Part 6/8:
He suggests that defining a clear standard—like always asking GPT-3 to "write a writing prompt for..."—could help systematize this process. Moreover, by repeatedly feeding prompts into GPT-3, one could generate increasingly complex or refined prompts, expanding the creative toolkit.
Future Potential and Conclusions
Shapiro concludes that the core insight for meta prompting is deceptively simple: instruct GPT-3 to "write a prompt." This basic instruction, when used repeatedly, can build a recursive system of prompt generation. He notes that while it may not produce infinite streams automatically, it provides a powerful foundation for developing autonomous prompt chains that can be used for brainstorming, training writing models, or exploring complex issues.