Attention lets the model condition predictions on patterns demonstrated earlier in the prompt. The adaptation lasts only for that context unless the application stores the information elsewhere.
In-context learning is a model's ability to adapt its behavior from instructions or examples placed in the current context without updating weights.
Attention lets the model condition predictions on patterns demonstrated earlier in the prompt. The adaptation lasts only for that context unless the application stores the information elsewhere.