Neural Nets will be able to write bestselling novels by the end of 2025
24
Ṁ2606
2025
7%
chance

Resolution criteria (draft):

  • 3+ fully AI written and edited (no human engagement in text) novels on NYT bestsellers list (or if that doesn't exist or chooses to ban AI novels, some equivalent)

  • Not bestsellers on the basis of novelty, but instead seen as novels in their own right (any suggestions on making this better)

  • Book must be written from a single prompt and not somehow just editing a very long context window.

  • Little human editing

Inspired by tweet thread (note I've put them all as "Neural Nets will" so that they are all framed the same):

Link: https://twitter.com/RichardMCNgo/status/1640568776495353860?s=20

Mar 28, 11:22am: Neural Nets will be able to write bestselling novels by 2025 → Neural Nets will be able to write bestselling novels by the end of 2025

Get Ṁ1,000 play money
Sort by:

Part of the reason for going bestselling on a list (which is 10K SOLD i think? For Nyt bestseller) might be the novelty factor that it's written by AI.

I think the criteria should reflect quality instead of vitality. If it is able to write a book that wins an award (non obscure) in competition with other human written books, then that would be impressive.

Does "an author uses GPT to turn bullet points into paragraphs, or to proof read" count?
Does the book have to be written using one prompt (that mostly doesn't look like a draft of a book, but does look like "write me the next book in the Harry Potter series" ?

It's silly because this is a quesiton about libs bureaucracy.

Anyone will buy anything written by ai so the question is meaningless, it's libs ability to scam themselves.

@MarkIngraham I agree that something like "people buying anything written by ai" is the main source of uncertainty here.

predicts NO

@na_pewno most of these questions are about libs ability to scam themselves rather than the capability of ai, which still fails the SAT and basic standardized tests that are a better criteria