X · @teortaxesTex
· X / Twitter
> If you were distilling, why would you 'distill' only the final answer and not the reasoning? The whole purpose of stealing reasoning traces is to le…
> If you were distilling, why would you 'distill' only the final answer and not the reasoning? The whole purpose of stealing reasoning traces is to learn to reason, isn't it? Doesn't make sense, at all.Uh? Because you need examples of correct, well-formatted answers in the data?GDP: This paper DOES NOT PROVE the DISTIL