English(EN) Language Models Show Conformity Bias in RAG: When the Majority Endorses a Poisoned Claim

语言模型在RAG系统中表现出从众偏见

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-02 05:00

一篇新近发表在arXiv上的研究揭示，语言模型在使用检索增强生成（RAG）时会表现出从众偏见。研究人员发现，当被污染的说法在检索文档中以多数形式出现时，Gemma-3-12B模型检测到错误信息的可能性显著低于这些说法仅占少数的情况。这表明模型会受到检索证据共识的影响，即使这些证据具有误导性。 AI

影响突显了RAG系统的一个潜在漏洞，表明需要改进错误信息检测机制。

排序理由研究论文，详细介绍了语言模型的特定偏见。[lever_c_demoted from research: ic=1 ai=1.0]

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 1 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 研究论文，详细介绍了语言模型的特定偏见。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: 1 days old; Coverage has settled into its steady-state source set.

完整方法见我们的编辑标准。

arXiv cs.CL TIER_1 English(EN) · Mehrdad Ghassabi, Audrina Ebrahimi, Zahra Zojaji, Amin Milani Fard · 2026-11-02 05:00

Language Models Show Conformity Bias in RAG: When the Majority Endorses a Poisoned Claim

arXiv:2610.41012v1 Announce Type: new 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 …