Definition
A prompting technique that forces an LLM to generate intermediate heuristic steps before arriving at a final answer, significantly improving performance on complex reasoning, planning, and multi-step retrieval tasks in AI Agents. While it increases accuracy and interpretability, the trade-off is higher token consumption and increased inference latency.
Focuses on the explicit sequential reasoning process rather than the direct mapping of input to output.
"A student's scratchpad showing every line of a mathematical proof before the final Q.E.D."
- ReAct (Reasoning and Acting)(Component/Refinement)
- Self-Consistency(Optimization Technique)
- Few-Shot Prompting(Prerequisite)
Conceptual Overview
A prompting technique that forces an LLM to generate intermediate heuristic steps before arriving at a final answer, significantly improving performance on complex reasoning, planning, and multi-step retrieval tasks in AI Agents. While it increases accuracy and interpretability, the trade-off is higher token consumption and increased inference latency.
Disambiguation
Focuses on the explicit sequential reasoning process rather than the direct mapping of input to output.
Visual Analog
A student's scratchpad showing every line of a mathematical proof before the final Q.E.D.