ReAct: Synergizing Reasoning and Acting in Language Models

Large language models have shown strong capabilities in language understanding and interactive decision making, but their reasoning abilities (such as chain-of-thought prompting) and their acting abilities (such as action plan generation) had largely been studied in isolation. ReAct explores using LLMs to generate reasoning traces and task-specific actions in an interleaved manner, creating synergy between the two: reasoning traces help the model induce, track, and update action plans and handle exceptions, while actions let it interface with external sources such as knowledge bases or environments to gather additional information.

Across diverse tasks ReAct outperforms strong baselines and improves human interpretability and trustworthiness. On question answering (HotpotQA) and fact verification (Fever), interacting with a simple Wikipedia API lets ReAct overcome the hallucination and error propagation common in chain-of-thought reasoning, producing more interpretable, human-like solution trajectories. On the interactive benchmarks ALFWorld and WebShop, ReAct outperforms imitation and reinforcement learning methods by 34% and 10% absolute success rate respectively, while using only one or two in-context examples.