On AI slop...

Hvast

Really Really Experienced
First. I want to make it clear that I am neither a bully nor a witch-hunter. I don't see AI use as unethical or evil by itself (even if some uses of AI are both, but it is true for any tool)

Second. I use LLMs to enhance my writing. I will not go into the detail of how I use it but be assured it is not "write me this" or "hey, here is my chapter, improve and rewrite it". I am fully aware that every time I delegate any creative decision (even a tiny one) to the AI, I dilute my own voice. But grammar and sentence structure were never my strong point - especially in English.

Third. I don't care if you use AI heavily in any form of art. As long as you are still making a lot of key creative decisions full of intent and vision. Dilution by AI is not that much of a problem when core things are not diluted.


Saying that. THERE IS TOO MUCH OBVIOUS AI SLOP ON THIS SITE. And I am getting slightly annoyed.

The most common type of slop I see is slightly edited roleplaying sessions from one of the many character.ai knockoffs (likely those advertized on the site). Guys. I understand the logic. I do. You enjoyed something. You want to share. After all, it is how most amateur stories are written - someone writes something for themselves and then go "I might as well share it." And many of those human-written stories ARE of very poor quality (I cringe when I look at stuff I wrote and had the audacity to post years ago. Nowadays I am at least mediocre.). The difference is that while those are bad, they are bad in a unique way and sometimes you can see unpolished brilliance. Plus it is a learning process. Your roleplay sessions aren't that. They are typical, average output of low-parameter models, full of style that everyone has encountered for thousands of times. And usually it is accompanied by very mundane, typical scenarios. And you can't easily fix it by light editing. Unless you are already a quite good writer, you just can't do much by editing. My suggestion: if you really like some roleplay session and want to share it... Rewrite it from scratch with your own words, and your own style, improving what you think may be improved.

Another type of AI stuff I tend to suspect is coherent stories that look like they were written by Grok (why Grok? Well, it is not too picky with adult content in inputs and outputs). They are too coherent for raw outputs and look like the result of "hey, rewrite this text in different words." prompts. And here comes a problem: I can't even tell whether someone used AI on their own text to 'enhance' it or if they are using works of other people to 'make their own stories'. In both cases, we can get quite good stories told in a bland AI-way (and this makes them not good). And the second case is one of the uses I am very against. Using AI to (almost) copy someone else's work is not cool.

Sometimes, there are also extremely lazy outputs that look like literal "Hey, AI, write a chapter continuing this story in context" and posted as is, with all crudeness and plot inconsistencies, with zero editing and probably no iterations. I have absolutely no idea why people are doing this. It is not like you are fooling anyone; even if people won't detect it is AI, they will detect that it is not good at all. If you do this, I am begging you - STOP. “If I wanted to see how an AI would continue a story, I would go and generate it myself. And would use a prompt that will fit my taste better.

Either don't use AI at all or learn to use it to supplement your creativity, not replace it.

PS. Speaking of detecting AI. Don't overevaluate your ability to do so. I am fully aware that many stories I suspect as AI or AI-assisted may be fully human-written. Neither human judgment nor software are reliable enough to be used for accusations. It is why I resisted the temptation to point at examples I suspect. You also should not overvalue your ability to hide AI use by cunning strategic edits like deleting em dashes.

PPS. Sorry if this outburst is too emotional or judgemental. I am generally pro-AI but some recent reading sessions on chyoa were VERY disappointing.

PPPS. I am also aware that few of slopers read the forum or would listen but if I can make one person to rethink their AI use, it was worth the time writing it.
 
On AI detection: I did decide a little while ago to start paying for a monthly Pangram subscription, and I installed the browser extension. I have a lot of faith in its accuracy for samples above 500 words--mostly because it always pegs my own samples as 100% human written. I only use it for my own personal edification; I don't plan to call out stories either. But, yeah...it feels like 70% to 80% of the time I feed a recently-added chapter from CHYOA into it, it comes back AI.
 
The first three chapter of High School fantasy redux I ran through Chatgpt for additional proofreading after using Grammerly. It refused to check the fourth as it involve teenagers and sex at party. That was the last time I used it. I use Grammerly for additional checks but that's the extent of me using AI with anything I write.
 
100% serious: mediocrity on an upwards curve is something we should all aspire to.
I am approaching half a million words at chyoa + at least as much in other places... Reaching mediocrity with this amount of written spam is not that much of an achievement :D

I have a strong itch to write something serious and high-effort using all accumulated experience. But if it happens, it will not be porn and it will not be in English.
 
I am approaching half a million words at chyoa + at least as much in other places... Reaching mediocrity with this amount of written spam is not that much of an achievement :D

I have a strong itch to write something serious and high-effort using all accumulated experience. But if it happens, it will not be porn and it will not be in English.

Currently I'm about 160 chapters and 100,000 words away from having written more than I did on my previous account. My previous account was stuff I wrote on and off over about ten years or so. That I've almost written more in under six months is something I'm still surprised at.
 
AI has become a very useful tool for me. It can't actually write, at least Chat GPT can't, which is the one I use. But used correctly, it's a great tool.

I plan the hell out of my stories and it takes me forever to get to the point of actually writing. AI has helped speed that up with simple stuff like fact checking my character bios and helping locate the right geographic setting for the story (for countries/places I'm not familiar with). And other stuff like 'I want this character to be a civil engineer, he's 40 and born in London, what would his education and entry into the professional world be?'. I can, and did do all that research myself, it's all there online. But AI just drops it in my lap in seconds. Frankly, it's madness not to use for that stuff. 90% of what i produce during planning is background info I'll never use on the page, but that was the same when i was doing the character research myself anyway.

I played with the idea of a time travelling character, and wanted them to meet famous people from history. AI was great and telling me who was famous in south America between 1800 and 1850 etc. The time it saves is fantastic.

I find it's editing hit and miss, I prefer Grammarly and my own eye, but if it could edit well I'd consider using it.

Where it's also become handy is with continuity, which surprised me. I can get it to check a doc and make sure I haven't messed up characters ages, or missed one of the changes I made, where my civil engineer is now a bricklayer and I've still got him at Uni learning engineering. AI can comb through a 100k word manuscript and find that stuff quickly. It's far from perfect, but it can be very handy.

And it's very helpful to have it to answer questions while I'm writing. How long does the train take from London the Manchester and what stations would be used? What's the next town North of Hull. Give me a Irish name for a bar tender. All that stuff that would have me stop writing and trawl the internet is now just a simple AI question and I'm back at the key board banging out the story.

AI has increased my output by a huge amount, without actually writing a word of it.

And as for those who do write stories with AI. Look, if it's good and i don't notice the difference then i don't give a shit. If it's crap, I'm not going to read it anyway, and I really don't give a shit. There's a lot of very poorly written stuff on every site like Chyoa, I'm not going to get in a tizzy because the crap stuff was written by AI and not a person. Crap is crap.

My only real issue is if we get to the point where AI is writing very good stuff that real authors who actually write can't match or keep up with, especially is it's passed of as written by the author. I don't think we're anywhere near there yet, but it probably isn't far off. I'll reassess my thoughts on AI at that point.
 
And as for those who do write stories with AI. Look, if it's good and i don't notice the difference then i don't give a shit. If it's crap, I'm not going to read it anyway, and I really don't give a shit. There's a lot of very poorly written stuff on every site like Chyoa, I'm not going to get in a tizzy because the crap stuff was written by AI and not a person. Crap is crap.

My only real issue is if we get to the point where AI is writing very good stuff that real authors who actually write can't match or keep up with, especially is it's passed of as written by the author. I don't think we're anywhere near there yet, but it probably isn't far off. I'll reassess my thoughts on AI at that point.

I almost agree with this approach. Almost. Slop makes this site worse. Harder to navigate. Probably slower and more expensive for Friedman. If someone knows that what they produce is lazy shit, they should not pour it on the public dining table just because it is easy to do and people can choose not to eat it. Yeah, humans write shit, too. But it is their learning process; we should not be too harsh to people who just can't make it better. Also, they are awful in somewhat unique ways. It is still a creative contribution to the collective body of works of humanity.

And the problem of plagiarism... Not the mythical one of "AI was trained on other people's works, it can't produce anything but plagiarism." Actual, real one. AI is, sadly, a great tool for taking other authors' work and creating derivatives from it with prompts like "translate it to English" "Rewrite this in other words." "Write a story based on this but change names and move it from New York to Paris." or other more elaborate prompts.



As for AI becoming better than humans... If AIs will achieve the level of being better than skilled humans in tasks like fiction writing, then we will also have very reliable AI-powered tools that will be able to detect and verify when we have true human writing, so everyone will be able to choose "organic" writing. Also, our society will be changed so much by AI that it will be the least of our worries.
 
Quality-wise, I don't believe LLMs will ever become super-human. After all, the technology inherently works by doing a great big averaging out of all of its inputs in order to generate a model of which words belong together. That can never produce anything that's better than the sum of its parts. And models need an absolute ginormous amount of data to function (tens of terabytes), so you can't use only the absolute best-written, most factually-accurate, most aesthetically pleasing prose humanity has ever produced: there's just not enough of that out there. You're always going to have to resort to using very large data sets that contain some good writing, some bad writing, but mostly just a lot of really average writing, which means that the outputs will mostly be average, too. At best there will be fewer and fewer obvious tells like em dashes, delving, and "it's not X, it's Y".

(This is mostly about art, in all its forms and shapes. There are other use cases that LLMs may improve continually at, particularly in areas more suited to formal verification or objective quality assessment.)

There's also the matter of quantity. As a very rough estimate, a human writing a story usually writes somewhere between 300-600 words of prose an hour (for non-writers: 1 page in a paperback holds about 300-350 words). Maybe you can do more, let's say 1000 words an hour. But literally no one can consistently write 10,000 words an hour. For example, NaNoWriMo set a goal of 50,000 words in a month, and it strongly emphasized quantity over quality. There's also thinking about the story and scenes you want to write and at least one round of editing your draft, but I'll be generous and assume vibers do that too, so I won't count that.

But an LLM has absolutely no issue producing stuff at that pace, hour after hour, day after day. And of course, when using LLMs, most of the human time and effort goes into crafting the prompts, and nothing stops you from just running the model twice. Or 10 times. Or 100. And so sites like CHYOA or Rule 34 are absolutely flooded with similar-looking slop, completely drowning out all the painstakingly human-created stories, images, and videos.

It's that, more than anything, I think, that puts people off and makes them rebel against AI. If only 10% or 20% of the CoNtEnT out there was generated by LLMs, it wouldn't be nearly as annoying as having to plough through page after page of the stuff. But it's so easy to do that it's closer to 80%.
 
Back
Top