few shot prompting

Few-shot prompting is a way to steer a model by including a small number of examples inside the prompt.

The model weights are not changed. The behavior is shaped at inference time by showing the model the pattern to follow.

A few-shot prompt usually contains:

instruction
example input 1
example output 1
example input 2
example output 2
new input

The model infers the task format from the examples and applies it to the new input. This is useful for formatting, classification, extraction, style transfer, and structured outputs.

Few-shot prompting is often the fastest way to test whether a model can do a task before fine-tuning. If a task works with a few clean examples, fine-tuning may not be necessary.

The tradeoff is context cost. Examples consume tokens, and behavior may be unstable if examples are poorly chosen or conflict with the instruction.