[marxplank]

that would help with decidable problems but would still be not generalisable for problems with non trivial rewards, or ones with none.

[astrange]

Reasoning seems to generalize, insofar as o1 and DeepSeek-R1 are better at answering questions than their base models.