AI300
  1. Cameron Wolfe / Researcher at Netflix
    61.
    I’m happy to share that I’ve been promoted to Staff Research Scientist at Netflix! It's hard to overstate how rewarding my journey at Netflix has been.›
    LLM score 15 · 7 days ago
  2. Damek Davis / Assoc. Professor Wharton Stats
    62.
    Codex took less than a week to formalize these 41000 lines and find the error.›
    LLM score 95 · 7 days ago
  3. Jim Fan / NVIDIA Director of Robotics
    63.
    This is pure nightmare fuel. Identity theft of the past would be nothing compared to what vibe agents can do.›
    LLM score 92 · 7 days ago
  4. Jonathan Ross / TPU Creator
    64.
    A pilot operates $100M in equipment and nobody blinks.›
    LLM score 82 · 7 days ago
  5. Leandro von Werra / Hugging Face Head of Research
    65.
    Auto-research for ML training models is all the rage now, but underrated is: auto-research for data!›
    LLM score 85 · 7 days ago
  6. Damek Davis / Assoc. Professor Wharton Stats
    66.
    'Proved' something new and had codex formalize it lean.›
    LLM score 92 · 7 days ago
  7. Andrej Karpathy / AI researcher
    67.
    Software horror: litellm PyPI supply chain attack.›
    LLM score 97 · 7 days ago
  8. Jerry Tworek / ex OpenAI VP of RL
    68.
    Todays AIs have a taste of maximally bland and median appeal RLHF.›
    LLM score 72 · 7 days ago
  9. Merve Noyan / Hugging Face ML Engineer
    69.
    my fav papers from 3DV because why not 🤝 MapAnything by Meta›
    LLM score 92 · 7 days ago
  10. Sayak Paul / Hugging Face Researcher
    70.
    Last year, I got to collaborate on a number of serious projects at the intersection of Diffusers x optimization ⚡️›
    LLM score 75 · 8 days ago
  11. Boris Cherny / Creator of Claude Code
    71.
    Little known fact, the Anthropic Labs team (the team I joined Anthropic to be on) shipped:›
    LLM score 85 · 8 days ago
  12. Cameron Wolfe / Researcher at Netflix
    72.
    Interesting observation on instruction following behavior induced by preference tuning versus RLVR.›
    LLM score 92 · 8 days ago
  13. Yang Chen / Nvidia Research Scientist
    73.
    We released Nemotron Cascade 2 30B A3B. What makes this release especially meaningful to me is that it reflects a 1.5-year journey at NVIDIA around one core idea: improving AI math reasoning through self-improvement at test time.›
    LLM score 92 · 8 days ago
  14. Jim Fan / NVIDIA Director of Robotics
    74.
    Teleop is so 2025. Ever since we unveiled EgoScale and the dexterity scaling law, it's been clear to us and the ecosystem that behavior cloning directly from humans is the way to break the curse of teleop.›
    LLM score 92 · 8 days ago
  15. Merve Noyan / Hugging Face ML Engineer
    75.
    "train rt-detrv2 on mobile-ui-design" is all it takes to train an object detector 🔥 ›
    LLM score 92 · 8 days ago
  16. Horace He / Thinking Machines Founding Engineer
    76.
    A conversation I've had several times lately: Them: "LLMs are so good for learning whatever you want — there's never been a better time to learn!"›
    LLM score 92 · 8 days ago
  17. Jason Weston / Meta Research Scientist
    77.
    🌐Unified Post-Training via On-Policy-Trained LM-as-RM🔧›
    LLM score 92 · 8 days ago
  18. Sharon Zhou / AMD VP of AI
    78.
    codegen is cheap now. performance isn’t: most generated kernels are kinda mid.›
    LLM score 92 · 8 days ago
  19. Lewis Tunstall / Hugging Face Researcher
    79.
    You can now pretrain LLMs entirely on the HF Hub 💥›
    LLM score 92 · 8 days ago
  20. Zixuan Li / Lead Z.ai
    80.
    Last time Chinese models missed the SWE-rebench top 10, they got slammed for "benchmaxing." Now, GLM-5 is back in a very "interesting" spot.›
    LLM score 82 · 8 days ago
  21. Yoshua Bengio
    81.
    Evidence of deceptive behavior has already appeared in widely used AI systems, and the risk is expected to grow as AI becomes more capable, more autonomous, and more embedded in everyday decision-making.›
    LLM score 92 · 8 days ago
  22. Ben Burtenshaw / Hugging Face Researcher
    82.
    share you agent harness on the hub. built a hf cli extension that backs up and shares coding agent workspaces on the hub.›
    LLM score 92 · 8 days ago
  23. Jonathan Ross / TPU Creator
    83.
    AI will split companies into two camps. Value Preservation — using AI to cut costs.›
    LLM score 12 · 8 days ago
  24. Sayak Paul / Hugging Face Researcher
    84.
    We released 🤗 Kernels 0.12.3 to support Flash-Attention 4! This means we now support kernels written in `cutlass.cute`.›
    LLM score 85 · 8 days ago
  25. Skyler Miao / MiniMax Head of Engineering
    85.
    We’ve pushed performance quite a bit here — M2.7 is showing strong results across both MLE and semi-autonomous RL.›
    LLM score 92 · 9 days ago
  26. Skyler Miao / MiniMax Head of Engineering
    86.
    This is an outstanding project.›
    LLM score 72 · 9 days ago
  27. Skyler Miao / MiniMax Head of Engineering
    87.
    Yes, we performed extensive optimizations and even established a dedicated benchmark for it.›
    LLM score 85 · 9 days ago
  28. Cameron Wolfe / Researcher at Netflix
    88.
    Very well-written post. Whether you think takeoff is imminent or not, it's worth deeply considering this trajectory and the bottlenecks that could arise.›
    LLM score 92 · 9 days ago
  29. Sharon Zhou / AMD VP of AI
    89.
    Mood: agents optimizing kernels Claude won on kernel optimization: gemm_bf16 at 1.19x vs Codex's 0.94x.›
    LLM score 92 · 9 days ago
  30. Alex Zhang / MIT CSAIL PhD
    90.
    soon all benchmarks of this form will be obsolete :) https://t.co/gzwPimFngT https://t.co/Hr1GH9COyS
    LLM score 72 · 9 days ago
More