Reproduce “ReSpAct: Harmonizing Reasoning, Speaking, and Acting Towards Building Large Language Model-Based Conversational AI Agents” 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

Vardhan Dongre, Xiaocheng Yang, Emre Can Acikgoz, Suvodip Dey, Gokhan Tur, Dilek Hakkani-Tür

ReSpAct lets an LLM agent speak with a user, incorporate replies into its context, and continue reasoning and acting. It can ask for guidance or verify assumptions. In AlfWorld, GPT-4o ReSpAct scores 85.3% success versus 83.6% for ReAct* with location priors, across six prompts and 134 unseen games; this advantage is not universal, and benchmark scope and quality limit generalization.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.