The Knowing-Saying Gap: When Probes See Errors that Confidence Misses

知行鸿沟：当探测器发现置信度所忽略的错误时

Abstract: Linear probes detect corrupted context in language models with near-perfect accuracy, yet this does not translate into reliable failure prediction. The result is a dissociation with direct implications for deployment monitoring. 摘要： 线性探测器（Linear probes）能够以近乎完美的准确率检测出语言模型中受损的上下文，但这并不能转化为可靠的故障预测。这种结果导致了一种“知行鸿沟”，对模型部署监控具有直接影响。

Across multi-hop arithmetic chains, probes that detect corruption turn out to be uninformative about final answer correctness; models forced into structured confidence formats collapse to two values with indistinguishable error rates; and probe persistence across hops fails to separate correct from incorrect outcomes, refuting our pre-registered “persistence beats peak” hypothesis. 在多跳算术链任务中，能够检测到上下文损坏的探测器对于最终答案的正确性并无参考价值；当模型被强制要求以结构化置信度格式输出时，其结果往往坍缩为两个数值，且两者之间的错误率无法区分；此外，跨跳的探测器持久性也无法有效区分正确与错误的输出，这反驳了我们预先注册的“持久性优于峰值（persistence beats peak）”假设。

This pattern of knowing but not saying generalises across model families including reasoning models. As a real-time monitor, probe-based interventions are sharply model and error-type dependent: branch-and-pick is net-positive across models and uniquely non-breaking on Llama-3.1-8B (4 rescued, 0 broken), while reprompt and replace-prior break correct traces at roughly the rate they rescue wrong ones. 这种“知而不言”的模式在包括推理模型在内的多个模型家族中普遍存在。作为一种实时监控手段，基于探测器的干预措施高度依赖于模型类型和错误类型：“分支选择（branch-and-pick）”策略在各模型中表现出净正向收益，且在 Llama-3.1-8B 上表现出独特的无破坏性（成功挽救 4 例，破坏 0 例）；而“重新提示（reprompt）”和“替换先验（replace-prior）”策略在挽救错误路径的同时，往往会以大致相同的比例破坏原本正确的路径。

Probe-based monitoring is a necessary complement to verbalised confidence, but no single intervention dominates, and the deployable answer is model-aware, error-type-aware routing. 基于探测器的监控是对模型口头置信度（verbalised confidence）的必要补充，但目前没有单一的干预手段占据绝对优势。因此，可行的部署方案应当是基于模型感知和错误类型感知的路由策略。