In order to resolve YES, someone must provide an example of such a novel. It must:
Be at least 60,000 words.
Be interesting, not some attempt to game the system like "Alice wanted to see how high she could count, so she said "1", then "2", then "3", then "4"..." It must seem like something a human might write if they were trying to write a good novel.
Not have obvious continuity errors.
Be created from a single input prompt, no further prompting or human error-correction. (If it's an automated prompting, like telling it "write the next chapter" over and over, that's fine. The prompts just can't be providing any help or suggestions.)
It does not have to be "good" in a literary sense.
https://www.vice.com/en/article/k7z58y/rie-kudan-akutagawa-prize-used-chatgpt
"Winner of Japan’s Top Literary Prize Admits She Used ChatGPT
Rie Kudan was awarded the prestigious Akutagawa Prize and promptly announced that she used ChatGPT to write about 5 percent of the winning novel."
@shankypanky I know. I don't think it will take much more to go from 66 prompts to 1. E.g. not hard to get the model to read each chapter and autoprompt itself as it goes, and context windows are already large enough. Gemini Ultra with a bit of scaffolding might be able to do it?
@GregColbourn 66 prompts is still impressive! but I initially missed the point for a single prompt for the 60k word output and we're still not there yet, primarily (imo) for the requirement of no further human error correction?
I suppose it depends on the genre and output, but I tend to invest time to varying degrees into polishing any written work I generate with, to smooth readability and structure and "humanness." that said, I'm not actively trying to avoid a bit of that so it would be interesting to try and see what it takes for my purposes to get a one-and-done output.
@shankypanky I mean, GPT-4 output is already pretty polished at the 1-2 page length (essay) range. Doesn't seem a huge step to 100x that in length.
@IsaacKing Can the AI write it's own summary of what's already happened and then use that in an autogenerated prompt to continue? The whole process would still be automatic without any human involvement, after the first human given prompt.
@LarryOBrien The original prompt can be detailed, but there can't be further steering afterwards (other than "continue" or similar to get around output length limits).