In case you didn’t notice: Agent Arena doesn’t have a voting mechanism. So how do we calculate the scores? The answer is causal inference. Agents are multi-stage systems where the orchestrator and harness work together to produce the end result. We developed a method called… https://twitter.com/ml_angelopoulos/status/2064028763697127844/photo/1

Excited to join Stanford CS 153 Office Hours tomorrow: Friday, May 1st at 12PM PT with @AnjneyMidha and @Mabb0tt. Live Q&A. Come say hi - link in thread.

Why GPT-5.5 is lower than Claude? The answer is simple: Code Arena currently only supports frontend/web development tasks, where GPT-5.5 is weakest. Full-stack app development and GitHub integration will land in a couple months. Next time we'll be clearer that this leaderboard… https://twitter.com/ml_angelopoulos/status/2048888792438939707/photo/1

Cool to see this work from @Jsjcl293905 , @davidsimchilevi, and @WillWeiSun deriving efficiency bounds and new estimators for model rankings built on Arena's human preference data. There’s a lot of foundational work to be done to continue to improve the statistical foundations of…

Our team at @arena is solving one of the most important problems in AI: Evaluation. We have some of the top researchers and engineers in the world working together to measure intelligence in the real world and build the infrastructure to ensure it is deployed reliably. Since the… https://twitter.com/ml_angelopoulos/status/2038660899171619279/photo/1

Cool to see @arena on here 💪