MIT study tests theory-based prompts for social agents
Published on September 22, 2025, the post describes a study using behavioral theory and small human datasets to create prompts. Agents were evaluated in related but distinct scenarios and reportedly had lower prediction error on new game variants than language models and baselines.
Published on September 22, 2025, the post describes an MIT study of prompts for social agents. The approach uses behavioral theory to construct prompts and small human datasets to adjust them. The text reports lower prediction error on new game variants than language models and baselines.
Evaluation in related but distinct scenarios is intended to test whether agents generalize beyond the data used for adjustment. The available summary does not specify the games, metrics, or result values. Consult the original post for those details, and check how the scenarios, comparisons, and measures were defined before interpreting the conclusion.