In-context learning
A model picking up a new task from examples or instructions in the prompt, without any change to its weights.
Also: ICL
The GPT-3 paper of 2020, titled "Language Models are Few-Shot Learners", showed that a large enough model could learn a task such as translation or unscrambling words just from a few examples placed in its prompt. Nothing is trained; the model adapts within a single response. In-context learning made general-purpose models useful without fine-tuning and is the basis of prompting. How it works inside the model is still studied; one proposed mechanism involves so-called induction heads that copy patterns seen earlier in the context.