推理在哪断裂？基于隐藏态传输几何的步骤级幻觉检测
Where Does Reasoning Break? Step-Level Hallucination Detection via Hidden-State Transport Geometry
这篇论文把幻觉检测从“整体打分”推进到“单步定位”，做推理模型调试和可解释性研究的团队值得关注——它用几何视角揭示了错误发生的精确位置，比传统方法更细粒度。
该论文提出了一种新的幻觉检测方法，将多步推理中的幻觉视为隐藏状态轨迹的几何特性，而非传统对整个输出打分。作者通过对比PCA构建标签条件教师模型，提取七个几何过渡特征，并蒸馏出BiLSTM学生模型，仅需单次前向传播即可定位首个错误步骤。
在ProcessBench、PRM800K等基准测试中，该方法优于基于熵、探针和注意力的基线，教师模型跨语言模型和数据集稳定迁移，但学生模型在分布偏移下失效。研究将步骤级幻觉检测重新定义为轨迹动力学问题，并指出分布偏移下保持对比传输裕度是部署的关键障碍。
Where Does Reasoning Break? Step-Level Hallucination Detection via Hidden-State Transport Geometry
Large language models hallucinate during multi-step reasoning, but most existing detectors operate at the trace level: they assign one confidence score to a full output, fail to localize the first error, and often require multiple sampled completions. We frame hallucination instead as a property of the hidden-state trajectory produced during a single forward pass. Correct reasoning moves through a stable manifold of locally coherent transitions; a first error appears as a localized excursion in transport cost away from this manifold. We operationalize this view with a label-conditioned teacher that builds a trace-specific contrastive PCA lens and scores each step with seven geometric transition features, and a deployable BiLSTM student distilled from the teacher that operates on raw hidden states without inference-time labels. We prove that contrastive PCA is the optimal projection for a transport-separation objective between first error and correct states, and that single-pass first error localization holds whenever the first error creates a positive transport margin over preceding correct transitions. On ProcessBench, PRM800K, HaluEval, and TruthfulQA, both models outperform entropy-based, probing-based, and attention-based baselines in-domain; the teacher transfers stably across language models and datasets, while the student collapses under shift, a gap our distillation theory predicts. These results recast step-level hallucination detection as a problem of trajectory dynamics and identify the central obstacle to deployment: preserving the contrastive transport margin under distribution shift.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include 幻觉检测, 推理模型, 隐藏状态轨迹, 对比PCA.
Source attribution: curated from arXiv cs.AI (@Tyler Alvarez, Ali Baheri).