不同评估方法让同一模型的可信度得分差超30%，说明可信度测量结果不客观。
Measuring Faithfulness Depends on How You Measure: Classifier Sensitivity in LLM Chain-of-Thought Evaluation
- 用三种不同方法评估12个模型的推理过程，发现结果差异巨大
- 同一模型在不同方法下可信度从69.7%到82.6%不等，最高差30.6个百分点
- 评估标准差异大，导致模型排名可能完全反转，需报告多方法敏感性范围
近期链式思维（CoT）可信度研究仅给出单一数值（如DeepSeek-R1承认提示词39%），暗示可信度是可客观衡量的属性。本文证明并非如此：对12个开源模型（7B至1T参数，覆盖9个系列）的10,276条受干扰推理轨迹，分别使用三种分类器（仅正则表达式、正则+LLM流水线、Claude Sonnet 4判官）进行评估，得到的可信度分别为74.4%、82.6%和69.7%。单模型间差异达2.6至30.6个百分点，所有两两比较均显著（p<0.001）。分歧系统性存在：对于奉承类提示，一致性系数仅0.06（微弱），而流水线将883例判定为可信，但Sonnet判为不可信，反之仅2例；模型排名也被逆转：Qwen3.5-27B在流水线下排名第1，在Sonnet下为第7；OLMo-3.1-32B从第9升至第3。不同分类器在词汇提及与认知依赖等层面操作可信度定义，导致相同行为测量结果迥异。因此，跨研究使用不同分类器时，可信度数值无法有效比较，未来评估应报告多种方法下的敏感性范围。
原文摘要 · Abstract (English)
Recent work on chain-of-thought (CoT) faithfulness reports single aggregate numbers (e.g., DeepSeek-R1 acknowledges hints 39% of the time), implying that faithfulness is an objective, measurable property of a model. This paper provides evidence that it is not. Three classifiers (a regex-only detector, a regex-plus-LLM pipeline, and a Claude Sonnet 4 judge) are applied to 10,276 influenced reasoning traces from 12 open-weight models spanning 9 families and 7B to 1T parameters. On identical data, these classifiers produce faithfulness rates of 74.4%, 82.6%, and 69.7%. Per-model gaps range from 2.6 to 30.6 percentage points; all pairwise McNemar tests are significant (p < 0.001). The disagreements are systematic: Cohen's kappa ranges from 0.06 ("slight") for sycophancy hints to 0.42 ("moderate") for grader hints, and the asymmetry is pronounced: for sycophancy, 883 cases are classified as faithful by the pipeline but unfaithful by the Sonnet judge, while only 2 go the other direction. Classifier choice can also reverse model rankings: Qwen3.5-27B ranks 1st under the pipeline but 7th under Sonnet; OLMo-3.1-32B moves from 9th to 3rd. Different classifiers operationalize faithfulness at different levels of stringency (lexical mention versus epistemic dependence), yielding divergent measurements on the same behavior. These results indicate that published faithfulness numbers cannot be meaningfully compared across studies using different classifiers, and that future evaluations should report sensitivity ranges across multiple classification methodologies.
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。