Definition
Task-execution systems that combine language models with tool use, environmental feedback, and multistep decision-making.
ReAct research had language models alternate between reasoning and actions, updating subsequent steps based on external information. It illustrates one approach to combining language models with interactive environments.
This entry covers only language-model-driven agents. Agent research existed much earlier, so ReAct should not be treated as the origin of the agent concept. Task success requires environmental feedback and external verification; reasoning text produced by a model does not itself prove that execution is correct.
Principles and impact
Problems solved
- Enabled language models to use tools and observations to advance multistep tasks.
Principles
- Repeatedly read task and environmental information, select and execute an action, and use feedback to adjust the next step.
Improvements
Not recorded
Significance
- Connected language processing with task execution in an environment.
History and applications
People
Shunyu Yao
Organizations
Not recorded
Events
Not recorded
Applications
Tool-assisted question answering、Interactive task execution
Limitations
- Errors can accumulate across steps; tool permissions, stopping conditions, and result verification directly affect reliability.
Upstream and downstream relations
Upstream
Sources: ReAct: Synergizing Reasoning and Acting in Language ModelsWhat is Artificial Intelligence? — Basic Questions
Sources: ReAct: Synergizing Reasoning and Acting in Language Models
Downstream
No direct downstream relations recorded