> The shorter "thinking" is, the less is the probability of it going astray
As long as the error introduced by more steps is less than the compounding error of sub-optimal token sampling, I would expect a better result.
I think your choice of "wrong" is extreme, suggesting such a token can catastrophically spoil the result. The modern reality is more that the model is able to recover.
As long as the error introduced by more steps is less than the compounding error of sub-optimal token sampling, I would expect a better result.
I think your choice of "wrong" is extreme, suggesting such a token can catastrophically spoil the result. The modern reality is more that the model is able to recover.