Reproduce “Joint Workflow and Prompt Optimization for User Behavior Simulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Nipun B Nair, Tongtong Wu, Hongzhi Yin, Hui Li, Weiqing Wang
SWORD jointly optimizes agent workflows and natural-language prompts for user simulation, using prediction errors from labeled examples to revise both. It starts with generic instructions, without task-specific initialization or domain-expert input. In a controlled Mask Adoption test, its mean RMSE was 0.0217, versus 0.1795 for staged optimization, a statistically significant difference. On Agent Society, its advantage over prompt-only optimization was not significant.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.