In-context learning is the ability to carry out a new task using only the instructions and examples placed in the prompt, without changing the model's weights. It became widely known through large language models, and few-shot prompting with a handful of examples is the typical form.
Unlike fine-tuning, which retrains the model, it needs little setup time. In robot AI it is also used by showing a demonstration of a person doing a task and having the robot follow it.