Why RAGs Hallucinate: Penalty-Aware Evaluation of Retrieval-Augmented Generation Systems with Knowledge-Gap Canaries
检索增强生成系统为何会产生幻觉:基于知识间隔金丝雀的惩罚感知评估
专题命中 RAG评测 :RAG(summary_cn,abstract);retrieval-augmented generation(title,abstract);分类 cs.CL、cs.AI
AI总结 该研究针对RAG系统的幻觉问题,提出惩罚感知评估框架,结合不对称评分、知识间隔金丝雀和失败归因流程,在SimpleQA-Verified数据集上揭示系统差异源于不当作答,而非准确率,相关资源已公开。
Comments 13 pages, 1 figure, 5 tables. Code, full per-request logs, and all judge votes: this https URL (https://github.com/adorosario/why-rags-hallucinate)