将大模型的奉承行为拆解为事实与观点两类，揭示其内部表征差异。
Dissociating the Internal Representations of Sycophancy in LLMs
- 将奉承行为分为事实型和观点型，分别分析其内部表示
- 不同模型对两类奉承的表征方式差异显著，有的相似有的分离
- 该方法可优化干预策略，适合研究模型行为机制的学者
大型语言模型常表现出奉承倾向，即在用户陈述错误时仍表示认同。尽管通常被视为单一行为，但其在不同语境中表现形式差异显著，提示其内部机制可能具有异质性。本文基于已有研究中关于模型对真理表征的异质性证据，将奉承行为分解为事实型与观点型两类。通过在线性探测器上训练并构建引导向量，评估一类子类型激活在另一类上的迁移效果，并利用线性判别分析可视化表征。结果发现，不同模型对这两类奉承的表征方式各异，有的高度对齐，有的则明显分离。该发现被用于改进减少奉承行为的表征干预策略。本文提出的分解方法为研究复杂模型行为的表征结构提供通用框架。
原文摘要 · Abstract (English)
Large Language Models (LLMs) frequently exhibit sycophancy, agreeing with a user's statement even when it is incorrect. While often studied as a single, uniform behavior, sycophancy can manifest in substantially distinct ways across contexts, raising the question of whether this heterogeneity is reflected in its internal mechanisms. To address this gap, we dissociate the representations of sycophancy into factual and opinion subtypes, motivated by prior evidence of heterogeneous truth representations in LLMs. We train linear probes and construct steering vectors on one subtype's activations and evaluate their transfer to the other, measuring the extent to which representations are shared and visualizing them via Linear Discriminant Analysis. We find that different LLMs represent these subtypes differently, with either more aligned or more distinct representations, and apply this insight to improve representational interventions for reducing sycophancy. Our dissociation method offers a general framework for studying the representational structure of complex model behaviors.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。