ReAct: the framework that unites reasoning and acting in LLMs
In one sentence Yao et al. introduce ReAct, a schema alternating explicit thoughts (Thought) and concrete actions (Act) in LLMs, the theoretical foundation of all modern agents.
Before ReAct, language models could reason or act, but rarely both in a structured way at once. The paper proposes a simple format: the model writes a thought first, then executes an action (like searching Wikipedia), then observes the result, and repeats.
This Thought-Action-Observation loop lets the model correct its reasoning on the fly by consulting external sources. Results on benchmarks like HotpotQA and Fever improve significantly over plain chain-of-thought.
ReAct becomes the conceptual blueprint for almost all subsequent agent frameworks: LangChain, AutoGPT, BabyAGI all replicate this logic.
Companies
Google, Princeton University
Tools
ReAct, LangChain
Tags
Sources