Chain of Thought

Chain-of-thought prompting elicits intermediate reasoning steps, which raises accuracy on math, logic, and multi-hop questions. Reasoning models internalize this with test-time compute — spending more tokens “thinking” before the final answer.

Worked example: adding ‘let us think step by step’ makes the model write intermediate steps before answering, and multi-step problems go from often-wrong to often-right — the reasoning tokens are where the computation happens. Gotcha: the written chain is a useful window, not a guaranteed audit trail — a model can produce a plausible chain that rationalizes an answer it reached another way.