ReAct Prompting
Domain: Prompt Engineering
🎯 Motivation
Reasoning + Acting: the model alternates between thinking (reasoning) and taking actions (calling tools, searching) to solve complex tasks.
📋 Concrete Example
Thought: ‘I need to know the weather.’ Action: search_weather(‘Hong Kong’). Observation: ‘28C, sunny.’ Thought: ‘Now I can recommend outdoor activities.’ Action: generate_response().
🔗 Analogy
A detective solving a case - they think (reasoning), then act (interview a witness), then think again based on what they learned. ReAct makes the LLM alternate between thinking and acting.