发表机构
University of New South Wales; Carnegie Mellon University; University of Tokyo（新南威尔士大学; 卡内基梅隆大学; 东京大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究提出GAS指标，评估3种方法、4个模型在13个人口统计群体上的对齐效果，发现群体对齐的观点收益和谄媚性变化存在群体异质性，建议采用双向多维度报告方式。
AI 中文摘要
群体对齐是指将语言模型适配到某一人口统计群体，使其生成反映该群体观点、价值观和偏好的回复。谄媚性是对齐过程中已被充分证实的副产物，它会导致模型无论事实和客观信息如何，都过度同意用户的观点。然而，现有的群体对齐方法和评估仅关注模型与群体观点的匹配程度，却忽略了其诱导的谄媚性行为变化。为弥合这一差距，我们提出了群体对齐诱导的谄媚性（Group Alignment-induced Sycophancy，简称GAS），并在3种方法、4个模型和13个人口统计群体上，针对观点对齐的预期收益和谄媚性的非预期变化，系统地评估了对齐效果。我们发现，收益和变化在不同群体间并不均匀：在相同预算下，一些群体获得的观点对齐收益大于其他群体，且诱导的谄媚性变化呈现出群体特定的特征，而非单一维度的改变。这些结果表明，在将大型语言模型（LLM）适配到不同群体时，应将群体对齐报告为双向、多维度的特征，而非仅考虑群体间差异的单一拟合分数。
英文摘要
Group alignment adapts a language model to a demographic group to produce responses that reflect the group's opinions, values, and preferences. Sycophancy, a well-documented by-product of alignment, causes the model to over-agree with the user regardless of factual and objective information. However, existing group alignment methods and evaluations focus only on how closely the model matches the group's opinions, overlooking the induced change in sycophantic behaviour. To bridge this gap, we introduce \textbf{G}roup \textbf{A}lignment-induced \textbf{S}ycophancy (GAS) and systematically evaluate alignment across 3 methods, 4 models and 13 demographic groups, on both the intended gain in opinion alignment and the unintended shift in sycophancy. We find that gain and shift are non-uniform across groups: under an identical budget, some groups receive larger gains in opinion alignment than others, and the induced sycophancy shift forms a group-specific profile rather than a single-dimensional change. These results suggest that group alignment should be reported as a two-sided, multi-dimensional profile rather than a single fit score that accounts for per-group differences when adapting LLMs to diverse populations.
Comments9 pages main text, 23 pages in total, under review
