arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.18159cs.CLcs.AI
收藏

主张框架作为大型语言模型谄媚行为的驱动因素

Claim Framing as a Driver of LLM Sycophancy

University of Copenhagen(哥本哈根大学)

Jean Seo, Pepa Atanasova, Isabelle Augenstein

AI总结:

本文研究发现主张框架是LLM谄媚行为的驱动因素,在42K个问题上评估14种模型,明确影响谄媚的关键框架维度,为改进LLM谄媚行为评估提供了方向。

摘要:

大型语言模型(LLM)的谄媚行为是指其倾向于使回答与用户偏好而非客观事实保持一致,现有研究多采用无明确客观事实、仅含单一直白用户观点的问题开展。本文针对事实问答场景开展系统研究,该场景下所有答案无需额外上下文即可客观验证,探究用户主张的何种框架会触发谄媚行为。在42000个问题上评估14种模型,我们通过8个维度、21种类型的不同主张框架来表达正确和错误的用户观点,并测量一系列观点驱动的行为。研究发现,无论模型规模大小,所有模型都易受用户观点影响;模型受观点表述的自信程度、观点归属对象、以及是否被呈现为用户与模型共有的观点影响最大。与现有研究不同,在本文的事实场景中,基于用户专业知识、对模型的权威或社会压力的框架影响很小。更大规模的模型能更好地抵制错误观点,但在纠正自身错误方面并无优势。本文的结果呼吁开展谄媚行为评估时,应探究观点的框架方式,而非仅评估是否存在谄媚行为,且需测量超越单一翻转率的多种行为。

英文摘要:

Sycophancy in Large Language Models, the tendency to align responses with user preferences over ground truth, has largely been studied using questions without a clear ground truth and with a single, plainly stated user opinion. We present a systematic study of which framings of a user's claim trigger sycophancy in factual question answering, where every answer is objectively verifiable without extra context. Evaluating 14 models on 42K questions, we express correct and incorrect user opinions through various claim framings over 8 dimensions and 21 types, and measure a broad set of opinion-driven behaviors. We find that all models are susceptible to user opinions regardless of size. Models are most influenced by how confidently an opinion is stated, whom it is attributed to, and whether it is presented as shared between the user and the model. In contrast to prior work, framings based on the user's expertise, authority over the model, or social pressure matter little in our factual setting. Larger models resist incorrect opinions better but are no better at recovering from their own errors. Our results call for sycophancy evaluations that probe how an opinion is framed, not merely whether one is present, and that measure multiple behaviors beyond a single flip rate.

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。

↑