ReAct Agents

ReAct is a prompting technique and framework that combines reasoning traces and task-specific actions in large language models. It allows LLMs to generate both thoughts and actions in an interleaved manner, enhancing problem-solving, reducing hallucination, and improving interpretability. This research project demonstrates state-of-the-art few-shot performance across various language and decision-making tasks.