AI300
  1. Vahid Kazemi / ex MTS at xAI
    991.
    IMO coding benchmarks aren't capturing the flaws of LLMs in coding.›
    LLM score 92 · 3 months ago
  2. Keller Jordan / OpenAI Researcher
    992.
    Modded-NanoGPT optimization result #12: Transferring good hparams from recent NorMuon records -- in particular, taking final val 25 steps early following @wen_kaiyue's NorMuonH, and lr=0.035 following Liming Liu's NorMuon -- improved the Muon baseline by 50 steps.›
    LLM score 95 · 3 months ago
  3. Keller Jordan / OpenAI Researcher
    993.
    Modded-NanoGPT optimization result #11: @nilinabra has achieved a new record of 3225 steps (-25) via a novel technique dubbed Contra-Muon, in which top SVD components are somewhat suppressed.›
    LLM score 92 · 3 months ago
  4. Jerry Tworek / ex OpenAI VP of RL
    994.
    This is unfortunately mostly a bearish take on RL https://t.co/bbUle9yq3B
    LLM score 75 · 3 months ago
  5. Cyris Kissane / Researcher at Flapping Airplanes
    995.
    I have an existence proof of a minimum of 384k layers https://t.co/wDMOGLmtPO
    LLM score 92 · 3 months ago
  6. Jerry Tworek / ex OpenAI VP of RL
    996.
    Laws of physics of companies are generally simple but specific for each industry.›
    LLM score 65 · 3 months ago
  7. Jack Clark / Anthropic Cofounder
    997.
    Gotta hand it to him - Juergen Schmidhuber had some amazing papers in the 2010s - early stuff on handwriting recognition (https://t.co/xus8EKhNd8), computer vision (https://t.co/R2saXFHMTy), suggesting intelligence be benchmarked via video games (https://t.co/MpupPARwQe), etc.
    LLM score 85 · 3 months ago
  8. Cameron Wolfe / Researcher at Netflix
    998.
    Great point about the utility of multi-teacher on-policy distillation and why this approach might be preferable to just including multiple domains in RL training.›
    LLM score 92 · 3 months ago
  9. Juntang Zhuang / MTS at xAI (pre-training lead)
    999.
    I left xAI earlier this year after two incredible years.›
    LLM score 75 · 3 months ago
  10. Andrew White / Edison Scientific Cofounder
    1000.
    Is there a legit complete IND publicly available on the internet anywhere? Like the 1k page document? It would be really instructive to see what these look like.
    LLM score 82 · 3 months ago
  11. Cameron Wolfe / Researcher at Netflix
    1001.
    Really interesting work. I like the idea of finding simple ways to encourage better routing between experts that adapts naturally based on the task / domain.›
    LLM score 82 · 3 months ago
  12. Merve Noyan / Hugging Face ML Engineer
    1002.
    read our response here https://t.co/oA7DAMyJql I wish journalists can do better in the future wrt avoiding baiting https://t.co/85whKvIJGZ
    LLM score 85 · 3 months ago
  13. Sander Dieleman / DeepMind Research Scientist
    1003.
    Great thread about multimodal models. Just because a single Transformer can do it all, that doesn't mean doing things this way makes the most sense economically.›
    LLM score 92 · 3 months ago
  14. Yoshua Bengio
    1004.
    Thank you to @robertwiblin for inviting me on the @80000Hours podcast to discuss the research progress we’re making at @LawZero_ to create safe-by-design AI systems.›
    LLM score 85 · 3 months ago
  15. Sayak Paul / Hugging Face Researcher
    1005.
    Together w/ the community, our initiative of profiling Diffusers pipelines & potentially improving them is going very strong 💪›
    LLM score 92 · 3 months ago
  16. Zhaocheng Zhu / Nvidia Research Scientist
    1006.
    With agents, coding habits become less about the abstraction, but more about the scaffold around the agents.›
    LLM score 92 · 3 months ago
  17. Sherwin Wu / OpenAI API, Head of Engineering
    1007.
    The new reasoning audio model is getting a lot of attention, but GPT-Realtime-Translate might be even cooler.›
    LLM score 85 · 3 months ago
  18. Sherwin Wu / OpenAI API, Head of Engineering
    1008.
    Our very first speech model that uses reasoning is now live! ›
    LLM score 92 · 3 months ago
  19. Aidan McLaughlin / OpenAI Research Scientist
    1009.
    ai becoming so powerful it interrupts market forces is discussed ad nauseam (birth of new species, postcapital, etc) but less discussed is the market becoming so powerful it interrupts ai flowering (tools not persons, egoless ai, etc)
    LLM score 92 · 3 months ago
  20. Sebastien Bubeck / OpenAI MTS
    1010.
    The top 10 Erdos problems by @thomasfbloom are really cool, completely different level from what we have seen so far with AI!›
    LLM score 85 · 3 months ago
  21. Aidan McLaughlin / OpenAI Research Scientist
    1011.
    agents are freeing you to hang out with your kids AND they make computer more fun›
    LLM score 75 · 3 months ago
  22. John Carmack
    1012.
    Space launch was a clear case where there was a large difference in efficiency between what was possible and what was done in practice before SpaceX.›
    LLM score 75 · 3 months ago
  23. Emmanuel Ameisen / Anthropic Interpretability Researcher
    1013.
    Interpreting model activations is important to understand why a model is doing what its doing.›
    LLM score 85 · 3 months ago
  24. Albert Gu / Cartesia Chief Scientist
    1014.
    Introducing a new sequence model Raven which pushes the boundary of fixed-state-size sequence models!›
    LLM score 92 · 3 months ago
  25. Eric Jang / ex VP of AI at 1X Robotics
    1015.
    I think of robot demos as being "upper bound" (how good is the best take / video clip shared on social media) and "lower bound" (how good is it under adversarial disturbances or unseen scenarios).›
    LLM score 92 · 3 months ago
  26. Cameron Wolfe / Researcher at Netflix
    1016.
    What are some good blogs / papers that cover different compaction techniques and their relative performance? This blog does a great job of covering various options and their downsides: https://t.co/nNISXsl0SE›
    LLM score 92 · 3 months ago
  27. Julian Schrittwieser / Anthropic Researcher
    1017.
    Very excited to be partnering with @elonmusk @SpaceX!›
    LLM score 75 · 3 months ago
  28. Jack Clark / Anthropic Cofounder
    1018.
    Very nice post pushing back on my "RSI 60% by EOY 2028" claim.›
    LLM score 92 · 3 months ago
  29. Asher Spector / Cofounder of Flapping Airplanes
    1019.
    lots of fun to chat with @bfspector about how new systems can enable new algorithms! We think opening up what the GPU can do gives a lot more room to explore new ideas :) https://t.co/mCrluPCjGX
    LLM score 75 · 3 months ago
  30. Kushal Thaman / Researcher at Flapping Airplanes
    1020.
    sneak peek at the Sequoia AI Ascent on what we're up to! including compiler crimes against GPUs in pursuit of godlike data efficiency :) https://t.co/i2a4IzTo5N
    LLM score 92 · 3 months ago
More