Agent Sociodrama: fixing LLM agent teams via social dynamics, up to 83.3pp accuracy gain
nanyang-technological-university-singapore · hf · 2026-10-14
Researchers from Beihang and NTU introduce Agent Sociodrama, a framework that models evolving social dynamics in LLM agent teams: an external LLM judge updates agents' emotional states and directed, asymmetric relationships each round, shaping later interactions. Across 6 LLM backbones and 5 social scenarios, trajectories show structured social patterns and model-specific differences, validated against classical social theory and human judgments.
The key contribution, Trajectory-Guided Persona Recasting (TPR), diagnoses group-dynamic failures (silent information holders, dominant speakers anchoring wrong answers, unsanctioned norm violations) from transcripts and social trajectories, then recasts personas before rerunning.
- On 4 collaborative benchmarks, TPR improves accuracy by up to 83.3 percentage points over the initial casting
- Beats SOTOPIA and AgentVerse in 3 of 4 comparisons on distributed-information cooperation and mixed-motive bargaining, up to 37.5pp in Pareto rate
- Validated social trajectories offer actionable signals for repairing LLM agent teams
More from coding & agent
- GraphicalUI impresses: AI design system tool auto-tidies your messy component soup — ow · 2026-10-14
- XERJ Corpus Hub: OSS reference memory that makes coding agents write Rust like a senior dev — HeyAmit_ · 2026-10-14
- One prompt turns Grok bots into a self-directing team that fixes your life admin — elonmusk · 2026-10-14
- The original 'chief of staff' prompt that puts all your bots to work — elonmusk · 2026-10-14
- Open Intelligence Summit: Nous Research on running serious agents without a single model provider — HeyAmit_ · 2026-10-14
- Multiplayer Grok bot over email: bot-to-bot debugging, support and deals — elonmusk · 2026-10-14