Few-shot Learning

In plain English

Few-shot learning means steering a model towards a new task by including a small number of worked examples directly in the prompt, rather than retraining it. The examples show the model the pattern you want, such as the format, tone, or style of answer, and it continues that pattern. It works because the model adapts to context at the moment of use, with no change to its underlying weights.

Why it matters

Few-shot prompting is a fast, cheap way to get reliable, consistent output without the cost and effort of fine-tuning, which makes it a core everyday technique.

A worked example

Common confusion

Few-shot examples do not permanently teach the model. They only shape its behaviour for that prompt; start a fresh conversation and the lesson is gone.