大模型答错题，是因为输出阈值太死，内部知识其实早就知道答案。
When Do Internal Probes Beat Reading the Answer? Miscalibrated Readouts and Behavior-Concealed Knowledge in Language Models
- 用隐藏层探针发现模型内部早有正确判断，但输出前被错误阈值抹去。
- 0.6B模型行为准确率仅50%，修正阈值后提升至81%；8B模型同样因阈值问题表现更差。
- 提出可复现的校准方法，让模型真正发挥出隐藏的知识能力。
一个0.6B的语言模型在验证1200个逻辑结论（一半有效，一半经单语义修改）时，始终回答YES，行为上无法区分。然而，对其隐藏状态进行线性探针分析，可在未见过的逻辑结构上达到0.96 AUC，甚至能区分由真结论词汇组成的干扰项（AUC 0.90）。研究发现，正确判断实际上存在于模型输出的对数概率中（边际AUC 0.89），但因决策阈值偏移+4.6σ而被完全掩盖。该现象在九种不同语义标签配置、五种模型三类架构下均成立：行为准确率与阈值偏移呈强负相关（Spearman -0.93），而边际排序变化较小。在13倍规模范围内，内部知识趋于饱和，但自由生成行为非单调——8B模型表现反而劣于4B模型，根源是答案通道失效而非阈值本身。通过单一参数校正，无需在评估结构上训练即可将0.6B模型行为准确率从50%提升至81%；边际解码校准使8B模型恢复至94%；少量提示也能实现类似效果，通过重置阈值（+4.6σ → 0.0σ）同时保持排序一致性。对比探针与边际解码，可识别出三种状态：隐藏、误校准、未察觉。在无表面线索的迷宫任务中，审计能准确识别第三类。标准生成场景下，仅凭答案表面特征和启发式标签即可复现已有探针结果，无需内部访问。
原文摘要 · Abstract (English)
A 0.6B language model, asked to verify 1,200 logical conclusions (half valid, half corrupted by a single semantic edit), answers YES every time. Judged by behavior it discriminates nothing; linear probes on its hidden states read the correct verdict at 0.96 AUC, transferring to unseen logical structures and separating foils built from exactly the words of the true conclusion (0.90). We ask where the verdict is lost, and find the dominant failure is a single scalar. The verdict survives to the model's own output logits (margin AUC 0.89) along a well-aligned readout direction; a saturated decision threshold, offset by +4.6 sigma, erases it. The diagnosis generalizes: across 90 semantic-label configurations of a five-model, three-family factorial, behavioral accuracy collapses onto a single function of threshold offset (Spearman -0.93) while margin ranking moves far less. Across a 13x scale range, internal knowledge saturates while free-form behavior is non-monotone: an 8B model underperforms its 4B sibling through an answer-channel failure rather than the threshold; forced-choice accuracy is monotone. The diagnosis is actionable: a one-parameter correction, never fit on evaluated structures, repairs behavior from 50% to 81% (0.6B); calibrated margin decoding recovers 94% at 8B; few-shot prompting works the same way, recentering the threshold (+4.6 sigma to 0.0 sigma) while preserving ranking. Comparing probe to margin separates three regimes: concealed, miscalibrated, and undetected. On a maze task built so foils carry no surface cues, the audit correctly reports the third. In the standard generation setting, answer-surface features and heuristic labels reproduce published probing results without any internal access.
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。