1
Causal Tongue-Tie: LLMs Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express
大模型能编码因果方向，但输出却返回反常识答案，揭示隐藏状态与表达的不匹配。
arXiv:2605.25891v1 Announce Type: cross Abstract: We find a mismatch between what large language models encode about a causal question and what they a…