Showing a model three examples of a classification and asking it to do a fourth is in-context learning. The GPT-3 paper, Language Models are Few-Shot Learners (2020), made the ability famous.
It lets people adapt general models to new tasks in seconds. See the few-shot prompting guide.