pama

18 ms·

1. 301.▲
   Understanding R1-Zero-Like Training: A Critical Perspective(github.com)

   160 points by pama 2y ago|21 comments
2. 302.▲
   Fin-R1: A Large Language Model for Financial Reasoning Through RL(arxiv.org)

   2 points by pama 2y ago|1 comments
3. 303.▲
   Compute Optimal Scaling of Skills: Knowledge vs. Reasoning(arxiv.org)

   2 points by pama 2y ago|0 comments
4. 304.▲
   Tapered Off-Policy Reinforce: Stable and Efficient RL for LLMs(arxiv.org)

   2 points by pama 2y ago|0 comments
5. 305.▲
   by pama 2y ago

   I would be curious to know if it would be possible to recunstruct approximate versions of popular common subsets of internet training data by using many different LLMs that may have happened to read the same info. Anyone knows pointers to m
6. 306.▲
   by pama 2y ago

   I know 3D is in the roadmap. Once the basic functionality is in place, it would be great to also consider integrating molecular visualization or at least provide enough fast primitives to simplify the integration of molecular visualization
7. 307.▲
   by pama 2y ago

   One big difference from most of the other companies is that George Church works in this company now.
8. 308.▲
   by pama 2y ago

   The reality is much simpler than what is often presented about science geniuses. Lorentz and Poincare had the math down to explain early experiments and the Maxwell’s equations predictions of the constancy of the speed of light, and Einstei
9. 309.▲
   by pama 2y ago

   Thanks.
10. 310.▲
    by pama 2y ago

    Can you elaborate on this point: “ We discovered that meaningful performance improvements, as high as 10–15%, can be achieved with as few as 16 training examples.” In particular, did you need to change the hyperparameters much, and did this
11. 311.▲
    by pama 2y ago

    Maybe strict compliance is not as big a threat for academic institutions. A wall ot click-through text does not technically limit liabilities against good lawyers on the other side, so you might need better measures to ensure compliance in
12. 312.▲
    by pama 2y ago

    Not GP, but in my opinion the reason why the license restriction is so important is that otherwise very few people will be able to try it. Huggingface or other commercial providers cannot put it up online (even if they didnt charge for the
13. 313.▲
    Mice have been genetically engineered to look like mammoths(economist.com)

    4 points by pama 2y ago|1 comments
14. 314.▲
    by pama 2y ago

    I wonder why it doesn’t mention aider, the earliest practical tool for vibe coding.
15. 315.▲
    by pama 2y ago

    Did you notice how much better things are today (eg Claude Sonnet 3.7) than they were 1 year ago? Don’t you expect things will not improve in the next year? Even R1, a public weights model, can add huge value when left to code in a loop.
16. 316.▲
    GPT-4.5: "Not a frontier model"?(interconnects.ai)

    166 points by pama 2y ago|161 comments
17. 317.▲
    by pama 2y ago

    I would contribute to a repo that works towards this goal.
18. 318.▲
    by pama 2y ago

    Glad it helps. Claudio Arrau thought of the fingers as “dead” with all the piano playing happening using larger muscles. It is a struggle to internalise this concept in some Chopin etudes (op 10, no 2 especially) but it almost immediately h
19. 319.▲
    by pama 2y ago

    This was a nice demonstration of letting these robotic hands learn to play a keyboard. The technique is limited due to constraints of the hands, and is closer to parts of the technique of early keyboard instruments like an organ or a harpsi
20. 320.▲
    by pama 2y ago

    I feel like a kid in a candy shop. Some of these tricks would take way too long to reverse engineer correctly based on the papers. I hope that the releases this week start a renaissance of the use of MoE as baseline academic models.
21. 321.▲
    by pama 2y ago

    Nanoparticle as a term has a specific technical definition that does not include a random gas made of organic small molecules, which include most molecules we perceive by smell.
22. 322.▲
    by pama 2y ago

    With the explosion of LLMs we are close to the final thoughts in her chapter: “this planet may be filled with beings who will be blind and deaf as the inmost rock, yet will execute changes as delicate and complicated as those of human langu
23. 323.▲
    by pama 2y ago

    Could this be abstracted enough with the right macros to make a subset of useful lisp commands play well with the shell? It could be a powerful way to extend the shell for interactive use.
24. 324.▲
    by pama 2y ago

    It is good to occasionaly display windows from a remote linux onto a macbook client. I wouldnt say I use it regularly but it certainly helps.
25. 325.▲
    by pama 2y ago

    Yeah I think this was a wrong puzzle to try according to: https://sudoku.com/sudoku-solver A bummer.
26. 326.▲
    by pama 2y ago

    Thanks for spotting this. The solution is indeed wrong. And I agree that the machine can regurgitate plausible reasoning in principle. If it run in a loop, I would bet that it could probably figure this particular problem out eventually, bu
27. 327.▲
    by pama 2y ago

    Here is o3-mini on a simple sudoku. In general the puzzle can be hard to explore combinatorially even with modern SAT solvers, so I picked one marked as “easy”. It looks to me like it solved it but I didnt confirm beyond a quick visual insp
28. 328.▲
    by pama 2y ago

    I am sure DeepSeek did optimize the inference cost of R1. They did not yet release an efficient MoE downscaling of it, ie an R1-mini.
29. 329.▲
    by pama 2y ago

    I wonder if the authors can explain the aparent inconsistency between what we now know about R1 and their statement “They don’t engage in logical reasoning” from the first lesson. My simple-minded view of logical reasoning by LLMs is that t
30. 330.▲
    by pama 2y ago

    Every time someone claimed “emerging” behavior in LLMs it was exactly that. I can probably count more than 100 of these cases, many unpublished, but surely it is easy to find evidence by now.