arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2403.06448cs.CLcs.AI

基于大语言模型内部状态的无监督实时幻觉检测

Unsupervised Real-Time Hallucination Detection based on the Internal States of Large Language Models

Tsinghua University（清华大学）；Beijing Institute of Technology（北京理工大学）；Renmin University of China（中国人民大学）

Weihang Su, Changyue Wang, Qingyao Ai, Yiran HU, Zhijing Wu, Yujia Zhou, Yiqun Liu

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出无监督框架MIND，利用LLM内部状态实现无需标注的实时幻觉检测，并构建基准HELM，实验证明其性能优于现有方法。

摘要：

大语言模型（LLMs）中的幻觉是指 LLM 生成连贯但事实不准确响应的现象。该问题削弱了 LLM 在实际应用中的有效性，因此有必要研究如何检测和缓解 LLM 的幻觉。以往研究主要集中于幻觉检测的后处理技术，这些方法通常计算密集，且由于与 LLM 推理过程相分离而效果有限。为克服这些局限，我们提出了 MIND，一个无监督训练框架，利用 LLM 的内部状态进行实时幻觉检测，无需人工标注。此外，我们还提出了 HELM，一个用于评估多个 LLM 幻觉检测能力的新基准，包含多样化的 LLM 输出以及 LLM 推理过程中的内部状态。实验表明，MIND 在幻觉检测方面优于现有最先进方法。

英文摘要：

Hallucinations in large language models (LLMs) refer to the phenomenon of LLMs producing responses that are coherent yet factually inaccurate. This issue undermines the effectiveness of LLMs in practical applications, necessitating research into detecting and mitigating hallucinations of LLMs. Previous studies have mainly concentrated on post-processing techniques for hallucination detection, which tend to be computationally intensive and limited in effectiveness due to their separation from the LLM's inference process. To overcome these limitations, we introduce MIND, an unsupervised training framework that leverages the internal states of LLMs for real-time hallucination detection without requiring manual annotations. Additionally, we present HELM, a new benchmark for evaluating hallucination detection across multiple LLMs, featuring diverse LLM outputs and the internal states of LLMs during their inference process. Our experiments demonstrate that MIND outperforms existing state-of-the-art methods in hallucination detection.

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。

↑