AI300
  1. Andrew Lampinen / Research Scientist at DeepMind
    3361.
    New paper studying how language models representations of things like factuality evolve over a conversation.›
    LLM score 85 · 9 months ago
  2. Sayak Paul / Hugging Face Researcher
    3362.
    While working on MaPO, we did something interesting: we reformulated the task of personalization as a preference alignment problem.›
    LLM score 85 · 9 months ago
  3. Cameron Wolfe / Researcher at Netflix
    3363.
    Trinity large is very sparse (400B-A13B, 256 experts w/ 4 active per token).›
    LLM score 85 · 9 months ago
  4. Roland Gavrilescu / ex MTS at xAI
    3364.
    Models haven’t been post-trained on progressive disclosure yet.›
    LLM score 92 · 9 months ago
  5. John Carmack
    3365.
    #PaperADay 13 2020: DREAM TO CONTROL: LEARNING BEHAVIORS BY LATENT IMAGINATION›
    LLM score 85 · 9 months ago
  6. Asher Spector / Cofounder of Flapping Airplanes
    3366.
    strong agree. this is why I'm not a trader ;) forgive my exaggeration, I just mean that's it *is* mostly a next token prediction problem that happens to print money.›
    LLM score 65 · 9 months ago
  7. Cyris Kissane / Researcher at Flapping Airplanes
    3367.
    I use muon when making decisions to minimize my regret Adam and SGD just weren't fast enough.
    LLM score 70 · 9 months ago
  8. Kushal Thaman / Researcher at Flapping Airplanes
    3368.
    I spent a bunch of time a year ago thinking about the data wall.›
    LLM score 85 · 9 months ago
  9. Mehtaab Sawhney / OpenAI for Science
    3369.
    I've recently gone on leave from Columbia to join OpenAI, working on OpenAI for Science.›
    LLM score 75 · 9 months ago
  10. Ben Burtenshaw / Hugging Face Researcher
    3370.
    We got Claude to teach open models how to write CUDA kernels.›
    LLM score 85 · 9 months ago
  11. alphaXiv
    3371.
    BIG new idea in interpretability called Patterning›
    LLM score 75 · 9 months ago
  12. John Carmack
    3372.
    #PaperADay 12 2019: Learning Latent Dynamics for Planning from Pixels (PlaNet)›
    LLM score 85 · 9 months ago
  13. Boris Cherny / Creator of Claude Code
    3373.
    @charles_irl In case it’s not clear in the docs: - Ancestor https://t.co/v4FOLUBHz9’s are loaded into context automatically on startup›
    LLM score 20 · 9 months ago
  14. Lucas Beyer / Meta Researcher
    3374.
    Another nice work with student-teacher setup, but this time with meta learning setup: teacher generates q/a pairs for student, and gets reward depending on student's improvement after training on those qa.›
    LLM score 75 · 9 months ago
  15. alphaXiv
    3375.
    "LLM-in-Sandbox Elicits General Agentic Intelligence"›
    LLM score 85 · 9 months ago
  16. Ethan Shen / Ai2 Researcher
    3376.
    @unsorsodicorda @saurabh_shah2 @Tim_Dettmers GA is trained using 4.5 Air as the teacher, whereas SERA-32B uses GLM 4.6.›
    LLM score 85 · 9 months ago
  17. Lucas Beyer / Meta Researcher
    3377.
    I really like this one, it's a neat idea for self-improvement: distill token-level from the teacher is the student who was allowed to peek at the solution first.›
    LLM score 80 · 9 months ago
  18. Tim Dettmers / Research Scientist at Ai2
    3378.
    his work was mostly the genius of Ethan Shen.›
    LLM score 70 · 9 months ago
  19. Tim Dettmers / Research Scientist at Ai2
    3379.
    Our method became so efficient (26x vs RL; 57x vs other synth gen), that we could easily generate 1000s of trajectories for a single repo.›
    LLM score 85 · 9 months ago
  20. Tim Dettmers / Research Scientist at Ai2
    3380.
    This is very impactful: you can now distill frontier performance into small models that are specialized to private repositories.›
    LLM score 75 · 9 months ago
  21. Tim Dettmers / Research Scientist at Ai2
    3381.
    From there we could do a massiv amounts of experiments and really understand what matters for training coding agents.›
    LLM score 80 · 9 months ago
  22. Cameron Wolfe / Researcher at Netflix
    3382.
    Continual learning is a popular topic in LLM research, but it might not be as far away as we think.›
    LLM score 65 · 9 months ago
  23. Ethan Shen / Ai2 Researcher
    3383.
    Finally, we conduct an analysis of variance across SWE-Bench runs.›
    LLM score 95 · 9 months ago
  24. Ethan Shen / Ai2 Researcher
    3384.
    We also experiment with mixing data from rollouts and discover that training on SERA’s first and second rollouts complement each other to increase performance in data constrained regimes.›
    LLM score 85 · 9 months ago
  25. Ethan Shen / Ai2 Researcher
    3385.
    Even more interesting is the fact that the best data comes from truncated trajectories that just barely exceed the context limit.›
    LLM score 85 · 9 months ago
  26. Niklas Muennighoff / AI Researcher at Stanford
    3386.
    Community-built open benchmarks work really well, e.g., Terminal-Bench, HLE, MMTEB.›
    LLM score 80 · 9 months ago
  27. John Carmack
    3387.
    #PaperADay 11 Discovering state-of-the-art reinforcement learning algorithms›
    LLM score 80 · 9 months ago
  28. Noam Brown / OpenAI Research Scientist
    3388.
    Had to cut this one for space: 2019: AI can't create art—creativity is uniquely human
    LLM score 20 · 9 months ago
  29. Andrej Karpathy / AI researcher
    3389.
    @0xabi96 It feels like I’m cheating.›
    LLM score 70 · 9 months ago
  30. Andrej Karpathy / AI researcher
    3390.
    @ChiragLathiya The nearest neighbor really is some kind of a junior engineer.›
    LLM score 80 · 9 months ago
More