[kristov]

Doesn't prompting "think step by step" simply increase the probability the model will generate its output in the style of places where it saw "step by step" before (eg: tutorial content)? It's not really "deciding" to think step by step. Maybe I am mistaken.

[pegasus]

It's more than just that. The intermediary steps act as a kind of working memory, ensuring that the model is less likely to get lost in the weeds.

[hobs]

You're not mistaken, its basically perturbing the seed for which the sentence grows.