AI300
  1. Ben Burtenshaw / Hugging Face Researcher
    1.
    yea tokenizers! super insightful thread on how and why a model's tokenizers can be changed without pretaining.›
    LLM score 82 · about 4 hours ago
  2. Eric Jang / ex VP of AI at 1X Robotics
    2.
    I enjoyed the Jensen + @dwarkesh_sp podcast.›
    LLM score 75 · about 5 hours ago
  3. Albert Gu / Cartesia Chief Scientist
    3.
    a dynamical systems point of view, which looks like an SSM applied along the residual stream, informs more principled ways to scale looped architectures https://t.co/Ry09zXCXY1
    LLM score 92 · about 11 hours ago
  4. Ronak Malde / DeepMind Researcher
    4.
    Ultra long horizon benchmarks are the next frontier,›
    LLM score 82 · about 11 hours ago
  5. Alex Zhang / MIT CSAIL PhD
    5.
    We need more super hard tasks to properly eval our models!›
    LLM score 92 · about 13 hours ago
  6. Sherwin Wu / OpenAI API, Head of Engineering
    6.
    Of all the launches in Codex today, this one was the most game-changing for me – computer use, but without taking over your whole screen!›
    LLM score 92 · about 13 hours ago
  7. Hyung Won Chung / Research Scientist at Meta
    7.
    Health care was built for a world where humans were the only interpreters of health data.›
    LLM score 92 · about 14 hours ago
  8. Chelsea Finn / Physical Intelligence Cofounder
    8.
    LLM post-training used to mean fine-tuning to a downstream task›
    LLM score 92 · about 14 hours ago
  9. Karol Hausman / Physical Intelligence Cofounder
    9.
    Today we're releasing our newest model π0.7: https://t.co/lk5H3Yv7Ru›
    LLM score 92 · about 14 hours ago
  10. Boris Cherny / Creator of Claude Code
    10.
    Dogfooding Opus 4.7 the last few weeks, I've been feeling incredibly productive.›
    LLM score 92 · about 15 hours ago
  11. Ben Burtenshaw / Hugging Face Researcher
    11.
    crucial new benchmark dropped for long reasoning we need this because even frontier models still can't think long even with 1M context.›
    LLM score 92 · about 15 hours ago
  12. Tri Dao / Chief Scientist at Together
    12.
    The dynamical system view gives very clean conditions for looped transformer to be stable https://t.co/md8fispK0Q
    LLM score 92 · about 15 hours ago
  13. Merve Noyan / Hugging Face ML Engineer
    13.
    new open-source Bonsai models are out 🔥 > ternary weights in 8B (1.75 GB), 4B (0.86 GB), and 1.7B (0.37 GB)›
    LLM score 25 · about 15 hours ago
  14. Sergey Levine / Physical Intelligence Cofounder
    14.
    We finished evaluating π0.7, our new model at Physical Intelligence.›
    LLM score 95 · about 15 hours ago
  15. Andrew White / Edison Scientific Cofounder
    15.
    Been using Opus 4.7 for a bit on vulnerability testing (mostly because of the unrelated mythos hype).›
    LLM score 92 · about 15 hours ago
  16. Jason Wei / AI Researcher at Meta
    16.
    Beautifully written piece by @FAbnousi about how AI for health might look like in the future›
    LLM score 85 · about 16 hours ago
  17. Boris Cherny / Creator of Claude Code
    17.
    Opus 4.7 feels more intelligent, agentic, and precise than 4.6.›
    LLM score 92 · about 17 hours ago
  18. Sander Dieleman / DeepMind Research Scientist
    18.
    A while back we had the "rotation trick" to improve VQ bottlenecks (https://t.co/w4XVeN9L0J), now we have DiVeQ, which seems to improve codebook coverage quite significantly.›
    LLM score 92 · about 18 hours ago
  19. Sayak Paul / Hugging Face Researcher
    19.
    Working at Hugging Face over the past 3.5+ years has allowed me to identify what technical areas truly interest me!›
    LLM score 65 · about 18 hours ago
  20. Boris Cherny / Creator of Claude Code
    20.
    Opus 4.7 is in Claude Code today.›
    LLM score 92 · about 18 hours ago
  21. Ben Burtenshaw / Hugging Face Researcher
    21.
    port any transformers model to mlx https://t.co/FtS6gnpVID
    LLM score 92 · about 18 hours ago
  22. Sander Dieleman / DeepMind Research Scientist
    22.
    FlexTok/Semanticist provided an elegant recipe to learn semantically coarse-to-fine sequence representations of images.›
    LLM score 92 · about 18 hours ago
  23. Damek Davis / Assoc. Professor Wharton Stats
    23.
    I wanted to understand the proof and ended up making a video using opus 4.6.›
    LLM score 92 · about 19 hours ago
  24. Sander Dieleman / DeepMind Research Scientist
    24.
    This work provides theoretical grounding for some of the design decisions (cross-entropy loss, learnable embeddings, self-conditioning, entropy-based schedule) in CDCD (https://t.co/B6tYr2AOUk), and brings it into the modern era.›
    LLM score 92 · about 19 hours ago
  25. Lewis Tunstall / Hugging Face Researcher
    25.
    New video from the HF journal club: we discuss Apple's self-distillation paper 🍎›
    LLM score 92 · about 19 hours ago
  26. Merve Noyan / Hugging Face ML Engineer
    26.
    this is the year of 3D World models 🔥 > Lyra 2.0 by NVIDIA: image → 3D world with Gaussians, 14B params, built on WAN-14B ›
    LLM score 85 · about 23 hours ago
  27. Michael Elabd / DeepMind Researcher
    27.
    Continual learning for MLE just hit a new milestone 🥁›
    LLM score 92 · 1 day ago
  28. Cameron Wolfe / Researcher at Netflix
    28.
    Strongly recommend the LLM-as-a-Verifier writeup.›
    LLM score 92 · 1 day ago
  29. John Carmack
    29.
    It is generally frowned upon to have LLMs precisely regurgitate part of their training set, but it is an interesting question how you could use LLM training to nearly losslesly compress a huge corpus like the entirety of the Internet Archive.›
    LLM score 92 · 1 day ago
  30. Dan Fu / VP of Kernels at Together
    30.
    📢 Super excited to announce Parcae! We've been thinking about scaling laws and the "right" way to get more FLOPs.›
    LLM score 92 · 1 day ago
More