arXiv cs.CL· Mehrdad Ghassabi, Audrina Ebrahimi, Zahra Zojaji, Amin Milani Fard·· 11 小时前AI 评分38
语言模型在 RAG 中表现出从众偏差：多数文档支持投毒声明时更难识别
Language Models Show Conformity Bias in RAG: When the Majority Endorses a Poisoned Claim
AI 导读
研究发现 RAG 场景下语言模型存在从众偏差：Gemma-3-12b 在检索文档中仅 1 篇投毒、6 篇健康时投毒识别率为 25%（50/200），而当 6 篇互相一致的投毒文档占据多数时识别率降至 10%（20/200）。在结果发生变化的 38 道 SciFact 题目中，34 道由识别转为未识别，仅 4 道反向变化（McNemar 检验 p = 6.04e-7）。
正文 · 原文
Abstract:Retrieval-augmented generation (RAG) can expose language models to misleading evidence in retrieved documents. We investigate whether increasing the number of mutually agreeing poisoned documents affects a model's ability to detect misinformation. We evaluate 200 randomly selected questions from SciFact (random seed 38), comparing contexts with one poisoned and six healthy documents against contexts with six mutually agreeing poisoned documents and one healthy document. For Gemma-3-12b, poison-detection rates decrease from 25% (50/200) in the minority condition to 10% (20/200) in the majority condition. Among the 38 questions with different outcomes across conditions, 34 change from detected to undetected, while only four change in the opposite direction (exact McNemar test, p = 6.04e-7). This pattern is consistent with a conformity-like response to a majority of mutually agreeing poisoned documents. Our results provide initial evidence of conformity bias in RAG-based misinformation detection: Gemma-3-12b is substantially less likely to detect poisoned claims when they dominate the retrieved evidence.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.41012 [cs.CL] |
| (or arXiv:2610.41012v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.41012 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mehrdad Ghassabi [view email]
[v1]
Thu, 29 Oct 2026 21:45:09 UTC (22 KB)
来源：arXiv cs.CL · arxiv.org