AI300
  1. Leandro von Werra / Hugging Face Head of Research
    1921.
    LLMs are generally not great at this but Opus 4.8 seems to be regressing on CAD tasks compared to previous generations: https://t.co/w3Lw17Hnrc
    LLM score 65 · 5 months ago
  2. Alex Zhang / MIT CSAIL PhD
    1922.
    Last thing and then I'll shut up because people on this website genuinely have trouble 1) reading and 2) remembering anything:›
    LLM score 25 · 5 months ago
  3. Sayak Paul / Hugging Face Researcher
    1923.
    We did a big release of Kernels, featuring important updates!›
    LLM score 45 · 5 months ago
  4. Ben Burtenshaw / Hugging Face Researcher
    1924.
    in student mode, this debate around renderers and message-in-tokens-out vs tokens-in-token-out is super interesting and I've learnt a lot from both sides of the argument.›
    LLM score 72 · 5 months ago
  5. Michael Elabd / DeepMind Researcher
    1925.
    5 days of Trajectory starts today!! https://t.co/s9tqtdgIc6
    LLM score 25 · 5 months ago
  6. Ronak Malde / DeepMind Researcher
    1926.
    5 days. 5 announcements. 5 days of Trajectory starts today.›
    LLM score 35 · 5 months ago
  7. Alex Zhang / MIT CSAIL PhD
    1927.
    most shocking part of this blogpost tbh on a more serious note, super valuable insight on mark's experience with AI for systems for AI and where it will be headed and a great first @CoreAutoAI blog.›
    LLM score 45 · 5 months ago
  8. Clive Chan / OpenAI Hardware
    1928.
    >hundreds of parallel subagents gonna hit your quota in like 3 seconds https://t.co/AwEPRRRLvj
    LLM score 35 · 5 months ago
  9. Vahid Kazemi / ex MTS at xAI
    1929.
    Me: @goal Implement these 3 items.›
    LLM score 35 · 5 months ago
  10. Ethan Shen / Ai2 Researcher
    1930.
    Raspberry Pi for AI is kinda dope https://t.co/ott8w1XvJT
    LLM score 25 · 5 months ago
  11. Clive Chan / OpenAI Hardware
    1931.
    Improving CPU speed by 10x should not affect training speed essentially at all.›
    LLM score 72 · 5 months ago
  12. Alex Zhang / MIT CSAIL PhD
    1932.
    In case you're curious about why dynamic workflows are so powerful and the future, read the RLM paper! Opus 4.8 + dynamic workflows in Claude Code is perhaps the first instance of a frontier model seriously trained to be an RLM.›
    LLM score 72 · 5 months ago
  13. Wes Griffin / Researcher at Flapping Airplanes
    1933.
    Awesome stuff from @JonSaadFalcon @Avanika15 and the team!! https://t.co/32bk9Z95L1
    LLM score 25 · 5 months ago
  14. Dan Fu / VP of Kernels at Together
    1934.
    Cool stuff! DC inference is supply bound - makes sense to offload intelligence locally when you can! https://t.co/rI0nKQFrzA
    LLM score 25 · 5 months ago
  15. Albert Gu / Cartesia Chief Scientist
    1935.
    Our new model Ink-2 tops AA's leaderboard for streaming speech-to-text!›
    LLM score 35 · 5 months ago
  16. Sander Dieleman / DeepMind Research Scientist
    1936.
    Great advice that also applies to writing blog posts! https://t.co/MIdY7bwABP
    LLM score 65 · 5 months ago
  17. Boris Cherny / Creator of Claude Code
    1937.
    Claude Opus 4.8 is out today. It's our strongest coding model yet: up on SWE-bench Pro (from 64.3 to 69.2) and noticeably more honest about its own work.›
    LLM score 42 · 5 months ago
  18. John Schulman
    1938.
    Glad to be advising https://t.co/QVsH1u528N, which uses AI to help authors and reviewers do deeper, more thorough analysis than unaided humans could practically do.›
    LLM score 65 · 5 months ago
  19. Sander Dieleman / DeepMind Research Scientist
    1939.
    Diffusion models are recurrent neural networks🧐 https://t.co/vq525ylrM7 https://t.co/H9Rq1A9I4i
    LLM score 65 · 5 months ago
  20. Lewis Tunstall / Hugging Face Researcher
    1940.
    Today's best coding models from Qwen, DeepSeek, Minimax etc are trained on 1000s of concurrent RL environments to simulate SWE tasks in real git repos.›
    LLM score 82 · 5 months ago
  21. Noam Brown / OpenAI Research Scientist
    1941.
    After AlphaGo, the skill of human Go players noticeably improved.›
    LLM score 72 · 5 months ago
  22. Sebastien Bubeck / OpenAI MTS
    1942.
    The point of a breakthrough is that suddenly it makes a lot of things that seemed impossible possible.›
    LLM score 72 · 5 months ago
  23. Mehtaab Sawhney / OpenAI for Science
    1943.
    A remarkable paper appeared on arXiv tonight by Thomas Bloom, Will Sawin, Carl Schildkraut and Dmitrii Zhelezov.›
    LLM score 72 · 5 months ago
  24. Andrew White / Edison Scientific Cofounder
    1944.
    Naming stuff is hard and I empathize with Evoscale/CZI/Bio[h]ub group on this.›
    LLM score 15 · 5 months ago
  25. Andrew White / Edison Scientific Cofounder
    1945.
    Be brave. Let's see that y-axis label https://t.co/XOZScuWXgR
    LLM score 15 · 5 months ago
  26. Andrew White / Edison Scientific Cofounder
    1946.
    Can someone give me the tldr on ESMFold vs ESMFold2? They also did an atlas release with ESMFold years ago.›
    LLM score 25 · 5 months ago
  27. Chelsea Finn / Physical Intelligence Cofounder
    1947.
    How can VLAs achieve 95+% reliability? Using RL post-training with EXPO-FT:›
    LLM score 78 · 5 months ago
  28. Damek Davis / Assoc. Professor Wharton Stats
    1948.
    It would be great if the algorithms curriculum in CS depts was reorganized around the GPU.›
    LLM score 65 · 5 months ago
  29. Keller Jordan / OpenAI Researcher
    1949.
    Modded-NanoGPT optimization result #19 (2026/05/08): @wen_kaiyue has achieved 3125 steps using KLSOAP-Hyperball, matching the record.›
    LLM score 25 · 5 months ago
  30. Jerry Tworek / ex OpenAI VP of RL
    1950.
    GEMM Processing Units https://t.co/OhFFLzdh1D
    LLM score 25 · 5 months ago
More