Do you see LLMs being able to generate a set of prompts where they can respond accurately? I'm thinking about self-augmentation and self-curation where LLMs need to know what they know before they can add content.