ReAct: Synergizing Reasoning and Acting in Language Models
Source
arxiv.org
Author
Shunyu Yao et al.
Date
Context
ReAct lets a LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → alternate between describing its next steps and taking actions that return new information. In the 2022 research, generated reasoning helps track and revise a plan, while actions query an external source or interact with an environment. For question answering and fact verification, the model uses a Wikipedia interface to obtain information during the task. The authors also evaluate interactive tasks involving household actions and shopping. The useful mechanism is the feedback loop: information obtained through an action can change what the model attempts next, keeping the evolving plan connected to the task's observations.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.