ReAct: Synergizing Reasoning and Acting in Language Models

1 Introduction

1. What does ReAct do?
   It prompts a language model to produce reasoning traces and task actions interleaved: reasoning helps it create, track and adjust plans and handle exceptions (“reason to act”), while actions query external sources such as a Wikipedia API or an environment to bring new information into the reasoning (“act to reason”).
   • Yao et al., 2023, ReAct (The Eleventh International Conference on Learning Representations (ICLR 2023)), §1
2. What weakness of chain-of-thought reasoning does ReAct address?
   Chain-of-thought is a static black box: the model reasons only from its internal representations, ungrounded in the external world, so it can’t update its knowledge and suffers from hallucinated facts and error propagation.
   • Yao et al., 2023, ReAct (The Eleventh International Conference on Learning Representations (ICLR 2023)), §1
3. What was missing from earlier work that used language models to act in environments?
   Those systems mapped observations to actions or plans using language priors but didn’t use the model to reason abstractly about high-level goals or keep a working memory to support acting.
   • Yao et al., 2023, ReAct (The Eleventh International Conference on Learning Representations (ICLR 2023)), §1
4. What human behaviour inspired ReAct?
   People interleave actions with inner speech, e.g. while cooking: tracking progress (“everything is cut, now heat the water”), adjusting plans (“no salt, so use soy sauce”), and noticing when outside information is needed (“how do I make dough? let me look it up”).
   • Yao et al., 2023, ReAct (The Eleventh International Conference on Learning Representations (ICLR 2023)), §1