I don’t necessarily see this as an obstacle. If you gave 80% there copy to any decent developmental editor and line editor, you could probably get to the level of any bulk genre novel on the shelves at B&N within 6 months. The content of most novels is derivative even when written by human authors and professionally edited and published. Theres very little art in those novels, as most of them tend to follow formulae for creating characters and plots and settings that are common enough to have worksheets being used in production. To be blunt, AI is going to absolutely decimate the writing industry because most of the published works are derivative and formulaic — and that’s exactly what AI is good at. Just to give an example, this (https://savethecat.com/beat-sheets) is the Save the Cat beat sheet page. Save the cat is a plot structure used by Hollywood rather extensively and is fairly common in novels. It’s also fairly specific in how a plot should be structured— down to the page number in the case of movie scripts (https://savethecat.com/beat-mapper). This isn’t people learning from experience, this is basically an algorithm for telling a story. And this is what is expected in the industry. I’m sure the there are niches in high literary fiction that are less derivative and more artistic, but this is only a very small part of the book industry, and furthermore it’s not easy to do well. One huge thing that AI detractors don’t like to admit is that AI doesn’t have to be perfect to be adopted for a purpose, it just has to create content that’s worth editing in this case. This is a pretty low threshold because of the economics— the AI is owned by the publisher and other than the software license, it’s FREE. And if the AI can produce 5000 novels that can be edited for publication, why bother with humans? If all that novels do is follow formulae then there’s no point to the human.
1) Blake Snyder was a casual friend of mine. 2) I was repped at William Morris for screenwriting. 3) I was repped at Darhansoff & Verrill for novels. 4) My letter of recommendation to grad school was written by Terry Rossio. 5) I was introduced to my agent by at D&V By Douglas Preston. 6) My novel was edited by Richard Marek. KCAL did a stand-up in 2006, 2007. Peak "Save the Cat!" time. They put a camera, a reporter and a microphone on the corner of Hollywood & Vine and asked random passers-by "how's your screenplay coming?" Eighty percent of them had an answer other than "I don't have a screenplay." And yet, none of them - not a one! Got made into a movie. I had a friend who would read eight screenplays a day. Did so for a couple years. Not a single one ever got greenlit. Is your argument that ChatGPT can do a better job than a human? Because I said they were 80% of the way there and your argument is that an editor could get them on the mass market shelves. So... all those writers out there banging away, unable to get published... are they worse than ChatGPT? It took M Night Shamalyan 9 rewrites before he figured out Bruce Willis was dead. Is ChatGPT going to figure it out in 10? 20? 30? Ever? It took Anne LaMott 20 drafts before her daddy's agent would shop around her first book. She ended up at a vanity press anyway. Are you suggesting that ChatGPT writes better than Anne LaMot? I might happen to agree; I find her tedious. Jeff Bezos tho So rather than go chapter and verse on the how and why of you being wrong, I will merely suggest that you're out over your skis, and if you disagree I suggest that you give this a try.If you gave 80% there copy to any decent developmental editor and line editor, you could probably get to the level of any bulk genre novel on the shelves at B&N within 6 months
I mean if you had the thing iterate 1000 scripts following the story beats that Hollywood likes, the chances that one of them would be interesting enough to edit is probably pretty good. And an AI could probably do 1000 in a day. It wouldn’t be AI spending 8 months writing a single draft of something, it’s AI making 1000 of them a day every day. So if AI for some reason were to generate 10,000 scripts based on a prompt like “generate a star trek movie in which Kirk defeats a god” the chances that one of that 10,000 would be interesting enough to edit into a shooting script is probably decent. Make 100,000 and I’d say there’s probably at least one that’s better than Star Trek V. This is what’s missing. The sheer scale of how much the computer can do and how fast. Yes, it took a human author 20 drafts to make something worthy of a vanity press. But an AI author bot could churn through 20 drafts in minutes, where it would take a human years to do the same. And again all of these drafts are free once you buy the AI license, where a human author will want to be paid for every successful piece they produce. It’s what a lot of people miss about AI in general. It doesn’t have to be as good as a human doing the task, it just has to be good enough that it’s no longer worth paying the premium to have the human doing that work. Depending on the field there might be reasons that you want a human involved — either for legal reasons (ai are already pretty good at reading mris, however if you want legal protection from having a trained human verify) or for luxury or premium products (there are markets for original signed art, hand made goods, etc. most people don’t care enough to pat the premium for the real thing, so they buy factory produced versions — prints instead of original art, factory made pasta instead of hand made pasta). I think the eventual shake out will be that there will be a premium book and movie market for human written material, and most bulk books will be written by AI that will be cheap and mostly disposable, forgettable stuff that people buy to read on the bus or train or while on breaks at work. There will still be luxury books written by exceptional people, but it would be the kind of books you pay a lot of money to own, and are probably collectible to some degree. Movies made by people will be more like how we see indie movies today, as something for highly educated cinema lovers who appreciate fine are. The general public wants Avengers and Star Wars and Chick Flicks, and don’t care if the movie was written by ChatGPT or similar bots. This isn’t going to happen by next Tuesday, but ask anyone who knows about technology how many times people thought a computer would never do that only to eat those words within ten years.
...I'm sorry, do you think that anyone is being paid for those drafts? That they have some sort of cost? Naaah. They're free. Cheaper than AI, they cost nothing. Wanna option one? that's a dollar. And that shit still doesn't get made. What "premium" are we talking about exactly? Okay, fine. We're gonna pay the screenwriter scale, which I think is $180k or so. That's $180k more expensive than ChatGPT. We're going to pay Will Smith $8m though so the screenwriter doesn't fucking matter. Oh, I'm sorry, did you think we were gonna make a movie with AI Will Smith? Yeah, Actor's Guild won that one, sorry. Frankly, it's worth paying the screenwriter $180k just so we can blame someone if it tanks. Where are our cost savings? That's what you're missing: the top of the top of the top of the cream of the crop are the guys who get published. Are the guys who get optioned. I was optioned, I was not published. I know one guy - ONE GUY - who had his first script made into a movie. Meanwhile, the whole of your argument - the whole of your thinking - is the infinite monkey theorem without recognizing that we don't want Shakespeare, we want a new twist on Shakespeare and ten thousand humans typing in their basements on evenings and weekends can't do what one Richard Linklater can do. We want the spark and the spark has been sucked right out of AI. Lemme introduce you to Georges Polti, who argued that there are only 36 different stories in 1895. "Save the Cat" is basically "the Hero with a Thousand Faces for dummies" which basically pointed out that from the Aborigines to the Zoroastrians, everyone has certain paradigms and motifs in drama. Reductionist thinking goes back to Beowulf; that doesn't mean fiction can be simplified to the point of mechanization. You pull this fantasy world of "luxury books" completely out of your ass as if there were any basis for any of it. There isn't. I have read more terrible writing in any given year than you have in your entire life and I'm here to say - terrible human writing often has glimmers of originality. It's what makes us keep reading. LLMs are the stochastic middle of whatever they're trained on and it is PHYSICALLY IMPOSSIBLE for them to be original. So tell me. What non "luxury book" would you read? Your mission: read 100,000 scripts to find at least one that's better than Star Trek V. They take about 45 minutes each. At 40 hours a week of reading, you're looking at 37 years to find "Star Trek 5 plus." It doesn't have to be vaguely better sometimes. It has to be markedly better, every time. And there is absolutely zero evidence or even reason to suggest that anything about it will ever be more than the mushy middle of mediocrity, by design. And the fact that you think an LLM will write better than a human some day says a lot about you. It doesn’t have to be as good as a human doing the task, it just has to be good enough that it’s no longer worth paying the premium to have the human doing that work.
Make 100,000 and I’d say there’s probably at least one that’s better than Star Trek V.