777

TL;DR

AI models often provide the most average or likely answer, leading to predictable results (the '7').
To get unique answers, give the AI your opinion first and ask it to argue against it using provided data.
When asking for ideas, request a range of options with probabilities of being common, aiming for less likely results.
Banning specific words, phrases, or styles helps refine AI output to avoid clichés.
Deeper results are achieved by asking the AI to consider what experts in the field would do, starting from raw material, and involving outside validation.
Teach AI your preferences by creating a 'taste file' detailing what you like and dislike, and setting it in the AI's profile or instructions.
Taste is subjective and evolves; AI should be trained on your current preferences, which requires continuous refinement.