1 paper
Zhuohan Wang, Haoran Ma, Tianyu Wu +3
Large language models (LLMs) can solve every intermediate step of a multi-step math problem on its own and still fail the full problem, even when given a roadmap of the steps and a…