New benchmark catches AI reasoning errors in chemistry tasks

Original: From Answers to States: Verifiable Process-Level Evaluation of Chemical Reasoning in Large Language Models

Source: arxiv.org ↗