跳到正文
arXiv cs.AI· Meng-Chen Wu, Umesh Bodhwani·· 4 小时前AI 评分40

大语言模型多元对齐研究综述:目标、人类参照与偏好表示的分类框架

Pluralistic Alignment in Large Language Models: Objectives, Human References, and Preference Representations

AI 导读

一篇 arXiv 论文(arXiv:2610.24471)提出大语言模型多元对齐的两层分类法:第一层区分 Overton、Steerable 和 Distributional 三种多元主义目标,第二层区分总体人群、国家地区和子群体三类人类参照。论文综合整理了 70 篇相关文献,指出当前研究集中于 Steerable 和 Distributional 多元主义,并为各目标给出相应评估框架。

正文 · 原文

View PDF HTML (experimental)

Abstract:Pluralistic alignment aims to represent meaningful variation in human preferences rather than collapse heterogeneous judgments into a single target. Yet existing work often mixes three distinct questions: what pluralistic behavior a model should exhibit, whose preferences define the reference, and what evidence is sufficient to demonstrate success. We organize the literature around a two-layer taxonomy. Layer 1 distinguishes representing a range of human-supported viewpoints (Overton Pluralism), adapting faithfully to a specified perspective or preference condition (Steerable Pluralism), and matching the response distribution observed in a target population (Distributional Pluralism). Layer 2 distinguishes aggregate populations, countries and regions, and subgroups as human references. Preference representation, implementation mechanism, modality, and evidence quality are treated as cross-cutting design choices rather than additional forms of pluralism. We synthesize a protocol-bounded corpus of 70 works on pluralistic alignment and neighboring areas, using the objective-by-human-reference matrix to map existing approaches and identify underexplored combinations. The coded literature concentrates on Steerable Pluralism and Distributional Pluralism, while eligible Overton Pluralism work appears only at the aggregate-population level under our stated protocol. We then connect each objective to the evidence needed to support it, including perspective validity, conditional fidelity, distributional fidelity, support preservation, and uncertainty in subgroup estimates. Together, the taxonomy and evaluation framework clarify what current methods establish, where their claims exceed their evidence, and how future systems can represent disagreement without mistaking variation for faithful representation.
Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
Cite as: arXiv:2610.24471 [cs.AI]
  (or arXiv:2610.24471v3 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2610.24471

arXiv-issued DOI via DataCite

Submission history

From: Meng-Chen Wu [view email]
[v1] Sat, 17 Oct 2026 05:32:24 UTC (6,598 KB)
[v2] Thu, 22 Oct 2026 02:38:58 UTC (6,598 KB)
[v3] Fri, 13 Nov 2026 01:22:27 UTC (6,598 KB)

来源:arXiv cs.AI · arxiv.org