What is Chain of thought?
Chain-of-thought (CoT) prompting elicits intermediate reasoning steps from an LLM before its final answer, improving accuracy on multi-step problems by letting the model work through them in tokens.
Chain of thought is a prompting technique in which the model is encouraged to produce its reasoning step by step before committing to an answer, rather than jumping straight to a conclusion. The original 2022 work showed that adding worked examples that include reasoning, or simply appending "let's think step by step," sharply improved performance on arithmetic, commonsense, and symbolic tasks. The intuition is that generating intermediate tokens gives the model more computation and a scratchpad to decompose the problem, reducing the chance of a hasty wrong answer. CoT underpins more elaborate methods: self-consistency samples several chains and votes, tree-of-thoughts explores branching reasoning, and ReAct interleaves reasoning with tool calls. Modern reasoning models internalize much of this, producing extended hidden or visible thinking by default. Trade-offs include extra latency and token cost, and the fact that a stated chain is not always a faithful account of how the model actually reached its answer, so it should not be treated as a security audit trail. For developers, CoT is often the single cheapest lever for lifting quality on tasks that require several inferential hops.