Few-shot learning is a model's ability to perform a task from only a small number of examples or demonstrations.
The term is used in model training, prompting, and evaluation when teams need useful behavior without large task-specific datasets.