What is Zero-shot / few-shot?

Zero-shot is asking a model to do a task with no examples; few-shot includes a handful of examples in the prompt to improve accuracy.

Zero-shot and few-shot describe how many examples you give a model in the prompt. Zero-shot means you just describe the task: "Classify this review as positive or negative." Few-shot means you include a handful of worked examples first, then the new input, and the model infers the pattern, including your preferred format, labels, and tone. This is called in-context learning: the model is not retrained, it just conditions on the examples in the prompt. Modern frontier models are strong zero-shot performers, so start there. Add few-shot examples when outputs are inconsistent, when the task has house rules that are easier to show than describe, or when you need a strict output format. Examples consume context window and tokens, so there is a small cost and latency trade-off, and badly chosen examples can bias results.

Example

To standardize support ticket tagging, you paste three example tickets with the correct tags into the prompt, then the new ticket; the model copies the tagging pattern far more consistently than with instructions alone.

Why it matters

Few-shot prompting is often the cheapest, fastest way to boost accuracy and consistency, worth trying before you pay for fine-tuning or switch to a more expensive model. Browse the AI tools directory or the model leaderboard to put it into practice.