1
TRACE: Toulmin-based Reasoning Assessment through Constructive Elements for LLM CoT Evaluation
用图尔敏模型拆解LLM推理过程，评估逻辑结构而非只看答案，为复杂推理评价提供新思路
arXiv:2605.29656v1 Announce Type: new Abstract: Evaluating open-ended outputs from large language models (LLMs) remains challenging due to the absence…