模型推理错误时，因果链大多不忠实，现有检测方法失效。
Two Regimes of Chain-of-Thought Unfaithfulness: Metric-Based Detection Fails Where Models Are Wrong
- 按答案正确性分层分析，发现检测分为两个截然不同的阶段。
- 69%的不忠实推理出现在错误答案上，此时所有信号均无法有效检测。
- 行为检测在正确答案上表现尚可，但对错误答案完全失效，且指标反向相关。
链式思维（CoT）解释仅在忠实的情况下才具备监督价值：推理过程必须真实产生答案。通过对比行为检测与 FaithCoT-Bench 的人工标注，我们发现答案正确性贯穿问题的各个层面。仅凭答案错误（作为理想诊断工具，非可部署检测器）的判断，其性能已优于所有专门设计的信号（AUROC 0.696），因为69%的标注不忠实情况发生在错误答案上。按正确性分层后，检测呈现两种不同模式：在正确答案中，行为信号能中等程度区分忠实推理与事后补全（AUROC 0.63–0.67）；而在大多数不忠实推理集中的错误答案上，所有测试信号均无法显著优于随机水平（在四个模型上均复现）。标准的步骤移除度量反而与人工标签负相关，这一反转现象在基准释放分数及依赖提示的反事实标注轨迹中均可复现。线性探针可在 Llama-3.1-8B 中解码行为盲区模式，在 Qwen-2.5-7B 中解码正确答案模式，但未发现跨模式共享的正向对齐方向；指令引导的先答后推轨迹（7个模型）未能迁移到任一标注模式，而提示诱导的未言明答案翻转则在模型和来源依赖条件下可迁移。此外，我们独立验证并修正了基准标签语义中的文档-数据不一致问题。
原文摘要 · Abstract (English)
Chain-of-thought (CoT) explanations support oversight only if they are faithful: the stated reasoning must actually produce the answer. Auditing black-box (behavioral) detection of unfaithful CoT against FaithCoT-Bench's human annotations, we find answer correctness structures the problem at every level. Answer incorrectness alone (an oracle diagnostic, not a deployable detector) outperforms every purpose-built signal (AUROC 0.696), because 69% of annotated unfaithfulness occurs on incorrect answers. Stratifying by correctness splits detection into two regimes: on correct answers, behavioral signals moderately separate faithful from post-hoc reasoning (0.63-0.67); on incorrect answers, where most unfaithfulness lives, no tested signal is detectably above chance (replicated on all four models for benchmark-wide signals). The standard step-removal metric anti-correlates with human labels; this inversion reproduces on the benchmark's released scores and on hint-dependent counterfactually labeled traces. Linear probes decode the behaviorally blind regime in Llama-3.1-8B and the correct-answer regime in Qwen-2.5-7B, with no shared, positively aligned direction detected across regimes; instructed answer-first traces (7 models) transfer to neither annotated regime, while hint-induced unverbalized answer flips do, in model- and source-dependent settings. We also independently verify and resolve a documentation-data mismatch in the benchmark's label semantics.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。