The Hallucination Signal Is a Mean Shift: Why Simple Probes Suffice
Computation and LanguageArtificial IntelligenceMachine Learning
Summary
The authors studied how language models' hidden states can be used to detect when these models make mistakes (hallucinations). They found that most of the useful signal is captured by a single main direction, and removing this makes detection fail. Simple methods like logistic regression work almost as well as more complicated models, meaning that complexity comes from handling data variations, not from complex patterns. They also developed a new technique called LayerMix that combines information from multiple layers effectively without needing prior knowledge.
Authors
Jungseob Lee, Jaehyung Seo, Heuiseok Lim
Abstract
Hidden-state probes effectively detect LLM hallucinations, but the geometry of the signal remains poorly characterized, driving increasingly complex probe architectures. Across three 7B-scale models and three datasets in a paired-example paradigm, we find the signal overwhelmingly dominated by a single mean-shift component, and removing this direction collapses detection to chance. Shrinkage linear discriminant analysis closes about 73% of the gap between 1D and full-dimensional classifiers, so apparent architectural complexity largely reflects high-dimensional covariance estimation difficulty rather than exploitable non-linearity. A simple L2-regularized logistic regression (0.952 AUROC) bounds or outperforms twelve controlled architectural alternatives, and our multi-layer aggregation exceeds CLAP cross-layer attention probing under matched paradigm. Because the signal spans a contiguous layer band, LayerMix aggregates it to match oracle-layer performance without oracle access. Our claims characterize the geometry within the controlled paired-example paradigm. Our code is available at https://github.com/js-lee-AI/LayerMix.