通往真实性的两条路径：关于LLM幻觉内在编码的研究
Two Pathways to Truthfulness: On the Intrinsic Encoding of LLM Hallucinations
AI总结：
本文探讨LLM幻觉的内在编码机制，通过两种信息路径揭示真实性信号的来源，发现其与知识边界和内部表示密切相关，并提出改进幻觉检测的应用。
摘要：
尽管大型语言模型（LLMs）具备出色的性能，但它们经常产生幻觉。先前的研究表明，其内部状态编码了丰富的真实性信号，但这些信号的起源和机制仍不明确。本文证明，真实性提示源于两种不同的信息路径：（1）问题锚定路径，依赖问题-答案信息流；（2）答案锚定路径，从生成答案本身提取自包含证据。首先，我们通过注意力删除和标记修补验证并分离这些路径。随后，我们揭示了这两种机制的显著特性。进一步的实验表明，（1）这两种机制与LLM的知识边界密切相关；（2）内部表示意识到它们的区别。最后，基于这些洞察，提出了两个应用以提升幻觉检测性能。总体而言，我们的工作为理解LLM如何内部编码真实性提供了新见解，为更可靠和自知的生成系统指明了方向。
英文摘要：
Despite their impressive capabilities, large language models (LLMs) frequently generate hallucinations. Previous work shows that their internal states encode rich signals of truthfulness, yet the origins and mechanisms of these signals remain unclear. In this paper, we demonstrate that truthfulness cues arise from two distinct information pathways: (1) a Question-Anchored pathway that depends on question-answer information flow, and (2) an Answer-Anchored pathway that derives self-contained evidence from the generated answer itself. First, we validate and disentangle these pathways through attention knockout and token patching. Afterwards, we uncover notable and intriguing properties of these two mechanisms. Further experiments reveal that (1) the two mechanisms are closely associated with LLM knowledge boundaries; and (2) internal representations are aware of their distinctions. Finally, building on these insightful findings, two applications are proposed to enhance hallucination detection performance. Overall, our work provides new insight into how LLMs internally encode truthfulness, offering directions for more reliable and self-aware generative systems.
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。
