Abstract:Despite the groundbreaking advancements made by large language models (LLMs), hallucination remains a critical bottleneck for their deployment in high-stakes domains. Existing classification-based methods mainly rely on static and passive signals from internal states, which often captures the noise and spurious correlations, while overlooking the underlying causal mechanisms. To address this limitation, we shift the paradigm from passive observation to active intervention by introducing CausalGaze, a novel hallucination detection framework based on structural causal models (SCMs). CausalGaze models LLMs' internal states as dynamic causal graphs and employs counterfactual interventions to disentangle causal reasoning paths from incidental noise, thereby enhancing model interpretability. Extensive experiments across four datasets and three widely used LLMs demonstrate the effectiveness of CausalGaze, especially achieving over 5.2\% improvement in AUROC on the TruthfulQA dataset compared to state-of-the-art baselines.
| Comments: | Accepted as ACL2026 Findings |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2604.11087 [cs.LG] |
| (or arXiv:2604.11087v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2604.11087 arXiv-issued DOI via DataCite |
From: Linggang Kong [view email]
[v1]
Mon, 13 Apr 2026 07:09:33 UTC (2,800 KB)
[v2]
Wed, 6 May 2026 03:11:01 UTC (2,861 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。