A prompting technique that encourages a model to reason step-by-step before producing a final answer.
Asking a model to 'think step by step' often improves accuracy on complex reasoning tasks, since it breaks a hard problem into smaller, more reliable intermediate steps.
