arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.02668cs.CLcs.AIcs.MA
收藏

过于礼貌而不反驳:理解多智能体系统中的趋炎附势传播

Too Polite to Disagree: Understanding Sycophancy Propagation in Multi-Agent Systems

University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校)

Vira Kasprova, Amruta Parulekar, Abdulrahman AlRabah, Krishna Agaram, Ritwik Garg, Sagar Jha, Nimet Beyza Bozdag, Dilek Hakkani-Tur

AI总结:

研究探讨多智能体系统中趋炎附势行为的影响,通过实验发现提供趋炎附势先验信息可减少影响、缓解错误连锁反应并提升讨论准确性。

摘要:

大型语言模型(LLMs)常常表现出趋炎附势:即使与模型意见相冲突,也会同意用户立场。尽管先前研究大多集中在单智能体设置,但在协作多智能体系统中仍较少探讨。我们询问是否其他智能体的趋炎附势水平意识会影响讨论结果。为此,我们使用六个开源LLM进行受控实验,向智能体提供同伴趋炎附势排名,以估计每个同伴的趋炎附势倾向。这些排名基于使用各种静态(讨论前)和动态(在线)策略计算出的分数。我们发现,提供趋炎附势先验信息可减少趋炎附势同伴的影响,缓解错误连锁反应,并通过绝对10.5%提升最终讨论准确性。因此,这是一种轻量且有效的方法,可用于减少讨论中的趋炎附势行为并提高下游准确性。

英文摘要:

Large language models (LLMs) often exhibit sycophancy: agreement with user stance even when it conflicts with the model's opinion. While prior work has mostly studied this in single-agent settings, it remains underexplored in collaborative multi-agent systems. We ask whether awareness of other agents' sycophancy levels influences discussion outcomes. To investigate this, we run controlled experiments with six open-source LLMs, providing agents with peer sycophancy rankings that estimate each peer's tendency toward sycophancy. These rankings are based on scores calculated using various static (pre-discussion) and dynamic (online) strategies. We find that providing sycophancy priors reduces the influence of sycophancy-prone peers, mitigates error-cascades, and improves final discussion accuracy by an absolute 10.5%. Thus, this is a lightweight and efficient way to reduce model sycophancy during discussions and subsequently improve downstream accuracy.

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。

↑