LLM Social Agents Need More Than Mimicry, Two Preprints Argue
Two new arXiv papers identify a common reliability gap in LLM-driven social simulations, then propose different fixes: tuned personalities versus a Bayesian belief layer.
Large language models are increasingly used to power social agents and robots in simulations, offering more flexibility than rule-based systems. Yet as two new arXiv preprints make clear, the ability to mimic human behaviour does not automatically translate into predictable or controllable social outcomes.
The first paper asks whether personality-tuned LLMs make better social agents. It acknowledges that these models imitate humans very well, but suggests a persistent gap remains — one that explicit personality tuning may or may not close. The second paper takes a different route, focusing on opinion dynamics. It argues that LLM agents currently revise opinions implicitly in context, meaning their openness to persuasion is neither specified nor verifiable, and collective results inherit the model's training priors. Its proposed fix is a Bayesian belief layer that makes opinion change more controllable.
Taken together, the two papers agree on the core diagnosis: implicit behaviour limits the reliability of LLM social agents. Where they differ is in the cure — one adjusts the model itself through personality tuning, while the other overlays a structured belief mechanism to govern how opinions evolve.
Sources · 7
- Mind or Message? Auditing Theory of Mind in Multi-Agent Social Simulation
- Do LLMs Choose Like Humans? Using Cognitive Theory to Evaluate LLM Decision-Making
- From Certain Doom to Survival: Agent-Driven Self-Governance in LLM Agent Societies
- Triggers and Diagnostics for LLM-Based Interpretability Failures in Active Inference Agents
- Recognition, Simulation, and Refusal: A Contamination-Aware Study of Classic Psychological Effects in LLM Agents
- Bayesian Belief Layer for Controllable Opinion Dynamics in LLM Agents
- Do Personality-Tuned LLMs Make Better Social Agents?
More in Research Digest
OpenRoIS middleware decouples HRI apps from robot hardware
A new open-source middleware aims to let human-robot interaction applications run across different robot platforms without rewriting code.
4DGS-Fixer: Video Diffusion Priors for Sparse-View 4D Scenes
A new method iteratively refines 4D Gaussian splatting with video diffusion priors to improve dynamic scene synthesis from sparse-view videos.
Two papers probe RL bottlenecks for LLM agents
New arXiv work targets two stubborn obstacles in reinforcement learning for LLMs: sparse credit in open-ended tasks and merging separately trained capabilities.
Proactive Robots Must Balance Helpfulness Against Intrusiveness
Three new papers examine how robots can anticipate human actions, time their assistance, and tailor it to individual users without becoming intrusive.