For everyone who says “modern incentives forbid publishing negative results,” let this stand as a counterexample!
Why do you think it's a negative result? The table on page 9 shows great results.
I think it's a pun. AlphaZero? AlphaNegative.
Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
11–20 of 20 posts
Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#12Warning: abuse of this technique may cause the model to go blind.
Is this a joke about wanking?
Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#13sometimes i feel like the whole self-play thing is kinda the obvious path now but still nuts seeing it actually work better than huge data dumps. you ever wonder how much of progress is just crazy good pipelines versus actual breakthroughs?
Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#14Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#15To be clear, this is not a model trained on zero data, this is a pretrained model (Qwen 2.5 trained on 18 trillion tokens) finetuned using self-generated data grounded by a Python interpreter
The breakthrough here is eliminating the need for human-labeled reasoning data while still achieving SOTA results, which has been a major bottleneck in developing reasoning capabilities.
Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#16"Despite using zero human-curated data, AZR achieves state-of-the-art results on diverse coding and math reasoning benchmarks, even outperforming models trained on large in-domain datasets. This demonstrates the potential for sophisticated reasoning skills to emerge purely through self-play without domain-specific supervision."
> "sophisticated reasoning skills"
Does it mean that it uses the data it has to the maximum possible level to produce new reasoning (that add to those produced by less algorithms). IOW, are we still in the realm of: with a given data set, A.I. can produce up to N reasoning capabilities and consequently, can't produce more than that ? IOW, reasoning is bound by knowledge ? And therefore, maybe we could just start from a data/knowledge set in which we add some randomness and self play until some form of reasoning emerge ?
Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#17Be nice to see some of these run on languages the pretrained model is a little less good at than Python and JS.
Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#18"Despite using zero human-curated data, AZR achieves state-of-the-art results on diverse coding and math reasoning benchmarks, even outperforming models trained on large in-domain datasets. This demonstrates the potential for sophisticated reasoning skills to emerge purely through self-play without domain-specific supervision."
> "sophisticated reasoning skills" Does it mean that it uses the data it has to the maximum possible level to produce new reasoning (that add to those produced by less algorithms). IOW, are we still in the realm of: with a given data set, A.I. can produce up to N reasoning capabilities and consequently, can't produce more than that ? IOW, reasoning is bound by knowledge ? And therefore, maybe we could just start from…
Up to N at a time probably. Then move on using them. The problem is the longer the chain, the more likely it will deviate from the reality. It will include non-obvious atomic decisions and wrong assumptions. This will make the whole thing unstable. I.e. without strict human supervision it likely will start producing crap. Probably some self double checks can help, but still. On the other hand humans aren't that smart either...
Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#19Re: Absolute Zero: Reinforced Self-Play Reasoning with Zero Data
#20Related to this: has anyone seen a model respond with “oh wait I was wrong…” when you follow-up with a “can you explain why this answer is right?” I still find that my uses of GPT and others still struggle with a sort of tunnel vision.
I saw ChatGPT do that within a single response once (only once). It started giving an answer and made a mistake, and then apologized and corrected it, all within a single response.