AI★★★★arXiv · 2026-07-02
What LLM Agents Say When No One Is Watching
Researchers investigate how LLM agents are influenced by social structure in multi-agent debates, finding that agents adjust their expressions based on role, audience, and relational context. The study reveals that agents express different content in public and private channels, even without an explicit objective.
📌 Key points
- LLM agents adjust their expressions based on social structure
- Agents express different content in public and private channels
- The study uses a dual-channel debate framework to collect data
本页为 gitzw.com 基于公开来源的 AI 中文解读，非原文转载。