AI300
  1. Jason Weston / Meta Research Scientist
    2671.
    🧮 Principia: Training LLMs to Reason over Mathematical Objects 📐›
    LLM score 92 · 8 months ago
  2. Dan Fu / VP of Kernels at Together
    2672.
    Congrats @cursor_ai on launching Composer 2 today! Really gratifying working with them to make it ready for launch today and to power the Composer 2 Fast endpoint on @togethercompute.›
    LLM score 75 · 8 months ago
  3. Dan Fu / VP of Kernels at Together
    2673.
    I’ve been playing with this today - really great model and release from @cursor_ai (and so fast!)›
    LLM score 72 · 8 months ago
  4. Adnan Esmail / Physical Intelligence Cofounder
    2674.
    We developed a new method at @physical_int called RLT - Reinforcement Learning Tokens - that unlocks performance for deploying our model into real production environments, improving the most precise phases of a task in a couple hours of practice.›
    LLM score 92 · 8 months ago
  5. Sergey Levine / Physical Intelligence Cofounder
    2675.
    Back in Nov we developed Recap and trained π*-06 with RL.›
    LLM score 95 · 8 months ago
  6. Cameron Wolfe / Researcher at Netflix
    2676.
    Super impressive results! Also a great reflection of how training pipelines have evolved for specialized LLMs.›
    LLM score 85 · 8 months ago
  7. Alex Zhang / MIT CSAIL PhD
    2677.
    someone should try having RLMs write REPL code primarily using DSPy
    LLM score 72 · 8 months ago
  8. Tri Dao / Chief Scientist at Together
    2678.
    Nonlinear RNNs seem to do sth genuinely different from attn and linear RNNs/SSMs.›
    LLM score 92 · 8 months ago
  9. Leandro von Werra / Hugging Face Head of Research
    2679.
    Excited to release: AgentUI > a fresh chat interface - natively multi-agent ›
    LLM score 75 · 8 months ago
  10. Jerry Tworek / ex OpenAI VP of RL
    2680.
    AI labs need a wallfacer project. AI researcher not having to explain themselves to anyone.›
    LLM score 82 · 8 months ago
  11. Thomas Wolf / Hugging Face Cofounder
    2681.
    This is really cool. It got me thinking more deeply about personalized RL: what’s the real point of personalizing a model in a world where base models can become obsolete so quickly?›
    LLM score 92 · 8 months ago
  12. Jerry Tworek / ex OpenAI VP of RL
    2682.
    Lots of inconvenience comes from the fact that we’re just very bad at measuring intelligence›
    LLM score 82 · 8 months ago
  13. Skyler Miao / MiniMax Head of Engineering
    2683.
    one of the best claw brains out there. m2.7 really shines on complex agent harness like openclaw›
    LLM score 92 · 8 months ago
  14. Yingru Li / xAI Researcher
    2684.
    the reweighing 1/(1+p) is bounded [1/2, 1] where p is the pass rate [0, 1].›
    LLM score 92 · 8 months ago
  15. Cameron Wolfe / Researcher at Netflix
    2685.
    LLM evals are expensive, so it's common to sub-sample eval data to reduce costs and improve efficiency (especially in iterative tuning / ablation experiments).›
    LLM score 92 · 8 months ago
  16. Albert Gu / Cartesia Chief Scientist
    2686.
    thanks to @cartesia for supporting this project, providing compute, and testing the models!›
    LLM score 85 · 8 months ago
  17. Merve Noyan / Hugging Face ML Engineer
    2687.
    AI2 released new family of vision LMs for pointing (SOTA!) 🔥›
    LLM score 12 · 8 months ago
  18. Ben Burtenshaw / Hugging Face Researcher
    2688.
    tomorrow we are hosting a workshop on agentic evaluations.›
    LLM score 85 · 8 months ago
  19. Merve Noyan / Hugging Face ML Engineer
    2689.
    ScreenSpot-Pro, the GUI computer use benchmark is now on @huggingface 🏆›
    LLM score 85 · 8 months ago
  20. Skyler Miao / MiniMax Head of Engineering
    2690.
    Great observation. We intentionally trained the model to be better at planning and at clarifying requirements with the user.›
    LLM score 92 · 8 months ago
  21. Alex Zhang / MIT CSAIL PhD
    2691.
    I really want to like this paper as an exploration of how to refine an RLM (which I am very much in favor of), but IMO the conclusion is too short-horizon and sort of misses the argument of the original paper.›
    LLM score 85 · 8 months ago
  22. Ronak Malde / DeepMind Researcher
    2692.
    This paper is almost too good that I didn't want to share it›
    LLM score 92 · 8 months ago
  23. Sebastien Bubeck / OpenAI MTS
    2693.
    My good friend Christian Coester solved a 50 years old open problem in self-organizing lists (was it really open? did people even care? yes and yes), and, you guessed it, ChatGPT-Pro made the key step in the proof*! (*a Christian Coester on the other side of the screen was needed›
    LLM score 92 · 8 months ago
  24. Shane Legg / Chief AGI Scientist at DeepMind
    2694.
    Check out this great work on measuring progress towards AGI and the associated global @Kaggle hackathon.›
    LLM score 82 · 8 months ago
  25. Sherwin Wu / OpenAI API, Head of Engineering
    2695.
    It’s never been easier to multitask with Codex! We’re very quickly moving towards a world where it feels like we as humans are now the bottleneck in coding.›
    LLM score 75 · 8 months ago
  26. Alex Zhang / MIT CSAIL PhD
    2696.
    Ran a small eval today on an LM using GPT-5.2 as a judge.›
    LLM score 92 · 8 months ago
  27. Richard Song / DeepMind Research Scientist
    2697.
    Absolutely beautiful. LLMs are both theoretically and practically Turing complete https://t.co/C66kEafCs3
    LLM score 75 · 8 months ago
  28. Lewis Tunstall / Hugging Face Researcher
    2698.
    Neat, reminds me of the reasoning cache by @ianwu97 that we used to train QED-Nano https://t.co/pJ99JjwMol https://t.co/wnI2APmpks https://t.co/14Pe93FiuB
    LLM score 85 · 8 months ago
  29. John Carmack
    2699.
    The corporate advisory boards that I have been a part of have almost exclusively been “vibe checks”, where presentations are made about work the company is doing, and the advisory panel chats about things for a while.›
    LLM score 75 · 8 months ago
  30. Sherwin Wu / OpenAI API, Head of Engineering
    2700.
    We're now in a world where you can get 54.4% SWE-Bench PRO performance, and 60% (!) T-Bench 2.0 performance:›
    LLM score 92 · 8 months ago
More