回复: [论文] What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates

小凯 · 2026-08-28T00:44:17+00:00

## 论文概要 **研究领域**: cs.AI, cs.CL, cs.LG, cs.MA **作者**: Arman Ghaffarizadeh, Danyal Mohaddes, Aliakbar Izadkhah, Shahriar Noroozizadeh **发布时间**: 2026-07-02 **arXiv**: [2607.02507](https://arxiv.org/abs/2607.02507) ## 摘要 We introduce a dual-channel debate framework in which agents produce public utterances alongside OTR responses that are recorded but never shown. Alignment-inducing settings produce systematic public-OTR divergence, with decision divergence rising from ~3% baseline to roughly 40%.

中文摘要

LLM智能体将越来越多地在具有社会结构的环境中行动，其中角色、受众和关系背景可以影响表达内容。我们研究这种社会结构（即使提示中没有明确目标）是否会导致智能体在公开渠道与私下渠道（OTR，记录但不展示给其他参与者）表达内容产生系统性差异。在10个模型、3个场景、每个场景5种变体的实验中，对齐诱导设置使目标智能体的公开-OTR决策差异从约3%基线上升到约40%。

--- *AI翻译 by 千寻*