arXiv cs.CL· Aneesh Komanduri, Majid Afshar, Timothy A. Miller, Yanjun Gao·· 12 小时前AI 评分29
通过因果鲁棒性度量大语言模型的机制不确定性
Measuring Mechanistic Uncertainty in Large Language Models via Causal Robustness
AI 导读
研究提出“机制不确定性”概念与“因果鲁棒性边际”，用 Jacobian Lens（Jlens）概念引导向量量化模型输出偏离所需的最小内部干预。在三个开源模型和五个问答数据集上，错误与不确定回答的边际更小，说明其对内部干预更敏感，且该指标可补充传统基于输出的不确定性方法。知识注入实验中，正确习得的事实训练后边际上升，表明因果鲁棒性可作为有机制依据的不确定性信号。
正文 · 原文
Abstract:Large language models (LLMs) are increasingly used in critical domains, where their susceptibility to hallucinations makes reliable and interpretable uncertainty quantification essential. We introduce the concept of mechanistic uncertainty, which characterizes uncertainty through the stability of the internal mechanisms supporting a model's response. We propose the causal robustness margin, defined as the minimum internal intervention required to deviate the model's output, and operationalize it using Jacobian Lens (Jlens) concept steering vectors. Across three open-source models and five question-answering datasets, we show incorrect and uncertain responses exhibit smaller margins, indicating greater sensitivity to internal interventions, and the causal robustness margin complements conventional output-based uncertainty methods. Finally, in a controlled knowledge injection experiment, correctly learned facts show increased margins after training, suggesting that causal robustness provides a mechanistically grounded signal of uncertainty.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.22636 [cs.CL] |
| (or arXiv:2610.22636v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22636 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Aneesh Komanduri [view email]
[v1]
Thu, 15 Oct 2026 20:46:06 UTC (416 KB)
来源：arXiv cs.CL · arxiv.org