AI300
  1. Keller Jordan / OpenAI Researcher
    2281.
    @yonashav This fact makes the present seem less interesting.›
    LLM score 20 · 7 months ago
  2. Emmanuel Ameisen / Anthropic Interpretability Researcher
    2282.
    We actually understand very few mechanisms inside LLMs, so the ones we do are a treat to explore.›
    LLM score 85 · 7 months ago
  3. Karol Hausman / Physical Intelligence Cofounder
    2283.
    Thanks - this was fun Tobi brings a very grounded perspective on how learning from experience translates to long-horizon robot performance in the real world.›
    LLM score 65 · 7 months ago
  4. Noam Brown / OpenAI Research Scientist
    2284.
    @MikCampanelli @mobiuspoker Making a poker solver is out of distribution, not poker itself.›
    LLM score 75 · 7 months ago
  5. Igor Babuschkin / Cofounder of xAI
    2285.
    @DavidSHolz Most APIs implement speculative decoding where a small model or a simple heuristic is used to predict a few tokens at a time.›
    LLM score 80 · 7 months ago
  6. Noam Brown / OpenAI Research Scientist
    2286.
    @giffmana It could be a combo of traditional spam/phishing detection and LLM detection
    LLM score 30 · 7 months ago
  7. Igor Babuschkin / Cofounder of xAI
    2287.
    @jadenitripp That’s where continual learning comes in.›
    LLM score 30 · 7 months ago
  8. Noam Brown / OpenAI Research Scientist
    2288.
    A family friend recently lost $1,000 to a phishing email.›
    LLM score 30 · 7 months ago
  9. Igor Babuschkin / Cofounder of xAI
    2289.
    I suspect the reason Claude Code doesn’t work as well for large codebases is that they post-trained it mostly on smaller repos (big corp sized repos are rare).›
    LLM score 80 · 7 months ago
  10. Sholto Douglas / Researcher at Anthropic
    2290.
    @deredleritt3r I still think it requires massive effort to not end up in that world - e.g.›
    LLM score 10 · 7 months ago
  11. Andrej Karpathy / AI researcher
    2291.
    New post: nanochat miniseries v1 The correct way to think about LLMs is that you are not optimizing for a single specific model but for a family models controlled by a single dial (the compute you wish to spend) to achieve monotonically better results.›
    LLM score 80 · 7 months ago
  12. Chelsea Finn / Physical Intelligence Cofounder
    2292.
    Studying generalist reward models is hard: robot datasets focus on successful demos, not failures.›
    LLM score 95 · 7 months ago
  13. Noam Brown / OpenAI Research Scientist
    2293.
    Got this DM: I appreciate that you posted this - increasingly my twitter feed feels out of whack, especially with people claiming Claude Code makes them 1000000x more efficient.›
    LLM score 30 · 7 months ago
  14. Andrej Karpathy / AI researcher
    2294.
    @thecsguy Slope of slope
    LLM score 70 · 7 months ago
  15. Jonas Geiping / AI Researcher in Tübingen
    2295.
    @ziv_ravid Yeah I think forecasting is one of the few domains where we can really evaluate reasoning under uncertainty, and this may be a bit of contrarian take, I do think Bayesian reasoning is a key capability that is not learnt from yet another code/math RL environment
    LLM score 80 · 7 months ago
  16. Jonas Geiping / AI Researcher in Tübingen
    2296.
    We just published a new open-source model that we trained with RL to be capable of open-ended forecasting!›
    LLM score 20 · 7 months ago
  17. Andrej Karpathy / AI researcher
    2297.
    @ChrisPainterYup All of the following are true: I came out sounding too critical in that moment on the pod and accidentally created a misrepresenting sound bite, I was positively surprised by Opus 4.5 over the break, and my overall change of mind overall is less dramatic than naively implied.
    LLM score 20 · 7 months ago
  18. Noam Brown / OpenAI Research Scientist
    2298.
    @zachcpa I'd give the win to Codex due to the way better optimizations.›
    LLM score 20 · 7 months ago
  19. Noam Brown / OpenAI Research Scientist
    2299.
    @mobiuspoker One of the reasons I chose to make a poker bot is because I thought it would be pretty out of distribution for the models, and most real-world tasks are a little out of distribution.›
    LLM score 70 · 7 months ago
  20. Demis Hassabis / CEO of DeepMind
    2300.
    We’re making great progress with our Gemini Robotics work in bringing AI to the physical world - a critical aspect of AGI.›
    LLM score 40 · 7 months ago
  21. Noam Brown / OpenAI Research Scientist
    2301.
    @swyx I don't remember the prompt but the initial prompt wasn't very detailed.›
    LLM score 20 · 7 months ago
  22. Noam Brown / OpenAI Research Scientist
    2302.
    @_thomasip I think it's similar to the hallucination problem.›
    LLM score 20 · 7 months ago
  23. Noam Brown / OpenAI Research Scientist
    2303.
    This solver is a project I've wanted to do for a while.›
    LLM score 40 · 7 months ago
  24. Noam Brown / OpenAI Research Scientist
    2304.
    I vibecoded an open-source poker river solver over the holiday break.›
    LLM score 75 · 7 months ago
  25. Igor Babuschkin / Cofounder of xAI
    2305.
    @mmt_lvt I personally find Scala too complex.›
    LLM score 20 · 7 months ago
  26. Igor Babuschkin / Cofounder of xAI
    2306.
    @y0b1byte I haven’t encountered any issues with writing proc macros so far with Opus.›
    LLM score 20 · 7 months ago
  27. Igor Babuschkin / Cofounder of xAI
    2307.
    Opus 4.5 is surprisingly good at writing decent Rust code and I expect this will make the language much more popular than it already is.›
    LLM score 40 · 7 months ago
  28. Igor Babuschkin / Cofounder of xAI
    2308.
    @DavidSHolz There are decades where nothing happens, and there are weeks where decades happen
    LLM score 20 · 7 months ago
  29. Sander Dieleman / DeepMind Research Scientist
    2309.
    Nice thread on the return of x-prediction in diffusion models.›
    LLM score 75 · 7 months ago
  30. Andy Jones / Anthropic Research Engineer
    2310.
    me: 10k tokens/hour (thinking mostly) @ 100W -> 10kWh/megatoken›
    LLM score 85 · 7 months ago
More