arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.01637cs.CLcs.AI
收藏

误导比纠正更容易:LLM 从众中的有害与有益修正

Easier to Mislead Than to Correct: Harmful and Beneficial Revision in LLM Conformity

Amazon(亚马逊);Georgia Institute of Technology(佐治亚理工学院);Illinois Institute of Technology(伊利诺伊理工学院)

Jiaming Qu, Lucheng Fu, Yibo Hu

AI总结:

通过控制实验,研究大语言模型在多智能体系统中面对同伴答案时的从众行为,发现同伴一致意见更容易误导原本正确的模型,而权威标签使模型更倾向于选择被认可的答案,且通用推理干预无法可靠地减少有害修正。

摘要:

大语言模型越来越多地用于多智能体系统,在这些系统中,它们会看到并回应其他智能体的答案。一个关键风险是从众:模型可能仅仅因为其他人同意不同的答案而放弃自己的答案。先前的研究表明,LLM 经常向多数答案修正,但仍不清楚这些修正是像引入新错误一样频繁地帮助纠正错误。在本文中,我们进行了一项受控研究,其中 LLM 首先回答一个问题,然后在做出最终决定之前看到模拟的同伴回应。我们操纵两个社会线索:共识结构和分配给同伴的权威标签,并测量它们如何影响有益和有害的修正。在四个开放权重的 LLM 和七个问答数据集上,我们发现同伴一致意见使得误导原本正确的模型比纠正原本错误的模型容易得多。权威标签使模型更可能选择被认可的答案,无论其是否正确。更令人担忧的是,通用的推理干预(如思维链和反思)并不能可靠地减少有害修正同时保留有益修正。这些发现表明,多智能体 LLM 系统应该验证同伴答案,而不是简单地聚合它们。

英文摘要:

Large language models are increasingly used in multi-agent systems, where they see and respond to other agents' answers. A key risk is conformity: a model may abandon its own answer simply because others agree on a different one. Prior studies show that LLMs often revise toward a majority answer, but it remains unclear whether these revisions help correct mistakes as often as they introduce new errors. In this paper, we conduct a controlled study in which an LLM first answers a question, then sees simulated peer responses before making a final decision. We manipulate two social cues: consensus structure and authority labels assigned to peers, and measure how they influence beneficial and harmful revisions. Across four open-weight LLMs and seven QA datasets, we find that peer agreement makes it much easier to mislead initially correct models than to correct initially wrong ones. Authority labels make models more likely to choose the endorsed answer, regardless of whether it is correct. More concerningly, generic reasoning interventions such as chain-of-thought and reflection do not reliably reduce harmful revision while preserving beneficial revision. These findings suggest that multi-agent LLM systems should verify peer answers rather than simply aggregate them.

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。

↑