Glossary

Few-shot — Glossary

Providing a handful of input–output examples in the prompt to steer model behavior.

Reviewed:

Full definition

Few-shot (in-context) learning was highlighted in the GPT-3 paper. Two to five consistent examples often stabilize labels and formats without updating weights, at the cost of extra tokens.

Related encyclopedia topic

Few-Shot Prompting — Few-shot prompting includes a handful of example input–output pairs in the prompt so the model can imitate the demonstrated pattern.

← Back to glossary · Learning hub