CoT

Chain-of-thought (CoT) prompting gets the model to reason step by step before committing to an answer, instead of jumping straight to a conclusion. The mechanism is simple: because an LLM generates one token at a time, everything it writes becomes context for what comes next. Asking it to “think out loud” lets intermediate steps — extract the numbers, apply the formula, check the units — scaffold the final answer, dramatically improving accuracy on math, logic, and multi-step problems.

For a builder, CoT is the cheapest reasoning upgrade available: often a single added line. It also produces an audit trail — when the answer is wrong, the visible reasoning shows where it went wrong, which beats debugging a bare “42.” The costs are more output tokens and higher latency, so reserve it for problems that genuinely need multi-step reasoning; a sentiment classifier doesn’t. Note that modern reasoning models bake this in, generating internal thinking before answering — with those, explicit CoT instructions matter less, but the technique still pays off with standard models.

In practice, zero-shot CoT is as simple as appending “Think step by step” or “Show your reasoning before answering.” Few-shot CoT includes worked examples whose answers demonstrate the reasoning. A production-friendly pattern: have the model write its reasoning inside <thinking> tags and the final answer inside <answer> tags, so you can parse the conclusion and log the rest. Where CoT follows one reasoning path, tree-of-thought — next — explores several.