Chain-of-Thought (CoT)

Overview

Introduced by Wei et al. in 2022, CoT asks the LLM to produce intermediate reasoning steps rather than a direct answer. It substantially improves accuracy on multi-step math, logic puzzles, and planning tasks. Even a simple 'Let's think step by step' cue in the prompt activates CoT behavior.

Extensions

Self-Consistency generates multiple independent reasoning chains and takes a majority vote. ReAct interleaves reasoning and tool use. Tree-of-Thought explores a branching solution space. All build on the core CoT insight that explicit intermediate reasoning improves final answers.