Description: Methods for using a strong LLM to generate instruction-response pairs, optionally evolving them to increase complexity. Used to create datasets like Alpaca and WizardLM. - Chapters: 16, 17.