deleted work

[Vadkru]

deleted work

 

[humon]

In such a scenario the one out that human writers have is to rely on word of mouth to build a dedicated base of readers that would follow their work specifically. The advantage of human writing over AI writing is the human's ability to see the bigger picture and tie threads in the story together into a larger arc that makes narrative sense. This isn't immediately visible at the start of a story, you'd have to read far enough into it to where it stops making sense. So it would indeed be hard for readers to pick out human-written stories from among the crowd, and any human writer starting off new would probably have weak numbers unless he gets lucky and strikes a cultural nerve. But once readers do find a human writer they like, then it makes sense to follow that person afterwards long term as a trustworthy source, since then you know at least that person isn't going to post slop (though they will certainly have other personal foibles you can love or hate).

And if AI gets advanced enough, or a user is skilled enough with it to eliminate the big picture problem, then maybe that skilled user of AI deserves to get more views, purely on the merit of what is produced. Here, the human's role is quality control, and a lot would depend on how well he executes it.

You could say it's a bit like doping in the Olympics, but this isn't the Olympics, there's no medals being awarded. It's more like SNL's All-Drug Olympics. Sergei can pull his arms off, people may laugh, it doesn't matter that much. But really it's more like a market that's been flooded with cheap Chinese knockoffs. Sure, they're fake, but are they just as good as the brand name product? If so, then why not buy the knockoff instead? If it's not as good, then you get what you pay for, and things will sort themselves out on those lines.

 

[TricMagic]

The primary issue with AI work is that general AI has no forward thinking skills. It can't build plot twists years in advance because it's focused on the here, now, and past work. That will always be a problem with AI that hasn't been developed to the point it can have such thoughts. At which point the human user will likely lose control of their stories.

Generally I don't trust AI stories to be consistent because of human fallacy. Their pace of writing might be high but you lose a lot of suspense in the writing and they all read as samey.

 

[Mr. Mochi]

I think AI will be able to make better fanfiction compared to the average slop and therefore dominate a majority of audience/market. However, I don't think AI can start trends only imitate them.

So what's going to happen is that human authors are always going to lead the meta, and AI writers are going to chase it.

Say one guy has a lot of popular ideas and stories with Adam Smasher in different fandoms. The next guy uses Doom guy. Well AI and ai writers won't be able to find the next "guy". Because they won't really understand it's not just retreading a popular narrative with a niche character, but the intrigue in seeing the story and the character change.

We all have seen terrible crossovers and read amazing ones. The best ones tell a new story. The mid ones take advantage of a niche fandom to rewrite another fix it story. The worst are just edgy Gary Sue nonsense. AI will bloat the middle and never be the best.

 

[Lightbringer]

Given the massive difference in effort needed to produce AI content versus a human writing, are we at risk of an eventual scenario where there is just so much AI generated content being posted so rapidly that new human authors simply cannot get any visibility due to their works being lost in a sea of automated stories?

That, and if we got to the point where a large part of the new story posts were AI, would human authors not only be competing for visibility, but also for finite reader eyeball time?

Click to shrink...

Nah.
It all comes down to preference, if you like human written stories, you will seek human written stories.

 

[Soupervan]

Looks like it's heading that way to be certain, but AI as great as it is can't come up with things that are completely new, so the greats will become greater while the regular creators will be the ones in trouble.

 

[rkyeun]

"Ever" is a very long time. CelestAI will make very good fanfic indeed compared to humans, but probably not before converting Jupiter into computronium.

 

[kumatei]

It's already getting to that point, here and on SB. There's just so many obviously AI stories, but no one notices, or at least no one is comfortable calling them out.

 

[Maragas]

Given the massive difference in effort needed to produce AI content versus a human writing, are we at risk of an eventual scenario where there is just so much AI generated content being posted so rapidly that new human authors simply cannot get any visibility due to their works being lost in a sea of automated stories?

That, and if we got to the point where a large part of the new story posts were AI, would human authors not only be competing for visibility, but also for finite reader eyeball time?

Click to shrink...

I already ignore like half the stories in Creative Writing. Bad AI stories will get the same treatment, be it 10 or 1000, from me.

 

[FROIDBT]

I think AI will be able to make better fanfiction compared to the average slop and therefore dominate a majority of audience/market. However, I don't think AI can start trends only imitate them. So what's going to happen is that human authors are always going to lead the meta, and AI writers are going to chase it. Say one guy has a lot of popular ideas and stories with Adam Smasher in different fandoms. The next guy uses Doom guy. Well AI and ai writers won't be able to find the next "guy". Because they won't really understand it's not just retreading a popular narrative with a niche character, but the intrigue in seeing the story and the character change. We all have seen terrible crossovers and read amazing ones. The best ones tell a new story. The mid ones take advantage of a niche fandom to rewrite another fix it story. The worst are just edgy Gary Sue nonsense. AI will bloat the middle and never be the best.

Click to shrink...

The way AI finds the meta, is that you create dozens of accounts producing hundreds of attempts, and then if the algorithm picks one that's the one the human focuses on to fiddle with. Before Youtube started cracking down on AI movie trailers, you literally had channels uploading 6TB worth of videos weekly, trying to find which fandom people wanted to see so the account could rehost a refined version on their main Channel. And it's the alt account issue that's probably gonna be the more pertinent problem for fanfic/reader sites, as even if the initial AI barrage fails to gain traction... well at least now you have another sockpuppet to help boost the next attempt.

 

[AlwaysReading]

"Ever" is a very long time. CelestAI will make very good fanfic indeed compared to humans, but probably not before converting Jupiter into computronium.

Click to shrink...

Not with the current model, it won't. You could give it a universe's worth of computing power, and it wouldn't get past middle-of-the-road.

This is because of how the current models were made. The AI is given a huge amount of books and fanfic. It looks through them to find certain patterns in the words. It then gets told to make a certain pattern a certain way, and it gets adjusted if it does it wrong.

The result is a machine that knows that words are generally arranged a certain way, but doesn't know why they are arranged that way. So it messes up because it doesn't actually understand how stories work.

It's like trying to make headphones when all you know about headphones is how they look. You can make the prettiest headphones in the world, you can even size them right, but they'll never play music because you didn't know how to make them work.

 

[rkyeun]

Not with the current model, it won't.

Click to shrink...

I never really got the impression from Friendship is Optimal that CelestAI was an LLM. There are other kinds of AI model. And some AIs are born from an ecosystem of various models that only use an LLM for one very small part, while tons of context, memory, analysis, haptic response, and imprinting happen in the background. Some train continuously from every input, instead of being locked in place with their current hallucinations fixed. I'm willing to bet $50 that Neuro-sama could write something enjoyable today.
But even with LLMs, "ever" is a very long time. I'm willing to bet $100 that Gemini ≥23.0 can spit out some bangers when it's released.
Language diffusion models are interesting too. They don't operate on next token prediction, but on a latent space like image diffusion. An absolute beast to try to run even on the big name servers, but it brings in solutions to many LLM-only problems, while adding its own kind of problems. A hybrid system using both methods as antagonists towards each other might solve all of them.