Probe Generalization as Subspace Selection for OOD Deception Detection

A new arXiv study (2609.02893v1) finds that projecting Llama-3.1-8B-Instruct activations onto a small subset of principal components from the training distribution enables cross-domain deception detection that nearly matches probes trained directly on the test distribution. Using an LLM judge to select transferable PCs closes the baseline-to-oracle gap by 78% on Insider Trading Report and 25% on Sandbagging, suggesting out-of-distribution robustness is largely determined by subspace selection.

arXiv:2609.02893v1 Announce Type: new 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. 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. 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 determined by subspace selection.