动摇根基的低语：分析并缓解大型语言模型中的错误前提幻觉
Whispers that Shake Foundations: Analyzing and Mitigating False Premise Hallucinations in Large Language Models
AI总结：
本文揭示错误前提问题会通过少数“错误前提注意力头”干扰LLM知识提取，并提出推理时约束这些头的FAITH方法，仅约束约1%注意力头即可提升近20%性能。
摘要：
大型语言模型（LLMs）已展现出令人瞩目的能力，但仍受幻觉问题困扰。该问题的一个重要类型是错误前提幻觉，我们将其定义为LLMs在面对错误前提问题时生成幻觉文本的现象。本文对错误前提幻觉进行了全面分析，并阐明其内部工作机制：一小部分注意力头（我们将其指定为错误前提头）会干扰知识提取过程，从而导致错误前提幻觉的发生。基于分析，我们提出FAITH（False premise Attention head constraIning for miTigating Hallucinations，即通过约束错误前提注意力头来缓解幻觉），这是一种新颖且有效的缓解错误前提幻觉的方法。该方法在模型推理过程中约束错误前提注意力头。令人瞩目的是，大量实验表明，仅约束模型中约1%的注意力头，就能使模型性能显著提升近20%。
英文摘要：
Large Language Models (LLMs) have shown impressive capabilities but still suffer from the issue of hallucinations. A significant type of this issue is the false premise hallucination, which we define as the phenomenon when LLMs generate hallucinated text when confronted with false premise questions. In this paper, we perform a comprehensive analysis of the false premise hallucination and elucidate its internal working mechanism: a small subset of attention heads (which we designate as false premise heads) disturb the knowledge extraction process, leading to the occurrence of false premise hallucination. Based on our analysis, we propose \textbf{FAITH} (\textbf{F}alse premise \textbf{A}ttention head constra\textbf{I}ining for mi\textbf{T}igating \textbf{H}allucinations), a novel and effective method to mitigate false premise hallucinations. It constrains the false premise attention heads during the model inference process. Impressively, extensive experiments demonstrate that constraining only approximately $1\%$ of the attention heads in the model yields a notable increase of nearly $20\%$ of model performance.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
