Dissecting Literotica’s AI Policy

[alohadave]

Could this all not be solved by allowing the stories to be published but identified as having AI used and allow the readers to decided?

That would open the floodgates to a ton of AI submissions. It would be like saying "we welcome AI, flood us with submissions."

 

[YmaOHyd]

I do not want to share space and compete for eyeballs with "Generate a story of between 10,000 and 15,000 words about Ashley and Sam having sex at a lake, DD boobs, explicit, porn, fucking, romance".

 

[DeusAudivit]

In the hiatus of not writing, as for whatever reason my works are not currently getting published, I’ve seen numerous theories here in AH on Lit’s current crisis and the assumption it’s related to a spike in AI driven content/volume.

There’s also the daily well known gripe posts/panic threads relating to “my story was rejected for use of AI”. Which are always contested with claims that no AI has ever been used.

Then there’s the issue on translation for the community who are not English speaking/typing and the complaint that translation tools are flagged as AI content (which they invariably are by necessity)

As in many other daily touch points AI is here and it’s not going away, it’s a threat to genuine writers and their originality. Its threat will only increase so a clear policy on AI is essential

Many around here might have overlooked or be unfamiliar with Lits actual AI policy that’s hidden away in the sites FAQ’s

it’s a wonderfully ambiguous policy that’s worth dissecting… so I gave it a go and added my personal thoughts on the policy that’s being used to reject story’s

**********************

Literotica does not currently have an official comprehensive policy on Artificial Intelligence (AI). We are waiting to see how this technology develops, and getting feedback from our community, before creating any comprehensive AI policy. However, we have put together the following points on where AI fits into Literotica at this time.

Well there’s a great place to start - it’s there in black and white; there is no ‘official comprehensive policy’ on AI usage so we start from a place of massive hypocrisy

1. Literotica is a storytelling community based around the publishing of humanadult fantasies. While AI tools - spellcheck, grammar tools, autocomplete, etc. - have long been used to help Literotica authors write their stories, the fantasies themselves come from the creative efforts, experiences, and fantasies of the real people who make up the Literotica community. As writing tools continue to evolve, we do not foresee a future where machine fantasies will replace the real experiences and creative efforts of Literotica’s community of human authors.

Pick from this what you can but there appears to be an acceptance that AI style tools to check grammar etc have long been used - if that’s the case why penalise those that use them, are we better off with little to no grammar and stories littered with isssues (personal note: as a dyslexic I rely heavily on word to check my spelling and grammar)

2. Literotica explicitly does NOT grant any person or entity (commercial, non-profit, or other) the legal right to train AI on any works published on Literotica. Each work published on Literotica is copyrighted by the author. Before using any work on Literotica for any purpose (including training AI or any other AI-related use) you are required by law to contact the author to request permission to use that work. Using works on Literotica for training AI without legal authorization may subject you and your AI (and any work generated by your AI) to future lawsuits from the original author(s), Literotica, or both.

Fair enough

3. We are monitoring the various ethical concerns around AI tools (some of which we have been contacted about directly from members of the Literotica Community). We plan to continue closely watching the development of AI, along with the development of public policies around AI, before creating our own official policies.

When will this policy be official then and until such time as it is official why are writers being rejected on the basis of AI if there is no official policy on AI?

4. Literotica’s Publishing Guidelines are clear - you must certify that you are the author of AND you own the copyright to any work published on Literotica. While simple AI tools (spelling and grammar tools, for example) do not usually interfere with an author’s copyright, there are unanswered questions around copyright when using some of the latest AI technologies that generate large blocks of text. If there are any questions about copyright related to any work you’ve used AI tools to help you create, we ask that you research and be 100% sure you own the full rights to the work before attempting to publish the work on Literotica. If you publish a work on Literotica to which you do not fully own the copyright, it may open you up to future legal repercussions.

Seems that this is a copyright issue not an AI issue

5. Literotica’s own use of AI is currently limited to improving the way we recommend related works to readers.

????

What is Considered an AI Generated Story at Literotica?

Here we go, this will start to clarify matters, surely

With the proliferation of “smart” writing software and spelling/grammar apps in the last few years, questions arise about whether a story using these tools should be considered AI generated or human written. If blocks of text are being generated by a machine, that’s clearly AI writing. If software helps correct spelling and grammar in text that a human author has written, that’s probably not an AI generated story.

Fair enough on the first point although the ambiguity and contradiction in the lack of a policy is jarring in relation to software that helps correct spelling and grammar

One area of particular concern is software or apps that “rewrite” your paragraphs or stories for you. The text that results from rewriting features is definitely AI generated. These type of apps are replacing original human written text with generic AI generated text, and should be avoided when submitting work to Literotica. Readers prefer to experience your worlds and your fantasies in your own unique voice, rather than having them smoothed into a generic artificial voice available to everyone else using the same software.

Valid point, but in the absence of an official policy how do those who pass judgement justify what’s unique from an author and what’s been generated by an AI. This is surely being based on an assumption that an author doesn’t type the same words/sentence /paragraph/chapter in the same manner as an AI might or could

If you have questions, suggestions, or comments about Artificial Intelligence and how it might impact the future of writing, we recommend that you visit the Literotica Author Support Forum to discuss the issue with other published Literotica Authors.

Yeah OK but without casting too much shade communication is a one way street around here, but let’s leave that complaint where it is for now


********************


And that’s it, that’s the policy (or lack of policy)

Feel free to dissect and interpret as you please and let the comments below descend into the usual AH chaos

If you only take one thing from this post/thread please let it be that as a writer you may fall foul of an AI Policy that is not an ‘official comprehensive policy’

You could conclude, it’s all just being made up as it goes along and is wheeled out as a very convenient broad brush excuse.

A very interesting insight. I am Dyslexic also and without a spell checker my writing is gibberish.

 

[qidniz]

My two cents, as a long time reader.


1. As a rule, AI should be frowned upon, because the product is not _creative_. So systems, where the author is allowed to mark the "help" of AI as part of the submission process, will probably not work in practice, as they will encourage submissions of this kind, and more importantly, still leave open the question of how much AI is "reasonable".


2. Spell check and grammar check are not AI in the classic sense of the term: they are relatively primitive "expert systems", using sets of preprogrammed rules. They are not _adaptive_, which is a distinguishing feature of AI.


3. We're still in the early stages of widespread AI use. Things will take time to settle down. Literotica's generally conservative approach is sound, IMO. But it remains vague as to outcomes, which can have the taint of seeming arbitrary. if they are to have a flagging system at all. it needs to be transparent.


4. The problem is that any algorithmic approach over time comes to be game-able, if not actually gamed. So, e.g,, if Literotica announced which engines they use to detect AI, the defects of these engines _will_ be exploited. The best - by no means optimal - is probably an adaptive process of reader feedback, based on the idea of "reputation" ratings.


5. Spitballing, a system like the following has a fair chance of working (until someone figures out a way of gaming it, of course):


(a) Report feedback must have concrete evidence: link(s) to the engine(s) used to detect possible AI, so that the report can be duplicated (verified). Any report to the effect "this feels like AI" will be ignored (and may adversely affect "reputation" - see below.)


(b) Reports are ranked by _both_ the "reputation" of the reporter and that of the engine(s), combined into a score. A median (not mean!) average score above a threshold earns the flag.


(c) "Reputation" is scored by performance history. Cry wolf too many times, and you won't be taken seriously.


That's the broad outline, The devil's in the details, of course.

 

[MillieDynamite]

But what is the difference between sharing eyeballs with bad writing as opposed to AI writing? The sit is filled with horribly passive-voiced dull stories. Do you think we should kill all stories that can't score a 4.0? But wait, a lot of 2.0 stories are actually good; they just offend one segment of the readership or another. HUM... What do you suggest to avoid your having to share eyeballs with those you consider inferior?

I do not want to share space and compete for eyeballs with "Generate a story of between 10,000 and 15,000 words about Ashley and Sam having sex at a lake, DD boobs, explicit, porn, fucking, romance".

 

[StillStunned]

But what is the difference between sharing eyeballs with bad writing as opposed to AI writing? The sit is filled with horribly passive-voiced dull stories. Do you think we should kill all stories that can't score a 4.0? But wait, a lot of 2.0 stories are actually good; they just offend one segment of the readership or another. HUM... What do you suggest to avoid your having to share eyeballs with those you consider inferior?

The difference is the ease with which the poor writing can be produced. Even the worst-rated human-written story on the site took at least several hours to write, and almost certainly left the writer drained for the rest of the day. This limits the number of stories that they can submit.

On the other hand, someone could enter a prompt and submit their AI-generated story within a matter of minutes - and then repeat the process again and again.

That's a flood that will drown out the human-written content. It's already difficult to get views in several categories. Imagine if there were ten times as many stories for readers to go through every day. And sooner or later, if they see that 90% is soulless AI-generated slop, they'll give up and nobody at all will read our stories.

 

[dirk2024]

I am just recently back to writing and it is because of the "AI tools". My spelling and grammar are not the best. I have issues that made it hard for me to learn some of this. I use the tools to compensate for this issue. I understand that this goes against the sites policy. Could this all not be solved by allowing the stories to be published but identified as having AI used and allow the readers to decided? In the end, I have always seen this site being about the reader/writer experience. As a reader, I can choose the type of stories and the writing styles I like to read. As a writer, these tools give me the ability too express myself in a way that works for me. I KNOW AI and the tools are not going away and are only going to become more prevalent (thank god for spell checker). I can understand wanting to block stories with large amounts written by AI but I think there has to be a middle ground. In the end the site can do as it wants, I am not arguing that. For people like me and some foreign speakers who use translators it would still be nice to be heard. That is why as I said above, mark the stories and let the readers decide. Thanks.

You could create your own website and publish there.

 

[MillieDynamite]

I'm not saying that AI should be allowed, but quality is the reason for that rejection; the same quality of work by humans should also be rejected, and @Ym0Hyd's complaint was eyeballs he's competing with. He'll still be compeating with bad writing for readership.

The difference is the ease with which the poor writing can be produced. Even the worst-rated human-written story on the site took at least several hours to write, and almost certainly left the writer drained for the rest of the day. This limits the number of stories that they can submit.

On the other hand, someone could enter a prompt and submit their AI-generated story within a matter of minutes - and then repeat the process again and again.

That's a flood that will drown out the human-written content. It's already difficult to get views in several categories. Imagine if there were ten times as many stories for readers to go through every day. And sooner or later, if they see that 90% is soulless AI-generated slop, they'll give up and nobody at all will read our stories.

 

[MillieDynamite]

I could give you three big assed thumbs up for that comment, but I can only do one like, or its equivalent. But there's five small ones.

You could create your own website and publish there.

 

[StillStunned]

I'm not saying that AI should be allowed, but quality is the reason for that rejection; the same quality of work by humans should also be rejected, and @Ym0Hyd's complaint was eyeballs he's competing with. He'll still be compeating with bad writing for readership.

Yes, but much less of it.

Also, Lit's reasons for not publishing AI-generated content include ethical concerns and questions about copyright (as per the FAQ).

 

[MillieDynamite]

No, their main complaint, and I agree, is that it isn't human-made literature. I wasn't addressing Lit's reasoning; I was addressing @YmaOHyd's objection to it. I don't like all the bad work on Lit. But see no reasonable way of controlling that. Still, I'm not the judge of what's good or bad here. I'm interested in how he'd determine what stays and what goes.

Yes, but much less of it.

Also, Lit's reasons for not publishing AI-generated content include ethical concerns and questions about copyright (as per the FAQ).

 

[YmaOHyd]

But what is the difference between sharing eyeballs with bad writing as opposed to AI writing? The sit is filled with horribly passive-voiced dull stories. Do you think we should kill all stories that can't score a 4.0? But wait, a lot of 2.0 stories are actually good; they just offend one segment of the readership or another. HUM... What do you suggest to avoid your having to share eyeballs with those you consider inferior?

It's not about quality of work; it's about the effort required to produce it. If someone sits down at their keyboard and writes something that I consider unreadable, they still wrote it. They're as entitled to eyeballs as I am, and the attention marketplace decides, or should. I disagree with what the readers like and reward a lot of the time, but that's life, and I'm mostly happy to take my chances.

AI outputs are about prompt engineering. In the AI art world, which I'm more familiar with than the text-gen side of things, I can brute-force something that looks pretty good with the right combination of negative prompts and positive ones, plus setting the model to run 64 times and leaving the computer on overnight. Probably 64x 10k words is a little much, but maybe I set it to run eight times, then feed the outputs into Grammarly or some other tool and pick whichever one has the highest scores on their quality metrics -- if I'm being discerning. If not, maybe I run the model with (Ashley and Sam having sex at a lake, DD boobs, explicit, porn, fucking, romance), then against with (Scott and Tara having sex in a hotel, interracial, wife cheating, explicit, porn, fucking, rough) then again with (Will and Grace bondage, femdom pegging, DD boobs, explicit, porn, fucking), and post all three every single day. It takes about five minutes to generate an image using StableDiffusion, depending on the resolution and the number of injections and stuff; I doubt text generation would be especially long. You could shit out hundreds in a day if you wanted to.

As for how to detect AI outputs? Good question. That's one of the Holy Grails right there. A company that developed a foolproof method for doing this would have every single publisher and editor beating down their door to buy that product.

More broadly, Lit is Lit and that's not going to change. Would I prefer a system where there were twenty unbiased impartial editors who sifted through everything submitted each day and raised the quality bar significantly? Yeah, I would. But that won't ever happen, so why bother worrying about it?

 

[AwkwardMD]

2. Spell check and grammar check are not AI in the classic sense of the term: they are relatively primitive "expert systems", using sets of preprogrammed rules. They are not _adaptive_, which is a distinguishing feature of AI.

Just so this is clear, this was 100% true years ago, and is maybe 50% true today. It's is a bit like saying "A Polaroid is an entirely mechanical piece of machinery that produces the classic Polaroid photo."

Just because the output is the same thing you recognize from years back does not mean that, under the hood, there aren't newer and more complex systems at work.

 

[MillieDynamite]

The thing is, you'll never be able to absolutely say this writing, whatever it is, was or wasn't AI-generated. It will only get worse as I loses it's AIisms. Hopefully, that's a long way down the road. Knowing those isms helps keep your writing free of them. The other issue is that AIisms change over time.

It's not about quality of work; it's about the effort required to produce it. If someone sits down at their keyboard and writes something that I consider unreadable, they still wrote it. They're as entitled to eyeballs as I am, and the attention marketplace decides, or should. I disagree with what the readers like and reward a lot of the time, but that's life, and I'm mostly happy to take my chances.

AI outputs are about prompt engineering. In the AI art world, which I'm more familiar with than the text-gen side of things, I can brute-force something that looks pretty good with the right combination of negative prompts and positive ones, plus setting the model to run 64 times and leaving the computer on overnight. Probably 64x 10k words is a little much, but maybe I set it to run eight times, then feed the outputs into Grammarly or some other tool and pick whichever one has the highest scores on their quality metrics -- if I'm being discerning. If not, maybe I run the model with (Ashley and Sam having sex at a lake, DD boobs, explicit, porn, fucking, romance), then against with (Scott and Tara having sex in a hotel, interracial, wife cheating, explicit, porn, fucking, rough) then again with (Will and Grace bondage, femdom pegging, DD boobs, explicit, porn, fucking), and post all three every single day. It takes about five minutes to generate an image using StableDiffusion, depending on the resolution and the number of injections and stuff; I doubt text generation would be especially long. You could shit out hundreds in a day if you wanted to.

As for how to detect AI outputs? Good question. That's one of the Holy Grails right there. A company that developed a foolproof method for doing this would have every single publisher and editor beating down their door to buy that product.

More broadly, Lit is Lit and that's not going to change. Would I prefer a system where there were twenty unbiased impartial editors who sifted through everything submitted each day and raised the quality bar significantly? Yeah, I would. But that won't ever happen, so why bother worrying about it?

 

[YmaOHyd]

The thing is, you'll never be able to absolutely say this writing, whatever it is, was or wasn't AI-generated. It will only get worse as I loses it's AIisms. Hopefully, that's a long way down the road. Knowing those isms helps keep your writing free of them. The other issue is that AIisms change over time.

No, you won't. And that's why I say software that could do that with 100% accuracy is the Holy Grail, because it doesn't exist and probably can't ever exist; in order to detect patterns it needs to be exposed to those patterns. Detection is always behind creation.

But that doesn't mean Lit (or indie publishers or trad-publishers, because in the Kindle world there is a lot of obvious AI garbage getting published) shouldn't try. If the bar to be published here goes up, well, I'm pretty confident I'd clear it. If it goes down even further, I'd think about going somewhere else. The only thing Lit has to offer me is eyeballs.

 

[old_prof]

I'm not saying that AI should be allowed, but quality is the reason for that rejection; the same quality of work by humans should also be rejected, and @Ym0Hyd's complaint was eyeballs he's competing with. He'll still be compeating with bad writing for readership.

I fear ruling out poor writing shutting out new writers who will develop into writers who will contribute to the corpus that lit offers. My first story was certainly not good, but I think my stories as a whole are a contribution even if I’m far from a great writer

 

[MillieDynamite]

Again, I am not suggesting we exclude anyone. I was only inquiring about one person's statement. I'd prefer that those who use AI be honest about and post to an AI-specific set of categories. Even if Lit did that, they'd still have to send back human-written content that pings as possibly AI-influenced. Then, if the person had AI help, they just need to admit it. If they didn't, they'd need to refine it to pass muster, even if they aren't sure what to change. You're supposed to tell Amazon whether you used AI and which AI you used. I believe they are building their own detector based on what they learn from the AI-assisted writing. The only thing we've discussed (the other writers at Red Kitty's and our publisher) about AI is for covers (which is sometimes used for the covers) and for translations for other markets. So far, we aren't doing translations.

I fear ruling out poor writing shutting out new writers who will develop into writers who will contribute to the corpus that lit offers. My first story was certainly not good, but I think my stories as a whole are a contribution even if I’m far from a great writer

 

[Britva415]

I'm not saying that AI should be allowed, but quality is the reason for that rejection

It’s not.

 

[MillieDynamite]

Again, I'm not saying to do that.

I fear ruling out poor writing shutting out new writers who will develop into writers who will contribute to the corpus that lit offers. My first story was certainly not good, but I think my stories as a whole are a contribution even if I’m far from a great writer

 

[MillieDynamite]

It should've read if quality is the issue. I know it isn't.

It’s not.

 

[blowhyoooge]

I do not want to share space and compete for eyeballs with "Generate a story of between 10,000 and 15,000 words about Ashley and Sam having sex at a lake, DD boobs, explicit, porn, fucking, romance".

"I can't generate explicit or adult content. However, if you're looking for a romance story or a story with romantic themes, I'd be happy to help craft something engaging and meaningful. Let me know if you'd like to explore that!"

 

[YmaOHyd]

"I can't generate explicit or adult content. However, if you're looking for a romance story or a story with romantic themes, I'd be happy to help craft something engaging and meaningful. Let me know if you'd like to explore that!"

Meanwhile the X corporation is about to paywall Grok's ability to put children in bikinis. I wouldn't put your faith in AI companies preventing NSFW content generation; NSFW text is pretty easy to generate locally or by conditioning the models for some of the big providers. It has been for at least two years.

 

[blowhyoooge]

I'd prefer that those who use AI be honest about and post to an AI-specific set of categories. Even if Lit did that, they'd still have to send back human-written content that pings as possibly AI-influenced. Then, if the person had AI help, they just need to admit it. If they didn't, they'd need to refine it to pass muster, even if they aren't sure what to change.

i wish there was more subtlety here. i do not want to see ai-generated porn on literotica... if i want AI-generated porn i'll generate it myself.

i do think that ai can be helpful to a writer. my employer is eager, i think, to supplement or replace human editors with ai and so they want us to experiment. we have learned that ai is not a good editor. it can steamroll right over a writer's voice.

for editing, i have plugged in paragraphs and asked it to make them more concise. but the important part is re-editing it afterwards, because ai always has to make changes, even if they aren't for the better. it cannot say "The way you have it is good enough." unfortunately it is likely that any ai verbiage, even if vetted by a human,and even using words the human would have come up with, might trigger the ai detector, and once in lit ai hell its difficult to get out. ai can, i will begrudgingly admit, make some smart editing choices, but until the detectors get more subtle we cannot take advantage of that. a human can figure this out but i guess there are toomany submissions for people to review each one.

i have also used it for brainstorming. i've asked it for ideas, suggestions, scenarios. its answers aren't necessarily good but just like brainstorming with another person, something it says may trigger a better idea for me. i use it for research. ethical issues aside (displaying other people's content without permission or compensation) it can get better answers faster than regular search. and it's helpful for working out things i can't try myself, at least not at the moment, like is it possible for a man in this position to penetrate the asshole of a woman in this position?

so... it's an imperfect situation, but if i must make a choice i would rather read imperfect prose from imperfect people.

grok did do a nice job with that suggested lakeside fuck tory. i had it generate pictures of the two characters. ashley is a knockout and sam is handsome but it won't show me his cock.

 

[blowhyoooge]

Meanwhile the X corporation is about to paywall Grok's ability to put children in bikinis. I wouldn't put your faith in AI companies preventing NSFW content generation; NSFW text is pretty easy to generate locally or by conditioning the models for some of the big providers. It has been for at least two years.

grok was clearly programmed by hetero cis males who love boobies. much harder to get it to show dicks but i have managed once. unfortunately when i tried to make it into a movie, he grabbed his dick and it turned into a baguette.

 

[Andreas_Kreuz]

I believe the original 'dissected' statements are clear enough and accurately reflect our current status considering AI's rapid evolution. It says generated blocks of text are against the rules and talks about the voice of an author.

The other aspect that has been mentioned is that of effort. If you use AI to reduce your writing effort, it is probably wrong to do so.
The trouble is that the typical AI style is the typical AI style, because a large volume of prose on the internet has this style. So there are a large number of human writers who write in that style. If they hand in stories, they will be flagged as AI.