem!” • @miramurati: “A great home for Hugging Face to continue expanding access to open models and the tools that let people build with them.” • @alighodsi: “This will be great for the industry and open source.” • @AravSrinivas: “absolutely necessary for AI to remain accessible and useful to the public.” He added that he was glad NVIDIA was supporting Hugging Face and the open-source community. • @MichaelDell: “a big win for the whole ecosystem.” • @levie: “Another huge moment for open weights AI.” • @adcock_brett: “This is good for everyone building and open source.” • @elonmusk: “Congrats!” • @osanseviero: “The impact of HF in the AI ecosystem has been massive” Even more importantly, the community has been leaning into the announcement. Now we’ll get to demonstrate through our actions in the coming months/years that we will keep up the good work 🔥 This is a chance to make open-source AI so much bigger.
#107 🔗 原推
2026-09-08 02:33:39
AK
RT @rafaelcaricio: First selfie together!!! #microduck https://twitter.com/rafaelcaricio/status/2097116865579401315/photo/1
中文: RT @rafaelcaricio:第一次合影!!!#microduck
media 0 共 1 项
#106 🔗 原推
2026-09-08 02:33:16
AK
RT @LeRobotHF: @nvidia and @huggingface are joining forces! At LeRobot, we believe the future of physical AI must be collaborative, open, and accessible to everyone. This partnership brings incredible momentum to the open robotics community while keeping our commitment to flexible, multi-platform development stronger than ever. Can't wait to see what you all build as we push the boundaries of embodied AI together. 🤗💚
中文: RT @LeRobotHF:@nvidia 和 @huggingface 正在联手! 在LeRobot，我们相信物理人工智能的未来必须能够协作、开放且人人都能被访问。此次合作为开放机器人社区带来了巨大的推动力，同时使我们对灵活、多平台发展的承诺比以往任何时候都更加坚定。 迫不及待想看到你们共同构建的，共同突破了人工智能的界限。🤗💚
🎬
视频
#105 🔗 原推
2026-09-04 23:24:53
AK
RT @HuggingPapers: NVIDIA just released the NVFP4 quantized Qwen3.8-Flash-Next on Hugging Face 125B MoE with hybrid attention, now 63% smaller with minimal accuracy loss. https://twitter.com/HuggingPapers/status/2095973698356609037/photo/1
中文: RT @HuggingPapers:英伟达刚刚发布了NVFP4在Hugging Face上对Qwen3.8-Flash-Next进行量化 带混合型的125B MoE，现在体积小63%，精度损耗最小。
media 0 共 1 项
#104 🔗 原推
2026-09-04 21:18:22
AK
RT @HuggingPapers: Terminal-Universe A framework that reconstructs realistic terminal environments from agent trajectories, yielding 37.3k task-sufficient workspaces for post-training and evaluation. https://twitter.com/HuggingPapers/status/2095848000677347566/photo/1
中文: RT @HuggingPapers:终端宇宙 一个从代理轨迹重建真实终端环境的框架，为训练后和评估提供37.3k的任务空间。
media 0 共 1 项
#103 🔗 原推
2026-09-04 18:02:37
AK
RT @HuggingPapers: Compile by Training Turn natural-language specs into small neural programs that run locally on a compact interpreter. Compilation takes about a minute, and on FuzzyBench-Hard it hits 83.6% semantic accuracy. https://twitter.com/HuggingPapers/status/2095907902858871136/video/1
中文: RT @HuggingPapers:通过培训进行编译 将自然语言规范转化为在紧凑型解释器上本地运行的小型神经程序。编译大约需要一分钟，在FuzzyBench-Hard上，其语理准确率达到了83.6%。
🎬
视频
#102 🔗 原推
2026-09-04 16:43:46
AK
RT @viskoai: Our evaluation results for Orbis 1.0: Across automated benchmarks, Orbis scores highest on DOVER aesthetic and technical quality, and on VideoAlign visual and motion quality. It also leads three physical-plausibility protocols: VideoPhy-2, Physics-IQ, and VBench-2.0 Physics. In a randomized human Arena study of state-of-the-art real-time interactive video systems, Orbis ranked first in overall preference and first in temporal stability, the prerequisite for long-form generation. Full report: https://www.visko.ai/blog/papers/files/Visko_Orbis_1_0__A_Live_Model_for_Real_TimeInteractive_Long_Video_Generation.pdf
中文: RT @viskoai:我们对Orbis 1.0的评估结果: 在自动化测试中，Orbis 在 Dover 的美学和技术质量以及 VideoAlind 的视觉和运动质量方面均获得最高评分。它还负责三项物理可理性规范:VideoPhy-2、Physics-IQ 和 VBench-2.0 物理。 在一项针对先进实时交互式视频系统的随机人类竞技场研究中，奥比斯在整体偏好方面排名第一，在时间稳定性方面排名第一，而时间稳定性是长篇大传的前提条件。 完整报告:
media 0 共 1 项
#101 🔗 原推
2026-09-04 16:36:54
AK
RT @victormustar: reflecting a bit on my (long) journey @huggingface - most big things start very small: > around 6 years ago @julien_c contacts me (on twitter) he's building a "hub for ml models community" (wtf is this). Still I feel it's interesting and I want to be part of it > day 18: we start iterating on design ideas and prototypes with the Hub team - trying a lot of stuff to understand what could work (and we are having fun doing it) > day 41: julien shares our mockups with the team, "everyone is pretty excited". 2 weeks later an offer email arrives, subject: "Offer 🤗😱" > day 167: we write "The AI community building the future", 2 days before launch, not sure anyone will care > day 169: launch day, julien: "we launch now or tomorrow?" me: "now, I'm done". we launch the new Hugging Face Hub > day 318: clem posts traffic charts in slack at 1am: "very nice growth in august!". Glimpses of something happening 👀 > day 687: it's not small anymore... also on this day dalle-mini goes viral, for a week the whole internet is on huggingface > years pass by, we keep shipping uninterrupted, the Hub team grows a bit (with exceptional people). The community shows up, 1M models, then 3M... > day 2221 (yesterday): surprise all-hands, Jensen Huang joins the call. We're joining NVIDIA.
media 0 共 1 项
#100 🔗 原推
2026-09-04 16:29:33
AK
RT @matth_lapeyre: Today we visited @seeedstudio robotics manufacturing and assembly lines to figure out how to set up lines that can scale Microduck production. BTW, that’s us standing in front of thousands of Reachy Minis ready to ship. https://twitter.com/matth_lapeyre/status/2095831515569483831/photo/1
中文: RT @matth_lapeyre:今天我们参观了@seeedstudio机器人制造和生产线，了解如何建立能够扩展Microduck生产的生产线。 顺便说一句，那就是我们站在成千上万个准备发货的Reachy Minis面前。
media 0 共 2 项 media 1
#99 🔗 原推
2026-09-04 15:18:19
AK
It Takes Two to Match Co-Evolving Generative Retriever with Reinforcement Learning paper: https://huggingface.co/papers/2609.00638 https://twitter.com/_akhaliq/status/2095668374437306641/photo/1
中文: 比赛需要两个 通过强化学习共同进化生成式猎犬 论文:
media 0 共 1 项
#98 🔗 原推
2026-09-04 00:20:50
AK
HarnessDev Can LLMs Create and Evolve Their Own Agent Harness? paper: https://huggingface.co/papers/2609.01437 https://twitter.com/_akhaliq/status/2095668039404728351/photo/1
中文: 哈内斯德夫 LLM 能否创建并进化自己的代理 Harness? 论文:
media 0 共 1 项
#97 🔗 原推
2026-09-04 00:19:30
AK
Repo-To-Skill Distilling GitHub Repositories Into AI4AI Skills paper: https://huggingface.co/papers/2609.02749 https://twitter.com/_akhaliq/status/2095667775830434099/video/1
中文: 从重头到达 将 GitHub 仓库提炼为 AI4AI 技能 论文:
🎬
视频
#96 🔗 原推
2026-09-04 00:18:28
AK
RT @HuggingPapers: SolarWM: Open Data &amp; Scalable Training for Long-Horizon Video World Models A fully open framework for interactive video world models, unifying 1.43M clips across 14 datasets and 4 backbones, enabling real-time rollouts for minutes to hours after training on just 5-second clips. https://twitter.com/HuggingPapers/status/2095425335915598198/photo/1
中文: RT @HuggingPapers:SolarWM:开放数据与功能;可扩展长地平线视频世界模型培训 一款完全开放的交互式视频世界模型框架，将1.43亿次视频片段整合到14个数据集和4个主干中，只需5秒的训练即可实现数小时的实时部署。
media 0 共 1 项
#95 🔗 原推
2026-09-04 00:17:40
AK
RT @HuggingPapers: Can LLMs build and evolve their own agent harness? ByteDance Seed introduces HarnessDev, a benchmark that shifts evaluation from task outputs to the runnable infrastructure itself. Across six creator LLMs and five benchmarks, generated harnesses lag human-engineered ones on code and search — and evolution gains are unstable.
中文: RT @HuggingPapers:LLM 能否构建并改进自己的代理线束? 字节跳动种子引入了HarnessDev，这是一个将评估从任务输出转移到可运行基础设施本身的基准。在六个创作者的LLM和五个测试中，生成的线束在代码和搜索上落后于人类设计的，而进化的收益却不稳定。
media 0 共 1 项
#94 🔗 原推
2026-09-04 00:15:47
AK
RT @xbxbyXREAL: Then: hours on a shared TV. Now: fitting in a game whenever you get the chance. We didn’t outgrow gaming. Life just got busier. xbx a01+ by @XREAL_Global brings a personal big screen to the moments you still have. What does your gaming then vs. now look like? #GamingThenVsNow
#93 🔗 原推
2026-09-04 00:14:40
AK
RT @huggingface: 🤗💚 https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/ https://twitter.com/huggingface/status/2095508870915338284/video/1
🎬
视频
#92 🔗 原推
2026-09-03 15:06:00
AK
RT @julien_c: Super happy to officially announce that we are a̶c̶q̶u̶i̶r̶i̶n̶g̶ joining forces with @nvidia 🔥 Here is a more personal take: AI is at an inflection point. Open source AI can become less relevant in the coming years if the big closed labs run away with it, OR it can become the foundational fabric of the next phase of human civilization. Those are vastly different outcomes, and we need the critical mass to ensure we give our collective best shot to the second outcome. Given @JensenHuang's stance on open source AI and how he stepped up to defend it when it was under threat earlier in the summer, NVIDIA was the only partner we truly considered. HF will remain an independently run, neutral platform. This gives fuel to our long-term vision and mission of unlocking the community's progress to ensure that AI, which is the greatest breakthrough of our lifetime, is accessible to as many people as possible.
中文: RT @julien_c:非常高兴正式宣布我们将与 @nvidia 合作 🔥 以下是一个更个性化的取: 人工智能正处于一个转折点。未来几年，如果大型封闭式实验室倒闭，开源人工智能可能会变得不那么重要，或者它可能成为人类文明下一阶段的基础结构。 这些结果截然不同，我们需要临界质量，以确保我们为第二结果提供最佳结果。鉴于@JensenHuang在开源人工智能问题上的立场，以及他在今年夏天初面临威胁时如何站出来为其辩护，NVIDIA是我们真正考虑过的唯一合作伙伴。 高频将保持独立运行、中性平台。 这为我们实现社区发展进程的长期愿景和使命提供了动力，以确保人工智能——这是我们一生中最大的突破——能够让尽可能多的人能够接触到人工智能。
media 0 共 1 项
#91 🔗 原推
2026-09-03 12:27:03
AK
RT @Gradio: Yeah! It's official now. https://twitter.com/Gradio/status/2095487532196905284/photo/1
中文: RT @Gradio:是的!现已正式宣布。
media 0 共 1 项
#90 🔗 原推
2026-09-03 12:24:21
AK
RT @Thom_Wolf: So happy to finally share the news in person It’s been a wild ride for Hugging Face. We certainly did not anticipate, back in 2016, as a tiny team of scrappy underdogs, that the field would grow so much or that the impact we could have on it would become so massive. I remember @julien_c joking that « code will be a subset of ML » several years ago. The joke turned out to be true, and the pleasure we’ve had being part of this transformation and pushing an alternative vision of AI as open, collaborative and distributed has been and still is immense. We’ve always built things seriously while not taking ourselves too seriously at Hugging Face (special congrats if you find the Hugging Face and Nvidia references hidden in our $12,930,300,000 acquisition price), and we plan to keep doing what we've been doing, just at a much bigger scale, backed by the resources, expertise and drive of Nvidia. And to be clear, nothing changes for our users today. No company in the world has been a more natural fit with our mission than Nvidia. From open-source, open-weights and open science to robotics and AI for science, they have been close partners across everything we care about. So when Jensen offered @ClementDelangue the opportunity to double down on building the Hub as an open, independent and compute agnostic platform, we decided the time was right to start the next 10 years of our journey together. We’re at an important inflection point for open-source AI, where scale and compute are becoming increasingly essential. We’re excited to have the resources to push further, build more ambitiously, and bring you even more projects and news in the coming months.
media 0 共 1 项
#89 🔗 原推
2026-09-03 12:15:09
AK
RT @ClementDelangue: Super happy to share our intention to join forces with NVIDIA in a $12,930,300,000 acquisition 💛💚 10 years after starting Hugging Face, open-source AI is at an inflection point. Thanks to the community, we’ve shown that it can be a complement, and even an alternative, to closed-source APIs. But for it to happen at larger scale, it needs more compute, more support, more collaboration and more visibility. That’s why we went to talk to Jensen, who offered to do exactly that with us. In addition to doubling down on NVIDIA’s massive contributions to open-source AI (I called them the “King of American open-source AI” earlier this year), they’ve committed to strongly supporting Hugging Face and our mission while keeping the platform open, independent and compute agnostic. The founders and the team are all staying to keep pushing this mission forward. Together, we think we can make open source the default way to build AI, with the goal of empowering 100 million AI builders to own their intelligence rather than rent it. Excited about the next 10 years! 🤗🤗🤗
中文: RT @ClementDelangue:非常高兴与英伟达(NVIDIA)联手，以1293030万美元的收购计划💛💚 在启动Hugging Face十年后，开源人工智能正处于一个转折点。得益于社区，我们证明了它可以成为闭源API的补充，甚至是一种替代方案。但要实现更大规模，就需要更多的计算、更多的支持、更多的协作和更多的可见性。这就是我们去和詹森谈话的原因，詹森主动提出要和我们一起做这件事。 除了加大对英伟达对开源人工智能(我今年早些时候称之为“美国开源人工智能之王”)的巨大贡献的投入外，他们还致力于大力支持拥抱面部和我们的使命，同时保持平台开放、独立和计算机不可知论。创始人和团队都将继续致力于推进这一使命。 我们共同认为，开源可以成为构建人工智能的默认方式，旨在赋能1亿名人工智能开发者，使其能够拥有智能，而非出租。 对接下来的10年感到兴奋!🤗🤗🤗
media 0 共 1 项
#88 🔗 原推
2026-09-03 12:06:29
AK
RT @RisingSayak: The only tutorial you should attend at #ECCV26 (kidding 😂). But I am incredibly psyched to be doing this with the best bunch out there. I strongly feel this is a timely tutorial! We will present general approaches alongside our learnings from (post)-training impactful models like Flux2/3. Tutorial website: https://pdm-tut.github.io/ Save your calendars!
media 0 共 1 项
#87 🔗 原推
2026-09-03 05:59:31
AK
RT @matth_lapeyre: Impressed by the precision of this analysis! Fortunately, Reachy Mini’s launch had already taught us a lot about production and supply chains. We also started preparing for Microduck’s mass production well before the announcement: we started working with Seeed on it since May. But the number of orders has been crazy. We weren’t expecting this much enthusiasm, and scaling up this fast with such a complex little robot is a serious challenge. For Reachy Mini, we WERE producing in batches, which made delivery times pretty uneven: sometimes short, sometimes painfully long. The RAM market chaos made things worse by throwing off our suppliers’ timelines. We’re now moving Reachy Mini to continuous production, and that’s the plan for Microduck too. The goal is a steady flow of robots leaving the factory each month, making supply easier to manage and delivery times more predictable for customers. Getting all the injection molds ready is a huge milestone. But the next big challenge is the ramp-up: how quickly can we go from zero to a steady production rate, while making sure every robot is properly assembled, calibrated and tested? Having learned so much from Reachy Mini gives us a head start. It still takes a lot of work to turn that experience into thousands of little robots walking out of the factory. This is going to be another epic adventure for our team and @seeedstudio 🦆
中文: RT @matth_lapeyre:对本分析的精确性印象深刻! 幸运的是，Reachy Mini 的发布已经让我们学到了很多关于生产和供应链的知识。我们在宣布之前就开始为Microduck的量产做准备:我们从5月起就开始与Seed合作。 但订单数量一直很疯狂。我们并没有预料到会有这么大的热情，而用如此复杂的小机器人来快速扩展这个速度是一个严峻的挑战。 对于Reachy Mini，我们分批生产，使得交货时间相当不均衡:有时短，有时疼痛。内存市场的混乱导致情况变得更糟，导致供应商的时间线被抛下。 我们现在正将Reachy Mini推进持续生产，这也是Microduck的计划。目标是每月有机器人持续离开工厂，使客户更容易获得供应管理和交货时间。 准备好所有注塑模具是一个巨大的里程碑。但下一个重大挑战是:我们能以多快的速度从零增长到稳定的生产率，同时确保每个机器人都能正确组装、校准和测试? 从 Reachy Mini 那里学到了很多，为我们带来了一个先发。将这种体验变成成千上万个从工厂里走出来的小机器人，仍然需要大量工作。 这将是我们团队和@seeedstudio的又一次史诗般的冒险🦆
#86 🔗 原推
2026-09-02 16:06:54
AK
RT @andimarafioti: Can you explain to me exactly what you want 4 microducks for?? https://twitter.com/andimarafioti/status/2094781803072467435/video/1
中文: RT @andimarafioti:你能准确地向我解释一下你想要4个微鸭的什么吗?
🎬
视频
#85 🔗 原推
2026-09-01 18:00:39
AK
Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement paper: https://huggingface.co/papers/2608.31046 https://twitter.com/_akhaliq/status/2094841129955176630/photo/1
media 0 共 1 项
#84 🔗 原推
2026-09-01 17:33:40
AK
RT @pablovelagomez1: Alright, I went ahead and built 4DAnyone into a proper @Gradio space. I thought it was cool enough that it deserved the effort and made sure to take advantage of our @rerundotio Gradio integration. TLDR takes in a single video and is able to output multiview consistent videos. Its kinda incredible how good the quality is. Got it working on my 5090 + DGX Spark, so it works with <32GB cards =] Sadly, I had to make this version only output 6 videos; otherwise, it would take way too long, but its able to go all the way up to 48 videos locally. Just takes like ~20-30 minutes. No splatting on this one, as again it would just take too much GPU time that ZeroGPU won't easily support and will eat up all your quota This also feels like the perfect fit for a new Gradio Workflow, something I'll look into at some point. Inference time ~5-8 minutes on the space, sped things up in the video to keep it digestible
🎬
视频
#83 🔗 原推
2026-09-01 17:26:16
AK
RT @datapointai: today, we're releasing the largest open-source human audio preferences dataset, focused on the customer support use-case - 300K+ annotations by real people - 15 SOTA TTS models ranked (Sonic 3.6, Grok TTS, Simba 3.2, Eleven Labs v3) - 8 categories (IVR menus, empathy, escalations, refunds etc) dataset + benchmark + frontier plot below:
中文: RT @datapointai:今天，我们发布了最大的开源人类音频偏好数据集，重点关注客户支持用例 - 真实人群的300K+注释 - 排名的15个SOTA TTS型号(Sonic 3.6、Grok TTS、Simba 3.2、Eleven Labs v3) - 8个类别(IVR菜单、同理心、升级、退款等) 数据集 + 基准图 + 边框图:
🎬
视频
#82 🔗 原推
2026-09-01 17:02:51
AK
RT @tomas_hk: Today we’re releasing our methodology for evaluating model routing with interactive benchmarks, which represent agent cost accumulation better than static benchmarks do. Across leading benchmarks, we achieve Pareto-dominance, exceeding Opus xhigh quality at 20–80% lower cost. https://twitter.com/tomas_hk/status/2094818918963761333/photo/1
中文: RT @tomas_hk:今天我们发布用于使用交互式基准测试来评估模型路由的方法，这些基准比静态基准更能代表代理成本累积。 在领先的基准测试中，我们实现了帕雷托-多伊丁级，以20%至80%的低成本实现了超高质量。
media 0 共 3 项 media 1 media 2
#81 🔗 原推
2026-09-01 16:21:09
AK
RT @viskoai: Today, we are introducing Orbis 1.0, our first Live Model! Create living worlds and stream them in real time, with persistent memory, interactivity, and physics-grounded generation of unbounded length. Try it now at https://www.visko.ai/models#orbis API available via @reactorworld Dynamic version: https://reactor.inc/sandbox?model=visko-orbis-dynamic Stable version: https://reactor.inc/sandbox?model=visko-orbis-stable
🎬
视频
#80 🔗 原推
2026-09-01 16:02:23
AK
RT @Thom_Wolf: seems like it https://twitter.com/Thom_Wolf/status/2094740928019783788/photo/1
中文: RT @Thom_Wolf:似乎在
media 0 共 1 项
#79 🔗 原推
2026-09-01 14:04:51
AK
LoopArena Benchmarking Models as Runtime Controllers for Loop Engineering paper: https://huggingface.co/papers/2608.28281 https://twitter.com/_akhaliq/status/2094511483639963904/video/1
🎬
视频
#78 🔗 原推
2026-08-31 19:43:46
AK
RT @aistasiia: Made a Microduck sticker pack because apparently owning the robot wasn’t enough. Thoughts? https://twitter.com/aistasiia/status/2094388078043390323/photo/1
中文: RT @aistasiia:制作一个Microduck贴纸包，因为显然拥有机器人是不够的。 想法?
media 0 共 1 项
#77 🔗 原推
2026-08-31 19:35:03
AK
RT @ClementDelangue: Over 4 petabytes of models and datasets have been uploaded to HF just last week by AI builders and their agents! That's the equivalent of ~800,000 HD movies. More than ever the storage and collaboration platform for AI!
中文: RT @ClementDelangue:上周，人工智能开发者及其代理已将超过4PB的模型和数据集上传至HF!这相当于大约80万部高清电影。 比以往任何时候都更加是人工智能的存储与协作平台!
#76 🔗 原推
2026-08-31 17:29:04
AK
RT @andimarafioti: MAKE IT SMALLER! MICROER! DUCKIER! https://twitter.com/andimarafioti/status/2094363440152203331/photo/1
中文: RT @andimarafioti:让它更小!微手!杜基!
media 0 共 1 项
#75 🔗 原推
2026-08-31 15:54:17
AK
RT @zizhpan: Weights open: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp
中文: RT @zizhpan:重量开启:
#74 🔗 原推
2026-08-31 14:42:19
AK
RT @ClementDelangue: 10,000+ microducks were pre-ordered in just five days! We weren't expecting that many (it's more than the total number of reachy mini in a year) and it will generate delays so from now on, it will be first-ordered, first shipped to make it fair for everyone! https://twitter.com/ClementDelangue/status/2094421922159210979/photo/1
中文: RT @ClementDelangue:仅五天内就预购了一万多只微鸭! 我们没想到会有太多(超过一年内达到的迷你车总数)，而且会因此产生延迟，因此从现在开始将首次下单，首次发货，让每个人都公平!
media 0 共 1 项
#73 🔗 原推
2026-08-31 14:23:54
AK
RT @NielsRogge: People have reported issues on https://paperswithcode.co/chat to me Apologies! Working on a more scalable architecture Should I just open-source the code base so we can collectively improve it? https://twitter.com/NielsRogge/status/2094323113810907170/photo/1
中文: RT @NielsRogge:人们在 上向我报告了问题 道歉!构建更具可扩展性的架构 我应该开源代码库，以便我们共同改进它吗?
media 0 共 1 项
#72 🔗 原推
2026-08-31 07:40:32
AK
RT @HannesVonEssen: How fast can we make it run?? My current best is 1.6 m/s https://twitter.com/HannesVonEssen/status/2093858085420875928/video/1
中文: RT @HannesVonEssen:我们能以多大速度让它运行?我目前最好的是1.6 m/s
🎬
视频
#71 🔗 原推
2026-08-31 04:47:33
AK
RT @matth_lapeyre: Numbers are way higher now
中文: RT @matth_lapeyre:现在数字更高了
#70 🔗 原推
2026-08-31 04:26:37
AK
Code as Worlds Agentic Discovery of Executable World Representations for Physical Reasoning paper: https://huggingface.co/papers/2608.27549 https://twitter.com/_akhaliq/status/2094278702532358362/photo/1
media 0 共 1 项
#69 🔗 原推
2026-08-31 04:18:47
AK
RT @Gradio: Now You Can Play with Microduck via Voice and Text too! Run all the 9 RL policies that we have actually shipped on Microduck. Ask for a barrel roll, to put on skates, or to kick, and it genuinely follows all the orders 👇👇👇
media 0 共 1 项
#68 🔗 原推
2026-08-31 04:15:57
AK
RT @Thom_Wolf: Microduck reached the Shopify UI limit https://twitter.com/Thom_Wolf/status/2094254389443584383/photo/1
中文: RT @Thom_Wolf:Microduck 达到了 Shopify 的 UI 限制
media 0 共 1 项
#67 🔗 原推
2026-08-31 03:45:18
AK
RT @RemiFabreRobot: We sold over $2,500,000 worth of Microducks in the first 24 hours. Is this one of the most successful robot launches in history? Insane achievement by @antoinepirrone and the team! https://twitter.com/RemiFabreRobot/status/2093465891191042057/photo/1
中文: RT @RemiFabreRobot:我们在前24小时内售出了价值超过250万美元的Microducks。这是历史上最成功的机器人发射之一吗? @antoinepirone 和团队的疯狂成就!
media 0 共 1 项
#66 🔗 原推
2026-08-29 03:12:35
AK
RT @Thom_Wolf: hockey-stick growth
中文: RT @Thom_Wolf:冰棒生长
#65 🔗 原推
2026-08-28 19:02:57
AK
RT @RisingSayak: One plausible future is not enough. A video model may generate a convincing die roll, yet produce the same few outcomes again and again. Our new paper asks a simple question: Does the model capture not only what can happen, but how often? Introducing PAWBench 🧵
media 0 共 1 项
#64 🔗 原推
2026-08-28 15:52:36
AK
RT @NielsRogge: You can try out GLM-5.3 Flash with High reasoning on https://t.co/VcX1LA5ARe, example below Working on improving the latency, currently using @baseten as it's the fastest provider on @huggingface P.S. Find all providers here https://huggingface.co/inference/models?model=zai-org/GLM-5.3-Flash https://twitter.com/NielsRogge/status/2093329958064022011/video/1
🎬
视频
#63 🔗 原推
2026-08-28 15:52:07
AK
RT @Escapation: Welcome to your new simulation Microduck. https://twitter.com/Escapation/status/2093240611008823621/video/1
中文: RT @Escapation:欢迎使用您的新模拟微鸭。
🎬
视频
#62 🔗 原推
2026-08-28 13:32:04
AK
RT @elephant_lumps: Getting all my ducks in a row https://twitter.com/elephant_lumps/status/2093115180741992589/photo/1
media 0 共 1 项
#61 🔗 原推
2026-08-28 13:01:12
AK
RT @NielsRogge: Seeing a major spike in users on Papers with Code What's happening?? https://twitter.com/NielsRogge/status/2093293041905799433/photo/1
中文: RT @NielsRogge:使用代码的纸质纸上用户数量大幅上升 发生了什么?
media 0 共 1 项
#60 🔗 原推
2026-08-28 12:47:41
AK
RT @n0mad_0: that's ducking 1B ARR! 📈
中文: RT @n0mad_0:这太划比了 1B ARR!📈
#59 🔗 原推
2026-08-28 12:29:52
AK
RT @ClementDelangue: someone trained microduck to do somersaults 😭 https://twitter.com/ClementDelangue/status/2093309872733315301/video/1
🎬
视频
#58 🔗 原推
2026-08-28 12:20:22
AK
RT @CongWei1230: Thanks @_akhaliq for sharing our work! How intelligent is your video generation model? Video generators are emerging as backbones for World Action Models (WAMs). VGI-Bench probes the reasoning and action-relevant priors they encode across 27 tasks and 810 instances. https://hexuan21.github.io/VGI-Bench/
中文: RT @CongWei1230:感谢@_akhaliq分享我们的工作! 你的视频生成模型有多智能?视频生成器正成为世界行动模型(WAM)的骨干。 VGI-Bench 负责调查它们在 27 项任务和 810 个实例中编码的推理和动作相关先验。
🎬
视频
#57 🔗 原推
2026-08-28 12:20:05
AK
RT @Thom_Wolf: we've ended at over $2.6M of Microducks ordered in the first 24h
#56 🔗 原推
2026-08-28 12:03:59
AK
VGI-Bench Probing Visual Intelligence in Video Generation Models paper: https://huggingface.co/papers/2608.19583 https://twitter.com/_akhaliq/status/2093154284095295685/photo/1
media 0 共 1 项
#55 🔗 原推
2026-08-28 01:50:44
AK
RT @lvwerra: FYI Microduck is currently generating $2.5B ARR
中文: RT @lvwerra:FYI Microduck 目前正在产生 25 亿美元的 ARR
#54 🔗 原推
2026-08-28 01:40:05
AK
RT @Liu_Zeyi_: I used to feel like companion robot products were kind of a scam. This one changed my mind — cute body, serious engineering inside! Really excited to see progress in this domain 🦆
中文: RT @Liu_Zeyi_:我曾经觉得伴用机器人产品有点像骗局。这个改变了我的想法——可爱的身材，内心严肃的工程!非常期待看到该领域的进展🦆
#53 🔗 原推
2026-08-27 20:46:59
AK
RT @Thom_Wolf: we've just passed $1,000,000 in sales for Microduck
中文: RT @Thom_Wolf:我们刚刚为Microduck售出了100万美元
#52 🔗 原推
2026-08-27 18:05:08
AK
RT @Thom_Wolf: currently selling one Microduck every 5 seconds
中文: RT @Thom_Wolf:目前每5秒销售一次Microduck
#51 🔗 原推
2026-08-27 17:46:44
AK
RT @Thom_Wolf: We have a huge news to share today! Today we are unveiling the first truly accessible RL robot - welcome Microduck A 25 cm tiny open-source biped with 15 actuators and packed with sensors (camera, speaker, LiDAR, NFC, bluetooth, wifi, etc) that you train yourself with reinforcement learning. It's also playable out of the box with more than half a dozen fun and playful pre-trained policies to have it walk, sit, crouch, roller-skate, pick up objects with its articulated beak, and recover on its own. And all for less than $400. See all the details, play with the simulator and order it at: https://pollen-robotics.com/microduck/ (video with sound on 🔊)
🎬
视频
#50 🔗 原推
2026-08-27 17:40:52
AK
RT @pollenrobotics: We built a small biped robot you can teach new tricks to. Train it in simulation, run it on the real thing. Meet Microduck 🦆 $399, shipping before Christmas. https://pollen-robotics.com/microduck https://github.com/pollen-robotics/microduck https://twitter.com/pollenrobotics/status/2092915032052879425/video/1
中文: RT @pollenrobotics:我们打造了一个小型双足机器人，你可以教新技巧。 以模拟训练，在真实事物上运行。认识 Microduck 🦆 399美元，圣诞节前发货。
🎬
视频
#49 🔗 原推
2026-08-27 14:49:31
AK
RT @ClementDelangue: raw footage https://twitter.com/ClementDelangue/status/2092947181124608115/video/1
中文: RT @ClementDelangue:原始视频片段
🎬
视频
#48 🔗 原推
2026-08-27 14:46:12
AK
RT @NielsRogge: You move to Hugging Face Jobs: https://huggingface.co/docs/hub/jobs-overview
中文: RT @NielsRogge:你转至“拥抱面子工作”:
#47 🔗 原推
2026-08-27 14:20:08
AK
RT @ClementDelangue: BIG ANNOUNCEMENT FROM HUGGING FACE TODAY: We're unveiling Microduck 🐥🤖 It's a tiny $399 open-source robot you can teach new tricks with reinforcement learning. It can walk, pick things up, get back up when it falls, and even roller-skate. Welcome to the era of open-source affordable robots to democratize physical AI and world models! 🤗🤗🤗
中文: RT @ClementDelangue:今日大幅宣布面对拥抱: 我们正在推出Microduck 🐥🤖 这是一个售价399美元的小型开源机器人，可以通过强化学习来教授新技巧。它可以走路，捡起东西，跌倒时重新站起来，甚至滑过滚轮滑冰。 欢迎来到开源经济型机器人时代，以普及物理人工智能和世界模型! 🤗🤗🤗
#46 🔗 原推
2026-08-27 14:05:19
AK
RT @ManycoreTech: Say hello to Lux3D 👋 Manycore Tech's new 3D generative model. See it deliver: ⚡ 3D generation in as little as 20s 🎨 High-quality materials and textures ⚙️ Cost-efficient batch creation with Harness Mode Learn more: https://lux3d.aholo3d.com/ #Lux3D #GenAI #3D #ManycoreTech https://twitter.com/ManycoreTech/status/2092946308667789483/video/1
中文: RT @MarycoreTech:向Lux3D打招呼 👋 Maycore Tech 的新型 3D 生成模型。 看到它交付: ⚡ 只需20多代3D 🎨 高品质材料与纹理 ⚙️ 采用光环模式的高效批量制作 了解更多: #Lux3D #GenAI #3D #MaycoreTech
🎬
视频
#45 🔗 原推
2026-08-27 12:18:37
AK
RT @Zai_org: More good news: GLM-5.3’s weights will be released tomorrow. https://huggingface.co/zai-org/GLM-5.3
中文: RT @Zai_org:更多好消息:GLM-5.3 的权重将于明天发布。
#44 🔗 原推
2026-08-27 04:05:16
AK
RT @huggingface: We use Inkling-Small to turn paper abstracts into quick, useful summaries. open weights × open science 🤝 https://twitter.com/huggingface/status/2092727360584135026/photo/1
中文: RT @huggingface:我们使用 Inkling-Small 将纸质摘要转化为快速、有用的摘要。 开放式权重 × 开放科学 🤝
media 0 共 1 项
#43 🔗 原推
2026-08-26 22:19:53
AK
RT @jeffboudier: 💥💥 2 banger open model drops today - (al)ready to deploy on-prem on @Dell &gt; Qwen 3.8 Flash Next from @Alibaba_Qwen &gt; GLM 5.3 Flash from @Zai_org You can host the frontier, you can build your own AI. Congrats team on the release! @ehcalabres @balaatdell https://twitter.com/jeffboudier/status/2092713057026007488/video/1
中文: RT @jeffboudier:💥💥 今天就会下降——(已)在 @Dell 上部署 on-prem Qwen 3.8 闪光 下传来自 @Alibaba_Qwen 来自@Zai_org的GLM 5.3闪存 你可以托管前沿，也可以构建自己的人工智能。 祝贺团队发布!@ehcalabres @balaatdell
🎬
视频
#42 🔗 原推
2026-08-26 20:44:33
AK
Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment paper: https://huggingface.co/papers/2608.23691 https://twitter.com/_akhaliq/status/2092702100270497858/photo/1
中文: 开放世界多智能环境中的自主数学发现 论文:
media 0 共 1 项
#41 🔗 原推
2026-08-26 19:53:55
AK
RT @remi_or_: same day new qwen and new glm drop life feels good in open source https://twitter.com/remi_or_/status/2092632359841792124/photo/1
中文: RT @remi_or_:同日新秋水 在开源中，生活感觉很好
media 0 共 1 项
#40 🔗 原推
2026-08-26 19:34:03
AK
RT @abidlabs: how we are building the best agent-friendly and human-friendly UI for a complex workflows: https://huggingface.co/blog/gradio-workflow-guide https://twitter.com/abidlabs/status/2092674170337550687/video/1
中文: RT @abidlabs:我们如何为复杂的工作流程构建最环保且人性化的用户界面:
🎬
视频
#39 🔗 原推
2026-08-26 19:30:29
AK
RT @aliansarinik: The future of AI belongs to the humans behind it. There’s a common fear that as AI gets better, people get pushed out of the picture. We believe the opposite is happening. AI is creating entirely new categories of work, and an entirely new economy around the people whose knowledge, judgment, and experience are helping these systems improve. Today, tens of thousands of experts are actively contributing to AI training projects through micro1. Over time, we believe this will grow to tens of millions+. And if humans are going to play such a critical role in building the future of AI, the companies they work with should raise the standard for how they’re supported. Today, I’m proud to announce the micro1 Expert Support Program. We’re building a new set of protections, resources, and support for our expert community, starting with: -Expert Bill of Rights: a clear set of commitments outlining what experts can expect when working with micro1. -Rest Credits: paid time away from projects when experts need or want a break. -Expert Emergency Fund: financial support for experts facing emergencies. -Mental Health & Coaching: new resources to support expert wellbeing and growth. -Confidential Support Line: a confidential way to ask questions, seek support, or raise concerns related to pipelines. Within the next two weeks, every expert currently active with micro1 will receive an email with the full program details and launch dates. AI is going to keep getting more capable. The opportunity in front of us is to make sure the people helping build it benefit from that progress too.
🎬
视频
#38 🔗 原推
2026-08-26 16:24:29
AK
RT @victormustar: here we go again: deployed a FREE public endpoint for Qwen3.8-Flash-Next 🚀 (going at +100 tok/s) No token needed, OpenAI-compatible, vision + tool calls, 262K context, thinking from xhigh → off. Light rate limiting, be nice to your neighbors 🤗 4× H200 · FP8 · SGLang cookbook · ~140 tok/s per stream · ~100 tok/s @ 16 concurrent · 0.8s TTFT Guide + chat UI 👇
🎬
视频
#37 🔗 原推
2026-08-26 14:25:31
AK
RT @tonygentilcore: https://x.com/i/article/2092581577637363712
#36 🔗 原推
2026-08-26 13:28:50
AK
RT @abidlabs: Really cool article: how did a cybersecurity risk analyst at a bank end up replicating more than 300 ICML papers and finding dozens of mistakes? Using agents, @TrackioApp, and @HuggingFace: https://www.science.org/content/article/researchers-presented-6000-papers-major-meeting-could-ai-reproduce-their-findings
#35 🔗 原推
2026-08-26 02:07:13
AK
RT @atomic_chat_hq: We officially surpassed 1M model downloads on @huggingface 🔥 Your 5 most downloaded models: 1. Qwen3.8 27B — 261K 2. Ling 3.0 flash — 158K 3. Ornith 1.5 35B A3B — 50K 4. Qwen3.5 4B DFlash — 49K 5. Gemma 4 26B A4B assistant — 48K A big thanks to all of you guys for using our quants, we will keep working hard to improve them even further!
media 0 共 1 项
#34 🔗 原推
2026-08-25 19:26:22
AK
RT @Gradio: Introducing Workflow -- make your AI pipeline the interface. A drag-and-drop canvas where every node runs on its own, every intermediate output stays visible, deploys to Spaces in one command, and the whole graph is also a REST API. Guide: https://huggingface.co/blog/gradio-workflow-guide
#33 🔗 原推
2026-08-25 18:54:43
AK
RT @NielsRogge: You can now also filter paper feeds based on a custom date range! e.g. here I'm checking out Document Understanding papers between 2023-2024 https://twitter.com/NielsRogge/status/2092321242342690967/video/1
🎬
视频
#32 🔗 原推
2026-08-25 18:52:13
AK
RT @SimonShaoleiDu: Our paper is top on HuggingFace Daily now: https://huggingface.co/papers/2608.23283 Also check out our Github Repo on Agent Team Harness: https://github.com/ApodexAI/FrontierAgent
#31 🔗 原推
2026-08-25 17:28:52
AK
RT @NielsRogge: Wrote up a blog on how the search engine of Papers with Code is built! 🔥 This involves a lot of @huggingface tooling > PostGreSQL is used with pgvector > the @Alibaba_Qwen 3 -Embedding-0.6B model is used > hybrid search gets the best results > embeddings are computed on an @nvidia L4 GPU powered by Hugging Face Jobs > artifacts are stored in Buckets > a live embedding endpoint is served on Hugging Face Inference Endpoints > this also powers the "related papers" feature on each paper page Learn more below! https://huggingface.co/blog/pwc-search
#30 🔗 原推
2026-08-25 16:09:04
AK
RT @ClementDelangue: Who's excited? https://huggingface.co/Qwen/Qwen3.8-Flash-Next https://twitter.com/ClementDelangue/status/2092275117539655835/photo/1
media 0 共 1 项
#29 🔗 原推
2026-08-25 15:54:08
AK
Apodex 1.1 Scaling Agentic Intelligence for Complex Work paper: https://huggingface.co/papers/2608.23283 https://twitter.com/_akhaliq/status/2092269176710631758/photo/1
media 0 共 1 项
#28 🔗 原推
2026-08-25 15:13:38
AK
RT @HuggingPapers: ByteDance's TLive-Omni An omni-modal understanding model for e-commerce live streaming, processing images, video, audio, and text. Achieves top results on live-commerce tasks and general benchmarks. https://twitter.com/HuggingPapers/status/2092223219172364658/photo/1
media 0 共 1 项
#27 🔗 原推
2026-08-25 14:36:09
AK
RT @EinsiaAI: The ultimate test for coding agents isn't local editing— it's whole-repo evolution, and right now, the survival rate is 5.4%. Today we’re releasing SWE Refactor Bench, a benchmark for long-horizon, whole-repository software stack migration. Coding agents are getting very good at fixing bugs. But can they refactor an entire system, C → Rust, Maven → Gradle, POSIX → WebAssembly? We built 20 real migrations across projects, including SQLite, zlib, libsodium, and GraphHopper. 520 runs. Only 28 survived all 3 stages. 13/20 tasks were solved by nobody. System-scale migration is still wide open. Full breakdown 👇 GitHub: [https://t.co/xXyLQ3qq0C] Paper Link: [https://t.co/lO1Enh63q7] Einsia Website: [https://t.co/AGn0hF5gwL]
media 0 共 1 项
#26 🔗 原推
2026-08-25 14:34:56
AK
RT @NielsRogge: In case you haven't heard of Buckets on @huggingface, it's a new repo type, besides models/datasets/Spaces, that lets you share and host extremely large files! e.g., the Bucket below is 33TB of data 😱 It's like @aws S3 but optimized for AI workloads, powered by the Xet technology. Learn more here: https://huggingface.co/docs/huggingface_hub/en/guides/buckets
media 0 共 1 项
#25 🔗 原推
2026-08-24 19:23:29
AK
RT @haodongli00: Thanks so much for sharing! @_akhaliq 🥰 The core idea is to learn how the world evolves, rather than directly predict what the future looks like. LDR maps past frames into structured latent states and rolls them forward with kinematic integration, learning only the higher-order motion residuals. To our knowledge, this is the first video world model that learns the underlying dynamics purely from pixels and extrapolates them beyond the training distribution. - Paper: https://huggingface.co/papers/2608.09926 - Code: https://github.com/adobe-research/LDR - Model: https://huggingface.co/haodongli/LDR - Data: https://huggingface.co/datasets/haodongli/LDR
#24 🔗 原推
2026-08-24 18:56:51
AK
Learning How the World Evolves Extrapolative Video World Models via Latent Dynamics Reasoning paper: https://huggingface.co/papers/2608.09926 https://twitter.com/_akhaliq/status/2091958146596041142/video/1
🎬
视频
#23 🔗 原推
2026-08-24 18:37:43
AK
InfinityEdit Infinite Video Editing with a Lightweight Edit-Ignition Adapter paper: https://huggingface.co/papers/2608.20910 https://twitter.com/_akhaliq/status/2091913229437915495/photo/1
media 0 共 1 项
#22 🔗 原推
2026-08-24 15:39:14
AK
RT @HuggingPapers: FACET: Building executable terminal tasks from agent skills while preserving intent and state. FACET reconstructs scenarios, builds the environment first, then aligns instruction, solution, and verifier to that state. Produces 6,078 validated tasks with dense checks. Fine-tuning with 1.2K trajectories improves Terminal-Bench 2.1.
media 0 共 1 项
#21 🔗 原推
2026-08-21 17:51:30
AK
RT @EinsiaAI: 1/ Recursive self-improvement (RSI) depends on agents improving how AI systems are trained —not just tuning hyperparameters, but improving the training algorithm itself. We tested this directly with AI4AI-Bench: 10 real research repositories spanning 10 distinct algorithm families. Full breakdown 👇 GitHub: [https://t.co/s0f0NY7PdQ] Paper Link: [https://t.co/x0qY8wnlwB] Einsia Website:[https://t.co/Rewt4FJwl8] 📊 The results: The average score is just 0.166. Even the best-performing model, Opus 5, reaches only 0.288. The median exploration cost per task rises from $1.69 to $34.60. #AI4AI #RecursiveSelfImprovement #AIResearch #AI4AI_Bench
media 0 共 1 项
#20 🔗 原推
2026-08-21 17:48:49
AK
RT @HuggingPapers: SWE-bench Science A new benchmark for scientific software engineering: 119 tasks across 98 repositories and 20 domains. Even the best agent, Claude Code with Opus-5, achieves under 50% pass@1. https://twitter.com/HuggingPapers/status/2090773411039457342/photo/1
media 0 共 1 项
#19 🔗 原推
2026-08-21 13:06:54
AK
RT @HuggingPapers: Google just released TIPS on Hugging Face A vision-language model with spatial awareness, built for dense understanding tasks like segmentation and depth estimation. https://twitter.com/HuggingPapers/status/2090480235078668615/photo/1
media 0 共 1 项
#18 🔗 原推
2026-08-20 16:58:14
AK
RT @datapointai: today, we’re releasing the largest open-source human image preferences dataset, along with a $1 million data grant - 2M+ annotations by real people - 30 SOTA image models ranked - 10 categories (marketing, product design, anime etc) dataset + benchmark + grant details below: https://twitter.com/datapointai/status/2090462861462065534/video/1
🎬
视频
#17 🔗 原推
2026-08-20 15:36:54
AK
RT @superwhisper: Check out S1-mini on Hugging Face 👇 https://huggingface.co/superwhisper/s1-mini
#16 🔗 原推
2026-08-20 08:58:00
AK
RT @nathanhabib1011: NEW BENCHMARK ON THE @huggingface HUB Your favorite LLM agent can write a script. Can it survive 300+ steps in a terminal without losing the plot? That's what Long-Horizon Terminal-Bench (LHTB) measures, 46 tasks, contamination-resistant, hidden verifiers that check real state, not vibes. 🏆 Leaderboard 🥇 @MiniMax_AI 's Minimax M3 🥈 @Kimi_Moonshot 's Kimi k2.7 Code 🥉 @Zai_org's GLM 5.2 Congrats to @zli12321 and team. Super excited to see how smaller local models perf on this benchmark @0xSero :)
media 0 共 1 项
#15 🔗 原推
2026-08-19 21:45:50
AK
RT @HuggingPapers: Agentic ESOpt: fine-tuning long-horizon LLM agents with minimal GPU memory This framework uses evolution strategies instead of backpropagation, enabling full-parameter optimization with only inference-level memory, and shows strong gains on WebArena-Lite and more. https://twitter.com/HuggingPapers/status/2090109325679140957/photo/1
media 0 共 1 项
#14 🔗 原推
2026-08-19 16:32:43
AK
RT @odysseyml: Today, we're introducing CaliBench, evaluating whether video world models reproduce the true randomness of our universe. Roll a dice or pick a card, the outcomes produced by a world model should match reality. Find out if do! https://odyssey.ml/introducing-calibench
#13 🔗 原推
2026-08-19 15:05:16
AK
RT @HuggingPapers: MOSS-VL: real-time video understanding that perceives while speaking OpenMOSS released MOSS-VL, an 11B open vision-language model that keeps watching live frames while answering — proactive silence and dynamic self-correction built in. https://twitter.com/HuggingPapers/status/2089686442364608855/photo/1
中文: RT @HuggingPapers:MOSS-VL:实时视频理解，即说话时的感知 OpenMOSS 发布了 MOSS-VL，这是一款11B开放式视觉语言模型，可在回答的同时持续观看实时画面——内置主动静音和动态自我纠正。
media 0 共 1 项
#12 🔗 原推
2026-08-18 17:13:54
AK
RT @datapointai: Introducing HumanEvals: open source library for adding real human judgment to your multimodal model evals. Send image, video, or audio outputs. Get pairwise preferences, ratings, or rankings from real people in seconds. https://github.com/impel-intelligence/humanevals https://twitter.com/datapointai/status/2089739361084489904/video/1
中文: RT @datapointai:介绍人形图像:用于为您的多式联运模型eval添加真实人类判断的开源库。 发送图像、视频或音频输出。 几秒钟内从真实人那里获得配对偏好、评分或排名。
🎬
视频
#11 🔗 原推
2026-08-18 17:07:29
AK
RT @HuggingPapers: Large Discovery Models: learning where to search next An LLM proposes, a Bayesian surrogate scores uncertainty, and the loop iterates — cutting validation error by 2.4x, binding energy by 18%, and boosting molecular objectives by 60%+ across programs, proteins, and molecules. https://twitter.com/HuggingPapers/status/2089747716427415901/photo/1
中文: RT @HuggingPapers:大型发现模型:下一步探索 一种LLM提议，贝叶斯代孕者对不确定性进行评分，该循环会将其验证误差降低2.4倍，将能量结合18%，并在项目、蛋白质和分子之间将分子目标提高60%以上。
media 0 共 1 项
#10 🔗 原推
2026-08-18 17:05:22
AK
RT @ClementDelangue: Something super exciting happened quietly on HF over the past month: AI agents became AI builders, and they did it in the open! During our ICML reproduction challenge, 1,221 humans teamed up with coding agents to verify and reproduce 2,226 papers. But here's the cool part: everything happened on the @huggingface hub: 6,816 reproduction logbooks published openly, 2,962 cloud jobs launched, 35,908 claims judged, all traceable, all public and transparent For years the hub has been where humans collaborate on models, datasets and demos. Now we watch agents use it the same way: writing logbooks, publishing results, building on each other's work. Closed labs run evals behind closed doors and ask you to trust the press release. Open science means anyone can check the receipts and now agents can too! The next million users of the hub might not be human. and that might be the best thing to ever happen to science! Full write-up about the hackathon: https://huggingface.co/blog/icml-2026-open-reproductions
media 0 共 1 项
#9 🔗 原推
2026-08-18 17:00:07
AK
RT @OJOaidesign: Today, we're introducing OJO: The first Design Agent Team Workspace. Build your own design agent team, add specialized skills, and turn ideas into product strategy, PRDs, interactive prototypes, and launch-ready designs on one editable canvas. Design is not just how it looks, it's how it works. When everyone can build, taste makes your product stand out. Now, your TASTE can be engineered. #OJO #DesignwithOJO #DesignAgentTeam
🎬
视频
#8 🔗 原推
2026-08-18 15:18:57
AK
RT @huggingface: We've just surpassed 3 million models on the Hub 🤗 the community is accelerating towards an open, distributed future where open AI is everywhere, for everyone 🚀 https://twitter.com/huggingface/status/2089673018737869242/video/1
🎬
视频
#7 🔗 原推
2026-08-18 14:30:28
AK
RT @HuggingPapers: HarnessEval-W: Agentifying the Evaluation of Visual Worlds A new benchmark that brings the harness paradigm to world model evaluation, using specialized sub-agents to produce transparent, auditable reasoning chains for every score. https://twitter.com/HuggingPapers/status/2089626408968491163/photo/1
media 0 共 1 项
#6 🔗 原推
2026-08-18 11:10:45
AK
RT @atlas_cloud_ai: We all know the pain of choosing the right model. You have an idea, testing it one model at a time in ten different tabs, using a spreadsheet to track usable outputs. Stop guessing and start building with Creator Central by Atlas Cloud. Model Explorer runs the same prompt across models and puts the results side by side, up to 10 at once. Start from a curated set or assemble your own. With Wan3.0 coming shortly, we're dropping the price of Seedance 2.5 so creators can compare and choose the right model for their projects. Create something now with our Generator templated workflow: Pick what you like and access the exact prompt and settings behind that result, ready for you to edit. #AIVideo #GenerativeAI #VideoProduction #Seedance #Wan3
🎬
视频
#5 🔗 原推
2026-08-18 11:06:55
AK
RT @NetEaseYouDaoAI: Our Confucius4-TTS paper is now on arXiv — and the open-source model has just received a major upgrade. 🎙️ Confucius4-TTS is built for multilingual, cross-lingual voice generation, with a focus on both model quality and real-world usability. https://arxiv.org/abs/2608.11650
#4 🔗 原推
2026-08-18 08:51:36
AK
RT @ggerganov: let that sink in https://twitter.com/ggerganov/status/2089409881858089102/photo/1
media 0 共 1 项
#3 🔗 原推
2026-08-17 23:01:25
AK
RT @NetEaseYouDaoAI: Our Confucius4-TTS paper is now on arXiv — and the open-source model has just received a major upgrade. 🎙️ Confucius4-TTS is built for multilingual, cross-lingual voice generation, with a focus on both model quality and real-world usability. https://arxiv.org/abs/2608.1165
#2 🔗 原推
2026-08-17 15:01:12
AK
time to follow: https://huggingface.co/MiniMaxAI https://twitter.com/_akhaliq/status/2035756614938972375/photo/1
中文: 时间关注:
media 0 共 1 项
#1 🔗 原推
2026-03-22 16:32:54