忠实性指标并不衡量忠实性：基于真实标签的元评估
Faithfulness Metrics Don't Measure Faithfulness: A Meta-Evaluation with Ground Truth
AI总结：
针对思维链忠实性度量缺乏真实标签验证的问题，构建了包含真实忠实性标签的数据集BonaFide，系统评估现有指标，发现多数指标表现接近随机、存在偏差且计算成本高。
摘要：
思维链（CoT）已成为解释和审计大型语言模型行为的核心工具。然而，越来越多的证据表明，这些轨迹往往未能忠实反映模型预测背后的计算过程。已有多种忠实性指标被提出，但它们是否真正衡量了忠实性仍不得而知。回答这一问题需要真实标签，但由于内部计算不可直接观察，真实标签难以获取。因此，大多数提出指标的工作仅报告绝对分数或与先前指标的对比，而少数现有基准依赖于似然性或重要性等代理指标，这些属性与忠实性正交，可能误导对CoT可信度的判断。我们通过构建任务来应对这一挑战，这些任务的输出揭示了哪些中间计算必然产生了它们，并开发了一个自动化标注流程，在步骤级和CoT级生成真实忠实性标签。基于这一方法，我们提出了BonaFide基准，包含来自13个任务和10个模型的3066个标注CoT，并利用它首次系统评估了主流忠实性指标。我们的实验表明，大多数指标表现接近随机，存在强烈的预测偏差，并且在更长的CoT上性能下降。最佳指标在CoT级仅达到0.70 AUROC，另一指标在步骤级达到0.59，且两者均无法跨设置迁移，同时计算成本过高。我们的结果暴露了当前忠实性评估中的根本性缺陷，并呼吁开发更可靠、更高效的指标。
英文摘要：
Chains of thought (CoTs) have become central in interpreting and auditing behaviors of large language models. Yet growing evidence suggests that these traces often fail to faithfully represent the computations behind a model's predictions. Several faithfulness metrics have been proposed, but whether they indeed measure faithfulness remains unknown. Answering this requires ground-truth labels, which are hard to obtain since internal computations are not directly observable. Consequently, most works proposing metrics report only absolute scores or comparisons to prior metrics, and the few existing benchmarks rely on proxies like plausibility or importance, properties orthogonal to faithfulness that can mislead about whether a CoT can be trusted. We address this challenge by constructing tasks whose outputs reveal which intermediate computations must have produced them, and developing an automated labeling pipeline that yields ground-truth faithfulness labels at both the step and CoT level. Building on this methodology, we present BonaFide, a benchmark of 3,066 labeled CoTs across 13 tasks and 10 models, and use it to conduct the first systematic evaluation of prominent faithfulness metrics. Our experiments show that most metrics perform near chance, exhibit strong prediction biases and degrade on longer CoTs. The best metric reaches only 0.70 AUROC at the CoT level while another reaches 0.59 at the step level, with neither transferring across settings, while entailing prohibitively high computational cost. Our results expose fundamental gaps in current faithfulness evaluation and call for the development of more reliable and efficient metrics.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
