arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2508.14314cs.CLcs.AIcs.LG

通过细粒度跨模型一致性实现零知识 LLM 幻觉检测与缓解

Zero-knowledge LLM hallucination detection and mitigation through fine-grained cross-model consistency

Amazon Web Services, USA（亚马逊网络服务）

Aman Goel, Daniel Schwartz, Yanjun Qi

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

提出黑盒框架 Finch-Zk，通过细粒度跨模型一致性检查策略和针对性缓解技术，在无需外部知识下检测并修正 LLM 幻觉，显著提升了事实准确率。

摘要：

大型语言模型（LLMs）在各类任务中展现了出色的能力，但它们仍易产生幻觉——即生成看似合理但包含事实不准确的内容。我们提出了 Finch-Zk，这是一个黑盒框架，利用细粒度跨模型一致性来检测和缓解 LLM 输出中的幻觉，且无需外部知识源。Finch-Zk 引入了两项关键创新：1）一种跨模型一致性检查策略，通过比较由不同模型从语义等价提示生成的响应来揭示细粒度不准确之处；2）一种针对性的缓解技术，对有问题的片段应用精确修正，同时保留准确内容。在 FELM 数据集上的实验表明，与现有方法相比，Finch-Zk 将幻觉检测 F1 分数提高了 6-39%。在缓解方面，当应用于 Llama 4 Maverick 和 Claude 4 Sonnet 等最先进模型时，Finch-Zk 在 GPQA-diamond 数据集上的答案准确率实现了高达 9 个绝对百分点的提升。在多个数据集上的广泛评估表明，Finch-Zk 为增强生产 LLM 系统的事实可靠性提供了一种实用、可即用的保障。

英文摘要：

Large language models (LLMs) have demonstrated impressive capabilities across diverse tasks, but they remain susceptible to hallucinations--generating content that appears plausible but contains factual inaccuracies. We present Finch-Zk, a black-box framework that leverages fine-grained cross-model consistency to detect and mitigate hallucinations in LLM outputs without requiring external knowledge sources. Finch-Zk introduces two key innovations: 1) a cross-model consistency checking strategy that reveals fine-grained inaccuracies by comparing responses generated by diverse models from semantically-equivalent prompts, and 2) a targeted mitigation technique that applies precise corrections to problematic segments while preserving accurate content. Experiments on the FELM dataset show Finch-Zk improves hallucination detection F1 scores by 6-39\% compared to existing approaches. For mitigation, Finch-Zk achieves up to 9 absolute percentage points improvement in answer accuracy on the GPQA-diamond dataset when applied to state-of-the-art models like Llama 4 Maverick and Claude 4 Sonnet. Extensive evaluation on multiple datasets demonstrates that Finch-Zk provides a practical, deployment-ready safeguard for enhancing factual reliability in production LLM 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。

↑