arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.10412cs.CLcs.AIcs.LG

在没有黄金标准答案的情况下测量和减少大语言模型幻觉

Measuring and Reducing LLM Hallucination without Gold-Standard Answers

UC Santa Cruz(加州大学圣克鲁兹分校);ByteDance Research(字节跳动研究院);Northwestern University(西北大学)

Jiaheng Wei, Yuanshun Yao, Jean-Francois Ton, Hongyi Guo, Andrew Estornell, Yang Liu

AI总结:

本文提出FEWL幻觉度量方法,在无黄金标准答案时通过加权多个现成大语言模型的答案来测量并减少幻觉,实验证明其有效性。

摘要:

大语言模型幻觉,即生成事实上不正确但看似令人信服的答案,目前是对大语言模型可信度和可靠性的主要威胁。解决这一复杂问题的第一步是测量它。然而,现有的幻觉度量指标要求有一个带有黄金标准答案的基准数据集,即由人类编写的“最佳”或“正确”答案。这种要求使得幻觉测量成本高昂且容易出现人为错误。在这项工作中,我们提出了通过加权大语言模型进行事实性评估(FEWL),这是一种创新的幻觉度量指标,专门为没有黄金标准答案的场景设计。FEWL利用现成大语言模型的答案作为黄金标准答案的代理。关键挑战是如何有策略地量化参考大语言模型的专业水平。我们证明FEWL具有一定的理论保证,并通过实验证明它比直接使用参考大语言模型能给出更准确的幻觉测量结果。我们还展示了如何利用FEWL通过上下文学习和监督微调来减少幻觉。在Truthful-QA、CHALE和HaluEval数据集上的大量实验结果证明了FEWL的有效性。

英文摘要:

LLM hallucination, i.e. generating factually incorrect yet seemingly convincing answers, is currently a major threat to the trustworthiness and reliability of LLMs. The first step towards solving this complicated problem is to measure it. However, existing hallucination metrics require having a benchmark dataset with gold-standard answers, i.e. "best" or "correct" answers written by humans. Such requirements make hallucination measurement costly and prone to human errors. In this work, we propose Factualness Evaluations via Weighting LLMs (FEWL), an innovative hallucination metric that is specifically designed for the scenario when gold-standard answers are absent. FEWL leverages the answers from off-the-shelf LLMs that serve as a proxy of gold-standard answers. The key challenge is how to quantify the expertise of reference LLMs resourcefully. We show FEWL has certain theoretical guarantees and demonstrate empirically it gives more accurate hallucination measures than naively using reference LLMs. We also show how to leverage FEWL to reduce hallucination through both in-context learning and supervised fine-tuning. Extensive experiment results on Truthful-QA, CHALE, and HaluEval datasets demonstrate the effectiveness of FEWL.

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。

↑