1 paper · 1 filter
Yiguo Wang, Ziyuan Yang, Yi Zou +3
Verifying multi-step LLM reasoning requires more than determining whether a trace is correct: a useful verifier should identify where the reasoning first goes wrong. However, exist…