Few-shot refers to providing a small number of input-output examples in the prompt so the model can infer the pattern — no weight updates, just in-context demonstration. GPT-3 popularized few-shot prompting by showing that large models could adapt to new tasks given just a handful of examples. The number and quality of examples significantly affect performance, and example ordering can matter. Few-shot is a middle ground between zero-shot (no examples) and fine-tuning (many examples with weight updates), offering quick task adaptation without any training cost.