How LLMs learn a task from the prompt alone: zero-shot, few-shot and chain-of-thought prompting, why in-context learning works, and prompt engineering.