arXiv cs.CL· Shubham Subhnil, Yannis Belkhiter, Erik Miehling, Adam Dahlgren Lindstr\"om, Elizabeth M. Daly·· 2 天前AI 评分22
隐藏状态探针能否定位推理失败步骤？4B 与 8B 推理模型上的有效性研究
Effectiveness of Hidden-State Probes for Localizing Reasoning Failures
AI 导读
研究测试了 4B 和 8B 推理模型的隐藏状态探针能否区分推理失败的三种情形并定位出错步骤。在 8B 模型的错误推理链上，置信度探针漏掉了 81% 的"单步导致不可恢复"案例和 47% 的"重试即可成功"案例，且换一种训练设计会反转这两类失败的排序。两个规模下，没有任何探针在定位失败步骤上优于仅基于推理链长度的规则，4B 上专门预测该步骤的探针在 10 次相同推理链运行中均未胜出。
正文 · 原文
Abstract:Verifying the final answer of a long reasoning trace may show that the trace is wrong but it does not distinguish three cases. The problem may lie beyond the model's capability, another attempt would succeed, or one step in an otherwise recoverable trace made the trace fail. Standard hidden-state probes are trained on final-answer correctness, a label that conflates these cases. Monte Carlo rollouts distinguish the cases and locate the failing step at additional inference cost. We test the hidden states of 4B and 8B reasoning models for the failure type and the failing step and compare each probe with the strongest baseline that uses no hidden states. On incorrect 8B traces, the confidence probe, a calibrated correctness probe at a midpoint cutoff, misses 81% of cases in which a single step makes the trace unrecoverable and 47% of cases in which another attempt would succeed. A probe with a different training design over the same states reverses the ordering of these two kinds of failure, so either ordering can be read from those states. At both scales, no single tested probe locates the failing step more accurately than a rule based only on trace length, and at 4B a probe trained to predict that step fails to outperform this rule in any of 10 runs on identical traces. The confidence probe therefore indicates that an answer is wrong, but its failure-type ordering depends on its training design.
| Comments: | Accepted at the NeurIPS 2026 Workshop on Interpretability as a Science (InterpScience). OpenReview: this https URL . 30 pages, 5 figures, 13 tables. Code: this https URL ; Data: this https URL |
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL) |
| ACM classes: | I.2.6; I.2.7 |
| Cite as: | arXiv:2610.22046 [cs.LG] |
| (or arXiv:2610.22046v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22046 arXiv-issued DOI via DataCite (pending registration) |
|
| Journal reference: | Interpretability as a Science: NeurIPS 2026 Workshop, 2026 |
Submission history
From: Shubham Subhnil [view email]
[v1]
Thu, 15 Oct 2026 15:15:40 UTC (119 KB)
来源：arXiv cs.CL · arxiv.org