Agent Sociodrama: Social-Dynamics-Guided Persona Recasting for LLM Agent Teams
The paper tests whether tracking social dynamics can help repair failing teams of LLM agents.
Agent Sociodrama uses an external LLM judge to update each agent’s emotional state and asymmetric relationships after every round of interaction. The authors say those trajectories match expected social-theory patterns and human judgments across six model backbones and five scenarios. They then use the traces to recast personas before rerunning tasks, reporting gains on four collaborative benchmarks, including up to 83.3 percentage points over the initial casting. HF Daily Papers' note
Agent Sociodrama uses an external LLM judge to update each agent’s emotional state and asymmetric relationships after every round of interaction. The authors say those trajectories match expected social-theory patterns and human judgments across six model backbones and five scenarios. They then use the traces to recast personas before rerunning tasks, reporting gains on four collaborative benchmarks, including up to 83.3 percentage points over the initial casting. HF Daily Papers' note
score 4