1
Breaking the Chain: A Causal Analysis of LLM Faithfulness to Intermediate Structures
用因果分析揭示大模型对中间推理结构的忠实度，为提升AI可信度提供新视角。
arXiv:2603.16475v2 Announce Type: replace Abstract: In schema-guided reasoning (SGR) pipelines, LLMs produce explicit intermediate structures -- rubri…