arXiv cs.CL· Alexandra Kuleshova, Andrey Volodichev, Daria Kotova, Alexey Zaytsev·· 12 小时前AI 评分31
LLM 如何藏在暗空间：从 Unembedding 暗空间读取上下文依赖
What LLMs Do in the Shadows: Reading Context Reliance from the Unembedding Dark Space
AI 导读
研究者提出 DarkLens，一种仅需单次前向传播的轻量忠实性检测器，通过将隐藏状态投影到与 unembedding 矩阵弱组合的"暗空间"来估计上下文依赖。该子空间中隐藏状态范数的占比被称为 Dark Matter Ratio（DMR），与参数知识得分结合后构成 DarkLens。
正文 · 原文
Abstract:Detecting faithfulness hallucinations in large language models (LLMs) is crucial for their safe deployment in high-stakes domains. Recent approaches address this by contrasting contextual against parametric knowledge utilization, yet measuring context reliance typically demands noticeable overhead, such as materializing attention maps, running additional forward passes, or training auxiliary probes. In this work, we show that context reliance can be estimated by projecting hidden states onto the dark space - directions that compose weakly with the unembedding matrix. Beyond its recognized function as an initial-token attention sink, we show that this subspace actively mediates context processing, and that its share of hidden-state norm - termed the Dark Matter Ratio (DMR) - provides a direct score of context utilization. Pairing DMR with a parametric knowledge score yields DarkLens, a lightweight faithfulness detector operating strictly within a single forward pass. Evaluated across three QA and summarization benchmarks on six open-weight LLMs (4B-13B parameters), DarkLens matches or outperforms a wide range of baselines without requiring extra forward passes or probe training.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.23182 [cs.CL] |
| (or arXiv:2610.23182v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.23182 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Alexandra Kuleshova [view email]
[v1]
Fri, 16 Oct 2026 09:21:49 UTC (300 KB)
来源：arXiv cs.CL · arxiv.org