cool pipeline for analyzing lots of screenshot data 🖼️ we need good tools to understand how we interact w/ complex algos

you never read the same codebase twice

Check out our new work: Generalization from context often outperforms generalization from finetuning. And you might get the best of both worlds by spending extra compute at train-time.

seems like a big theme lately (e.g. also "RL for Reasoning w/ One Training Example") is that approaches don't get nearly enough bang for each training point's buck - cool!

40% with just 1 try per task: SWE-agent-LM-32B is the new #1 open source model on SWE-bench Verified. We built it by synthesizing a ton of agentic training data from 100+ Python repos. Today we’re open-sourcing the toolkit that made it happen: SWE-smith.

NaN sample efficiency x.com/AndrewZ4573249…

fun note: heiner once described my env config as "the final boss of python venv issues" -- has been mostly issue free for a few months now, thanks mostly to uv 🤞

It turns out that a lot of the most interesting behavior of LLMs can be explained without knowing anything about architecture or learning algorithms. Here we predict the rise (and fall) of in-context learning using hierarchical Bayesian methods.

building reasoning agents w/ Yuchen He Qian Huang was so fun, and the next paradigm will be even cooler -- agents will solve far harder problems far faster

CollabLLM won #ICML2025 ✨Outstanding Paper Award along with 6 other works! icml.cc/virtual/2025/a… 🫂 Absolutey honored and grateful for coauthors Microsoft Research Stanford AI Lab and friends who made this happen! 🗣️ Welcome people to our presentations about CollabLLM tomorrow

🚀 Excited to share that the Workshop on Mathematical Reasoning and AI (MATH‑AI) will be at NeurIPS 2025! 📅 Dec 6 or 7 (TBD), 2025 🌴 San Diego, California

x.com/i/article/1954…

thank you Igor Babuschkin for all your inspiration and vision for xAI -- we're all lucky you cared so deeply about a world where AI benefits humanity

✨ Optimas is fully open-sourced at github.com/snap-stanford/… 🏋🏻 Contribute by: 1) Submitting PRs for your own compound AI systems. 2) Extending the optimization infra. 3) Creating issues (we're happy to help optimize your systems, explore why they work or fail, and go from there

🚨 Yanzhe Zhang's new work uses a search-based framework that evolves both sides of the game to find privacy risks --- alternating between improving attacker & defender instructions in privacy-critical LLM agent interactions!!! Attack agents evolve from blunt data grabs to

I am (slowly) re-reading the Tolkien legendarium (of which Lord of the Rings is a small part). The whole body of work is so incredible and there's nothing else like it... it dilutes other worlds of fiction. Wait - your story doesn't have a comprehensive history/mythology spanning