跳到正文
arXiv cs.AI· Isaac Song, Mohammed Rehan Parwani, Glenn Matlin, Emile Anand, Akhil Theerthala, Arjun Chatterjee, Anthony Wen-Ming Zang, Maria Kostylew, Yonadav G. Shavit, Sebastien Krier, Mark Riedl·· 4 小时前AI 评分31

语言模型角色引导用于社会模拟:CastVectors 与角色级筛选工作流

Role Steering of Language Models for Social Simulations

AI 导读

研究提出一套面向角色条件智能体的激活引导筛选流程:定义角色画像、提取角色特定方向、扫描四个引导系数并评估对齐度,在 OLMo-3-7B-Instruct 上对 275 个角色、228 个角色无关问题完成测试。

正文 · 原文

Authors:Isaac Song, Mohammed Rehan Parwani, Glenn Matlin, Emile Anand, Akhil Theerthala, Arjun Chatterjee, Anthony Wen-Ming Zang, Maria Kostylew, Yonadav G. Shavit, Sebastien Krier, Mark Riedl

View PDF HTML (experimental)

Abstract:Social simulations built from language-model agents need role-conditioned behavior that can be checked before agents are placed into a simulated population. We introduce an activation-steering screening workflow for role-conditioned agents: define a role profile, extract a role-specific direction, sweep four steering coefficients, evaluate role-profile alignment, and pass or flag each candidate configuration. On OLMo-3-7B-Instruct, we apply the workflow to a mixed 275-role inventory with 228 role-agnostic questions, GPT-4.1-mini prompted role references, and GPT-4.1-mini judges. We also introduce CastVectors, which receive higher judged role-profile alignment than an assistant-axis directional control from prior persona-vector work, with mean overall scores of 63.2 versus 41.1 across the tested grid. They also preserve high lexical diversity, while the control drops sharply at larger coefficients. The role-level screen is the main contribution: most roles improve as steering increases, but 38 roles decline across all six measured dimensions, showing why simulation builders should choose coefficients per role rather than deploy a uniform high-strength setting. Our code is available at this https URL, and the role vectors and evaluation data at this https URL.
Comments: 34 pages. Published at the Social Sim'26 Workshop, COLM 2026. Code: this https URL
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
ACM classes: I.2.7; I.2.11
Cite as: arXiv:2608.00023 [cs.CL]
  (or arXiv:2608.00023v3 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2608.00023

arXiv-issued DOI via DataCite

Submission history

From: Glenn Matlin [view email]
[v1] Thu, 9 Jul 2026 22:07:27 UTC (529 KB)
[v2] Thu, 6 Aug 2026 06:50:50 UTC (2,077 KB)
[v3] Tue, 17 Nov 2026 04:15:01 UTC (2,083 KB)

来源:arXiv cs.AI · arxiv.org