Moreover, the ridiculous ones are unusually useful for testing the prompt, as they demonstrate the true strength of the GPT’s imagination.
Pretty sure writers care way more about craft than “imagination”. Use a reasonable premise to figure out what causes the LLM to spit out good writing, then throw in the unreasonable premise to see what happens, if you like. Anyway that’s theorycrafting. I’ll post a candidate prompt and one-shot results for my reasonable premise, the tomato premise, and the meta-absurdity premise.
Pretty sure writers care way more about craft than “imagination”. Use a reasonable premise to figure out what causes the LLM to spit out good writing, then throw in the unreasonable premise to see what happens, if you like. Anyway that’s theorycrafting. I’ll post a candidate prompt and one-shot results for my reasonable premise, the tomato premise, and the meta-absurdity premise.
Edit: https://www.lesswrong.com/posts/goq3HfxZaQcomuH6d/usd300-for-the-best-sci-fi-prompt?commentId=nG99eewTtxrPM2sEk