Functionally slush piles are just ignored and have been for awhile. Literary agencies and publishers look for writers with a social media presence now, as it has a major impact on whether the book can be marketed successfully.

I don’t think a model will be able to handle the abstract concepts that make a good book worth reading.. it would be selecting for something predictable/boilerplate…

That said there’s zero doubt in my mind that they will be used heavily in the publishing industry.

I don't know if an LLM could make such a judgement "a lot better" in comparison to an intern, but maybe it can be used in a more useful manner? For e.g. maybe it is valuable for an LLM to tell the agency "67% probability of being similar to the mid-performing books of the last 2 years".

I don't know anything about this world, but judging a book of the length of the first paragraph seems incredibly shallow and easily gamed.

Maybe can randomly sample it instead. Would be similar to opening a book to a random page and reading it; not sure how well that would work though.

AI might be good at flagging books that deserve a second look. It can match similarly of a manuscript to commercially successful books.

Indirectly, of course, as all new fiction is written by AI, the "slush pile" problem will go away.