Probe Generalization as Subspace Selection for OOD Deception Detection

将探针泛化视为分布外（OOD）欺骗检测的子空间选择

Abstract: Linear probes can be used to detect behaviors and concepts inside language model activations, but may fail to transfer to out-of-distribution examples. When studying the generalization performance of Llama-3.1-8B-Instruct probes over 3 held-out deception detection datasets, we find that projecting inputs onto a small subset of principal components (PCs) from the training distribution of activations enables cross-domain transfer that nearly matches the performance of probes trained directly on the test distribution.

摘要： 线性探针（Linear probes）可用于检测语言模型激活内部的行为和概念，但在面对分布外（OOD）样本时往往难以迁移。在研究 Llama-3.1-8B-Instruct 探针在 3 个留出（held-out）欺骗检测数据集上的泛化性能时，我们发现将输入投影到训练分布激活的一小部分主成分（PCs）上，能够实现跨域迁移，其性能几乎可以媲美直接在测试分布上训练的探针。

Furthermore, we find that PC interpretations can be used to find a subset of those transferable PCs. By using an LLM judge to score each PC on whether its most/least activating examples imply a transferable deception direction, then probing on the highest-scoring PCs, we close the baseline-to-oracle gap by 78% on Insider Trading Report and by 25% on Sandbagging.

此外，我们发现可以通过对主成分进行解释来筛选出这些可迁移的子集。通过使用大语言模型（LLM）作为评判员，对每个主成分进行评分，判断其激活程度最高/最低的样本是否暗示了可迁移的欺骗方向，然后针对高分主成分进行探测，我们在“内幕交易报告”（Insider Trading Report）数据集上缩小了基准与最优解（oracle）之间 78% 的差距，在“消极怠工”（Sandbagging）数据集上缩小了 25% 的差距。

The directions a source probe weights heavily appear to encode source-specific surface features, while the directions that actually transfer appear to encode the same contrast more abstractly, in a way natural language descriptions can capture. Broadly, our results suggest that the OOD robustness of probes is largely the determined by subspace selection.

源探针赋予高权重的方向似乎编码了特定于源数据的表面特征，而真正能够迁移的方向则似乎以更抽象的方式编码了相同的对比关系，且这种方式能够被自然语言描述所捕捉。总的来说，我们的研究结果表明，探针的 OOD 鲁棒性在很大程度上取决于子空间的选择。