AI300
  1. Jerry Tworek / ex OpenAI VP of RL
    1.
    Spending lots of times these days thinking about human eval curves
    LLM score 35 · about 1 hour ago
  2. Jerry Tworek / ex OpenAI VP of RL
    2.
    Always the case. Attention is a surprisingly powerful and useful inductive bias.›
    LLM score 65 · about 2 hours ago
  3. Merve Noyan / Hugging Face ML Engineer
    3.
    btw I genuinely enjoy mech interpret papers from Anthropic›
    LLM score 65 · about 11 hours ago
  4. Merve Noyan / Hugging Face ML Engineer
    4.
    my favorite interval timer went paid 🥲 so I asked hugging chat @huggingface to quickly make one, took 6 secs and used DeepSeek V4.1 Flash›
    LLM score 15 · about 13 hours ago
  5. Yingru Li / xAI Researcher
    5.
    It's a subtle issue, but highly worth fixing! https://t.co/Vx8gg3QFVL
    LLM score 65 · about 17 hours ago
  6. Sergey Levine / Physical Intelligence Cofounder
    6.
    I had a fun conversation with @adityaag at SPC a few weeks ago.›
    LLM score 15 · 1 day ago
  7. Sander Dieleman / DeepMind Research Scientist
    7.
    Interesting contrarian take! While everyone in diffusion LMs is seemingly trying to figure out how to get rid of self-conditioning, they asked: what if self-conditioning is all you need?›
    LLM score 72 · 1 day ago
  8. Sander Dieleman / DeepMind Research Scientist
    8.
    The magic of flow map language models is that they capture dependencies between tokens in just 1-2 model passes.›
    LLM score 72 · 1 day ago
  9. Chelsea Finn / Physical Intelligence Cofounder
    9.
    Generative value classifiers are a new parametrization of Q-functions that don't require action conditioning.›
    LLM score 78 · 1 day ago
  10. Horace He / Thinking Machines Founding Engineer
    10.
    I'm doing DP. I'm doing TP. I'm doing SP.›
    LLM score 15 · 1 day ago
  11. Christian Szegedy / ex xAI Cofounder
    11.
    Yep: increasing formal verification of all code will be essential.›
    LLM score 45 · 1 day ago
  12. Clive Chan / OpenAI Hardware
    12.
    Kimi intentionally pacing the open weights cyber frontier is a big win.›
    LLM score 68 · 1 day ago
  13. Sander Dieleman / DeepMind Research Scientist
    13.
    Interesting perspective on the increasing importance of sequence parallelism in LLM inference.›
    LLM score 65 · 1 day ago
  14. Cyris Kissane / Researcher at Flapping Airplanes
    14.
    imagine posting this and making a select few projects a target...›
    LLM score 35 · 2 days ago
  15. Christian Szegedy / ex xAI Cofounder
    15.
    This is a clear demonstration of what is still missing from AI-generated movies; it is not the visuals or cinematics, but the additional individual pieces that make a real creation memorable and unique.›
    LLM score 65 · 2 days ago
  16. Eric Jang / ex VP of AI at 1X Robotics
    16.
    woah, surprised this wasn't found sooner! https://t.co/Da3rCZIYGT
    LLM score 35 · 2 days ago
  17. Andrew White / Edison Scientific Cofounder
    17.
    - senior red teamer - links to k8s repo - m5sum of some prediction›
    LLM score 15 · 2 days ago
  18. Yuhuai (Tony) Wu / xAI Cofounder
    18.
    Wow https://t.co/Falvs2sp5W
    LLM score 15 · 2 days ago
  19. Sander Dieleman / DeepMind Research Scientist
    19.
    Fresh take on diffusion distillation: the student learns mean flow by matching the teacher with the sum of a straight-line prediction and a curved zero-integral path at random sub-intervals.›
    LLM score 72 · 2 days ago
  20. Dan Fu / VP of Kernels at Together
    20.
    Looped models are cool - you can scale training FLOPs independently from model size - but memory scales with the # of recurrences from the KV cache.›
    LLM score 82 · 2 days ago
  21. Kai / OpenAI Researcher
    21.
    After a model used DNS to access live internet during training in September, we again paused frontier RL training and inference for 11 days while we further red-teamed our research infrastructure with fleets of agents.›
    LLM score 72 · 2 days ago
  22. Albert Gu / Cartesia Chief Scientist
    22.
    to prepare for an upcoming release (👀) we dug into the pretokenizer a little more.›
    LLM score 72 · 2 days ago
  23. John Carmack
    23.
    I always thought it was a shame that modest default stack setups made alloca() a generally-frowned-upon technique in C/C++.›
    LLM score 65 · 3 days ago
  24. Asher Spector / Cofounder of Flapping Airplanes
    24.
    Great being on @SquawkCNBC today.›
    LLM score 35 · 3 days ago
  25. Lewis Tunstall / Hugging Face Researcher
    25.
    With Carbon-X and llama.cpp, you can literally chat with your DNA at home https://t.co/bJDQbzU12j
    LLM score 25 · 3 days ago
  26. Thomas Wolf / Hugging Face Cofounder
    26.
    The Jensen effect 🦆 launch day: about 3,000 MicroDucks sold›
    LLM score 35 · 3 days ago
  27. Merve Noyan / Hugging Face ML Engineer
    27.
    if you're a start-up or SME chances of you getting access to cyber programs of these big AI companies are low and you're vulnerable›
    LLM score 65 · 3 days ago
  28. Demis Hassabis / CEO of DeepMind
    28.
    Honoured to be featured in the @WSJ Innovators Issue.›
    LLM score 15 · 3 days ago
  29. Lewis Tunstall / Hugging Face Researcher
    29.
    POV https://t.co/GVonf1jYWO https://t.co/defMj1NRye
    LLM score 25 · 3 days ago
  30. John Carmack
    30.
    It is unfortunate that so much of the robotics world uses 50 hz joint targets, while the video data collection is at 60 hz.›
    LLM score 72 · 3 days ago
More