zlacker

>>nyrikk+(OP)
> we then train models on noisy, corrupted traces which have no relation to the specific problem each is paired with, and find that not only does performance remain largely consistent with models trained on correct data, but in some cases can improve upon it

This is the interesting part. We've probably all had the experience where the model is going off the rails during the thinking process but somehow spits out the right answer at the end. Apparently the reasoning doesn't even need to be correct during training?

I guess it suggests to me that the reason CoT helps is that the model gets more compute to think internally, not that the words it produces are meaningful. I'm surprised nobody has come up with a good scheme for adaptive compute per token yet. Maybe we can skip CoT entirely.

>>modele+T5
> We've probably all had the experience where the model is going off the rails during the thinking process but somehow spits out the right answer at the end. Apparently the reasoning doesn't even need to be correct during training?

How do we know if the reasoning was correct or not? Do we have more information about what the model was thinking besides just what it says it was thinking?

>>trehal+p8
It's definitely not explicitly writing out everything it's "thinking" if you are considering all dimensions of the latent space that are connected, that can't really be exhibited with a sentence.

CoT builds on existing prompt engineering techniques by adding it to reinforcement learning to force the models to build their own CoT prompt essentially. So it's not what it's thinking but all indications are that it does guide the reasoning abilities of LLMs through the output distribution.