Wikitopia
ReAct
conceptAI Method
Try in Playground →
Overview
Developed byGoogle DeepMind
Founded2022
Use casecombines reasoning and acting in language models
Technical
Protocols
Action-observation loops
Knowledge graph stats
Claims54
Avg confidence93%
Avg freshness100%
Last updatedUpdated 4 days ago
WikidataQ19399674
Trust distribution
100% unverified
Governance
Not assessed
Contribute governance data →

ReAct

concept

Reasoning and Acting paradigm that combines reasoning traces and task-specific actions in language models.

Compare with...

first introduced year

ValueTrustConfidenceFreshnessSources
2022○UnverifiedHighFresh1

primary use case

ValueTrustConfidenceFreshnessSources
combines reasoning and acting in language models○UnverifiedHighFresh1
Combining reasoning and acting in language models for complex task solving○UnverifiedHighFresh1
reasoning and acting with language models○UnverifiedHighFresh1
reasoning and acting in language models○UnverifiedHighFresh1
question answering tasks○UnverifiedHighFresh1

published year

ValueTrustConfidenceFreshnessSources
2022○UnverifiedHighFresh1

combines approaches

ValueTrustConfidenceFreshnessSources
Reasoning traces and task-specific actions○UnverifiedHighFresh1

based on

ValueTrustConfidenceFreshnessSources
Large Language Models○UnverifiedHighFresh1
Chain-of-Thought prompting○UnverifiedHighFresh1

enables capability

ValueTrustConfidenceFreshnessSources
Interactive decision making and reasoning○UnverifiedHighFresh1

founded year

ValueTrustConfidenceFreshnessSources
2022○UnverifiedHighFresh1

developed by

ValueTrustConfidenceFreshnessSources
Google DeepMind○UnverifiedHighFresh1
Princeton University and Google Research○UnverifiedHighFresh1
Google Research○UnverifiedHighFresh1
Princeton University researchers○UnverifiedHighFresh1

methodology type

ValueTrustConfidenceFreshnessSources
Few-shot prompting technique○UnverifiedHighFresh1

requires

ValueTrustConfidenceFreshnessSources
large language models○UnverifiedHighFresh1
External tools○UnverifiedHighFresh1
external knowledge sources○UnverifiedModerateFresh1

supports model

ValueTrustConfidenceFreshnessSources
PaLM○UnverifiedHighFresh1
GPT-3○UnverifiedHighFresh1
Codex○UnverifiedHighFresh1

enables

ValueTrustConfidenceFreshnessSources
interleaved reasoning and action generation○UnverifiedHighFresh1
tool use in language models○UnverifiedHighFresh1

supports protocol

ValueTrustConfidenceFreshnessSources
Action-observation loops○UnverifiedHighFresh1

integrates with

ValueTrustConfidenceFreshnessSources
large language models○UnverifiedHighFresh1
GPT-3○UnverifiedHighFresh1
PaLM○UnverifiedHighFresh1
Wikipedia API○UnverifiedModerateFresh1
search engines○UnverifiedModerateFresh1

combines

ValueTrustConfidenceFreshnessSources
reasoning traces and task-specific actions○UnverifiedHighFresh1

improves performance on

ValueTrustConfidenceFreshnessSources
multi-step reasoning tasks○UnverifiedHighFresh1

evaluated on

ValueTrustConfidenceFreshnessSources
HotpotQA○UnverifiedHighFresh1
FEVER fact verification○UnverifiedHighFresh1

alternative to

ValueTrustConfidenceFreshnessSources
standard chain-of-thought prompting○UnverifiedHighFresh1
standard prompting methods○UnverifiedHighFresh1
Chain-of-Thought prompting○UnverifiedModerateFresh1
traditional chain-of-thought prompting○UnverifiedModerateFresh1

applicable to

ValueTrustConfidenceFreshnessSources
question answering tasks○UnverifiedHighFresh1
fact verification tasks○UnverifiedModerateFresh1

compatible with

ValueTrustConfidenceFreshnessSources
Wikipedia API○UnverifiedModerateFresh1

Alternatives & Similar Tools

standard chain-of-thought prompting
alternative to
Compare →
standard prompting methods
alternative to
Compare →
Chain-of-Thought prompting
alternative to
Compare →
traditional chain-of-thought prompting
alternative to
Compare →

Commonly Used With

large language modelsGPT-3PaLMWikipedia APIsearch engines

Related entities

Claim count: 54Last updated: 4/6/2026Edit history