MEGA Hub

HalluTracer: Hallucination Detection via Depth-Averaging Truth Signals

Authors

Do you know Zhihao Guo?You can claim authorship or link another user.Do you know Zonghan Wu?You can claim authorship or link another user.Do you know Huan Huo?You can claim authorship or link another user.Do you know DaYong Ye?You can claim authorship or link another user.Do you know Junwei Zhang?You can claim authorship or link another user.Do you know Weiran Yao?You can claim authorship or link another user.Do you know Zhiwei Liu?You can claim authorship or link another user.Do you know Qingsong Wen?You can claim authorship or link another user.Do you know Yilei Shao?You can claim authorship or link another user.

Abstract

Even well-aligned large language models confidently generate factually incorrect text, making hallucination a persistent reliability risk in high-stakes deployments. These models nonetheless carry linearly separable truthfulness signals in their internal representations. Existing white-box detectors, however, collapse this evidence to isolated components or a single depth, discarding discriminative information distributed across the full forward pass. We introduce HalluTracer, a detection framework that reads and aggregates truthfulness evidence across every layer of the forward pass before the model emits any answer token. A geometric analysis reveals that the per-layer signals are weakly correlated, so that simple depth averaging suppresses layer-specific noise and captures nearly all linearly accessible information. Across six open-source language models and five hallucination benchmarks, HalluTracer consistently outperforms matched white-box baselines, with gains ranging from one to fourteen points. Collectively, our work recasts hallucination detection from a layer-selection problem into a depth-aggregation problem governed by the geometric sparsity of the truthfulness signal.

Community

00