in-context learning
When a model picks up a new skill from examples placed in the prompt, without any change to the model itself.
There are two ways to teach a model something new. You can retrain it, which changes the model’s internal numbers and costs time and money. Or you can put a couple of examples straight into the prompt and let the model work out the pattern for itself. The second one is in-context learning, and it is the reason “here are three examples of the format I want” works so well as a prompting habit.
Nothing is actually learned in the permanent sense: close the conversation and the skill is gone, because it only ever lived in the context window. That is a limitation and a feature at once. It costs nothing, it is instant, and it is now spreading beyond text. Generalist AI’s GEN-1.5 does the same trick with robots, loading a short video demonstration into the context window instead of retraining on hours of footage.