AI written subsection?

[pizer]

There are a number of both ongoing and completed stories that are heavily AI-written and not labelled so. It's quite clear if you've ever played around with an LLM.

Personally, I feel like these should be in their own subsection and labelled.

What do you all think?

(Posting this here since I think it counts as discussion - hope it's the right place.)

[ValentinesStuff]

I agree. I don't mind AI written, or AI aided stories being posted, mut they should be so labeled. Whether  new section is neededm that would be up to @DailyDi.

[Elfy]

Could you link an example or two of these unmarked AI stories so I can take a look at them?

[pizer]

4 hours ago, Elfy said:

Could you link an example or two of these unmarked AI stories so I can take a look at them?

Sure. Crossing Worlds 2, Eternal Child, and Tricked into the Diaper Dimension are a few. Recent updates to LLMs gave rise to them overusing this particular literary device, so it's especially noticeable. These LLM-isms tend to change over time, though some are consistent.

(I can detail some others if you want. They're really hard to unsee once you know; perhaps ignorance is bliss.)

From each story, in order:

Quote

Naomi made everything feel good.

And yet—

Something was wrong.

Something Ash hadn’t thought about yet.

Somewhere in the far, distant corners of her mind, the thought began to form, slow, creeping, nagging at the edges of her awareness.

A prickle of unease.

A whisper of something off, something missing, something she had overlooked.

And then—

The credits began to roll.

The happy, cheerful jingle filled the air, the bright, colorful letters dancing across the screen.

The episode was over.

It was over.

It was supposed to be over.

Ash’s body relaxed further, her muscles turning to jelly, her mind preparing to drift into content, blissful emptiness—

And then—

The next episode began playing.

Quote

She blinked. Once. Twice.

It was still there.

She turned her head, but no one else seemed to notice the literal glowing notification floating in front of her face.

Before she could even begin to process, a strange warmth spread through her body. A warmth that sank into her bones. Her heart stuttered.

Then—

Her clothes started glowing.

“Oh, no.”

The fabric shifted around her, morphing into something totally different. Her skirt shrank. Way, way shorter. Her blouse melted away, twisting into something pastel and frilly. Her shoes transformed into tiny, cartoonish sneakers.

The hallway around her felt wrong. Like it was too big. Like the world had just zoomed out.

Her stomach dropped.

Wait—did everything get bigger?

She held up her hands. Smaller. A bit pudgy.

Nope.

It wasn’t the hallway.

She was shrinking.

Panic clawed at her throat. Her legs wobbled. Her entire body felt alien.

And then—

Her stomach flipped. A wave of pure nausea slammed into her.

Before she could fight it—

She threw up. Everything. Her entire breakfast. Right there. In the middle of the hallway.

Her vision spun. Her legs gave out.

And then—

Darkness.

Quote

Slowly. Subtly.

Just enough to make her forget why she ever wanted to fight.

Paula’s eyes widened in terror. She still had time. She could hold on. She wouldn’t let this happen—

Then, she felt it.

Her stomach gurgled.

Her body relaxed.

And the second her diaper grew warm and heavy beneath her, the pleasure hit harder than ever before.

Paula gasped.

Her back arched.

A shudder of pure bliss ran through her, blanking out every thought for one horrifying, endless moment.

And when it passed—

She blinked.

What… what was she… thinking about again?

Mommy kissed her forehead.

“There’s my happy baby,” she cooed.

And Paula?

She giggled.

[ValentinesStuff]

@pizer That's kind of funny. They are improving. One of the first AI written stories I read, unlabeled and very early in the process, used virtually the same sentence in every paragraph.

[Babypants]

At a minimum, we should have a requirement that AI stories must be tagged, although this does not answer the question of who should be credited with authorship.  Personally, I agree with Pizer's suggestion that these works be separated out and housed in a silo of their own, if not banned altogether.     

[Babypants]

15 hours ago, Elfy said:

Could you link an example or two of these unmarked AI stories so I can take a look at them?

Take a look at the most recent chapter of Crossing Worlds 2.  What can one say?

[Elfy]

Thank you. Let me take a look at things and discuss with the other admins  

[DAQ]

22 hours ago, pizer said:

There are a number of both ongoing and completed stories that are heavily AI-written and not labelled so. It's quite clear if you've ever played around with an LLM.

Personally, I feel like these should be in their own subsection and labelled.

What do you all think?

(Posting this here since I think it counts as discussion - hope it's the right place.)

My vote would be subsection: No. Labeled: Yes - Either by title and/or tag

[TBCuri]

 

15 hours ago, pizer said:

Sure. Crossing Worlds 2, Eternal Child, and Tricked into the Diaper Dimension are a few. Recent updates to LLMs gave rise to them overusing this particular literary device, so it's especially noticeable. These LLM-isms tend to change over time, though some are consistent.

(I can detail some others if you want. They're really hard to unsee once you know; perhaps ignorance is bliss.)

From each story, in order:

It's sad tho, but thanks for the info 🙂

[Cute_Kitten]

Some readers do not like AI written stories, so either a rule requiring them to be labeled or a separate section for them would be very helpful IMO. 

Maybe this is a separate topic, but if someone uses AI to write a story for them, can they really say they're the author of that story? Just touching on what Baby pants mentioned in a post about who gets credit of authorship for AI stories. (But maybe that's a subject for a different post or discussion?)

[Pelusban]

I'm kind of picky with AB/DL literature, because I'm not native English. It's harder to read in a foreign language so I like well crafted, compact writing and twisted, unpredictable storyline. Maybe that's why I prefer comics, but even in comics the speech bubbles are often wordy. I checked these AI examples and I think these are exactly the opposite of my taste. AI's writing is lengthy and boring. But maybe I'm wrong, so are there any popular, preferred AI story examples? (I'm intentionally not writing "good", as it's so subjective).

[Babypants]

1 hour ago, Cute_Kitten said:

Maybe this is a separate topic, but if someone uses AI to write a story for them, can they really say they're the author of that story?

In academia, this has been a subject of intense debate over the last couple of years.  A line seems to be emerging that separates generative AI products as plagiarism from products like auto correct that allow students to express their own thoughts with fewer grammatical or stylistic mistakes.  But programs like Grammarly evolve over time, and once they straddle the line, their use raises difficult ethical issues for students and faculty alike.

Bringing this issue home to this site, we have a steady stream of individuals writing their first stories here, and given that it's a learning process, it seems to me that those who ask for constructive criticism should be applauded and rewarded for contributing their time and effort to our community.  Asking them to compete for the limited amount of time that we can spare them with AI generated stories IMO is grossly unfair.  At a minimum, we should require tags for AI work, and I would prefer to divide it off and give it a separate heading akin to the Poetry Corner .      

[BabySofia]

I think tags are probably the way to go so people can see before they view. I think it should be the precursor tag honestly. 

I don't see a problem with using Grammarly as an editor, but it now has AI generation features. On that note though, Microsoft now has it too... The line is going to quickly blur even more I'm afraid. 

I don't see an issue with using some AI as a tool for little things, just not actually writing the story. I personally use it for giving me a list of character names or traits with guidelines every now and then. I've used it to help me come up with titles for fictional devices at times too. Not actually story telling though... I get frustrated when I click on a story and it's obvious it's been written by AI. (To be honest I don't even know how they get it past some of the safety protocols in place - I've gotten blocked on a ton of pretty innocuous things in the past!) 

[Babypants]

20 minutes ago, BabySofia said:

I don't see a problem with using Grammarly as an editor

I find it ironic that these tools are so poorly designed.  There is an auto-correct feature on Open Office that red lines everything it sees as a spelling error.  There have been occasions when I have gone on line to run its objection against other tools.  And guess what ... to date, Open Office auto-correct has proven wrong 100% of the time.  I am not exaggerating here.

Another thing that astonishes me is the number of outright factual errors that these programs commit.  Case in point.  A couple of days ago, I asked google how long Mother-in-Law was on the UK charts in 1961, and how high it climbed.  It answered 11 weeks, reaching #1.  No.  The correct answer is 7 weeks, topping out at #29.  I found this in (a) a book in my library, and (b) in a separate but equally massive on line guide.  Google must have been eavesdropping on my research, because when I looked it up again some hours later, it had corrected its mistake.

The old saw used to go "trust but verify."  How quaint the concept seems today.  

[BabySofia]

2 minutes ago, Babypants said:

everything it sees as a spelling error. 

It's always happened that way. One thing I do like about Grammarly is the ability to get it to easily add words to its dictionary. Microsoft yesterday was marking 'entendres' as misspelled (as in double entendres), and I had to go look it up. My spelling skills are still pretty good, despite autocorrect hurting them some. I do triple check things sometimes through other sources because of it. 

Yes, doing research and depending on it being accurate through AI is asking for trouble. Judges are starting to have issues with lawyers citing fake cases thanks to it. I can only imagine what the world of an English teacher must be like...

[pizer]

1 hour ago, BabySofia said:

I think tags are probably the way to go so people can see before they view. I think it should be the precursor tag honestly. 

I don't see a problem with using Grammarly as an editor, but it now has AI generation features. On that note though, Microsoft now has it too... The line is going to quickly blur even more I'm afraid. 

I don't see an issue with using some AI as a tool for little things, just not actually writing the story. I personally use it for giving me a list of character names or traits with guidelines every now and then. I've used it to help me come up with titles for fictional devices at times too. Not actually story telling though... I get frustrated when I click on a story and it's obvious it's been written by AI. (To be honest I don't even know how they get it past some of the safety protocols in place - I've gotten blocked on a ton of pretty innocuous things in the past!) 

That's where I draw the line as well. Asking an AI to suggest a name for a character is effectively the same thing as Googling "list of [characteristic] names," but, as soon as it starts doing the writing part, the authors' style (if they're even writing any of it) is lost. That has always been one of the most interesting and exciting aspects for me as a reader. Francine Prose has a fantastic book called Reading Like a Writer that gets into this step-by-step with hundreds of examples; I'm not a writer but it's still one of my favorite books.

As for safety protocols, they're not using the major AIs. They're either creating their own (which doesn't take as much tech savvy as you'd think--plenty of open source models and training data) or using one of the available unrestricted ones typically made for nsfw roleplay.

6 hours ago, Cute_Kitten said:

Some readers do not like AI written stories, so either a rule requiring them to be labeled or a separate section for them would be very helpful IMO. 

Maybe this is a separate topic, but if someone uses AI to write a story for them, can they really say they're the author of that story? Just touching on what Baby pants mentioned in a post about who gets credit of authorship for AI stories. (But maybe that's a subject for a different post or discussion?)

In terms of ownership, I view it as akin to commissioning someone to write an idea. The idea may be theirs but bringing it to life--the important, difficult part-- has nothing to do with them. Then add in the harmful effects like reducing demand for commissions from actual writers and what Babypants said about competition for attention/interaction and it's not a pretty picture. Also, removing the barriers to entry for any idea to be made into a story isn't great.

[Olympiczero]

17 hours ago, pizer said:

Asking an AI to suggest a name for a character is effectively the same thing as Googling "list of [characteristic] names," but, as soon as it starts doing the writing part, the authors' style (if they're even writing any of it) is lost.

I use AI for a number of things when writing: As @BabySofia mentioned, names are one. I am a very visual learner too, so I have used it in the past to generate floorplans of homes/schools/other locations my characters frequent, town maps, generating images for specific bedrooms I describe and create. SketchUp is a great tool for floorplans, not sure if anyone else uses that. I don't rely on AI for the description of any building, but just to help me visualize what I am creating when I write, show where doors/hallways are located, etc... This also helps me stay consistent (though I fail regularly) with where dressers, beds, rooms, shops and other locations, are throughout the story. I usually print the layouts out once done. That way I have them handy with me when I write, and don't need to consult more than one screen.

The most helpful thing AI does for me though is helps me keep track of my characters relationships. I am not great with remembering names, especially if I name a character a few chapters earlier who does not make a reappearance for a bit. Rather than hunt the name down by trying to remember where I wrote about them, I use AI to keep a list of characters, usually by Family Group, and print those out as well. If a character (as mine tend to) changes in a story, I can note next to their name in multiple Family Groups what chapters/ages they are while in that group.

I know I could do all of this by hand, but using AI helps for a few reasons but one more than others 1) my handwriting and artistry are terrible. When I was writing my first story, I started with hand drawn floorplans and family lists, and by Chapter 15 or so, couldn't figure out half of what I had written down. I also don't have to fear losing all my papers which happens enough with my actual job, I don't need to exacerbate the problem. Having everything in one place where I can pull it up quickly is helpful. I

I thought about this a long time before responding. I am generally impartial to the idea of requiring AI tags FWIW. Among those suggested, I like the idea of tagging stories best. But I see a potential issue with it; who is the arbiter of what is and is not AI written? If an author doesn't tag their story as AI, can a mod compel them to do so if the mod feels that the story is AI? Does it get put to a vote among mods? I'd hate the forum to devolve into infighting about what it and isn't AI, and the potential to chill writers from wanting to participate in contributing to the story forum out of fear that they. Writing here has been so therapeutic for me at times, I'd hate to feel as though I was denying someone that benefit. 

Personally, my solution is simply to not read stories i don't enjoy. Just thought I would offer my two cents, but I defer to more seasoned members here as I am sure there are consequences I've not considered in coming to my conclusion. 

[Cute_Kitten]

I think there is a difference between using AI as a support tool to help an author with a story- things already mentioned like spell and grammar check, keeping track of character names, settings and places, and even keeping track of plot lines etc in longer works. That's different than using AI to do the actual writing part- like "Hey AI, write me a story set in Ancient Egypt about Pharaoh Tutankhamun the boy king wetting his bed and his nanny puts him back in cloth diapers at night." And letting AI do all the work and heavy lifting. Then there are the "hybrid" attempts I've seen where a person will take an AI written story from a prompt they fed the AI, tweak or "edit" the story and claim they 100% wrote the story.  I've seen people who will feed AI prompts scene by scene, cobble those AI written scenes together and call that a story they wrote and call themselves an author. 🤷‍♀️. Sure anyone can do that and is free to do that. Heck what I've heard through the digital grapevine is places like Fiver are having problems with "authors" and "artists" selling "story commissions" and "art commissions" that turn out to actually be AI generated and leaving the customers unhappy. 🤷‍♀️ 

If someone wants to generate themselves an AI written story they can. I think the problem comes in when they want to publicly post that story or put it up for sale on a site like Amazon and pass themselves off as an author without letting people such s readers know it is AI. It's the same as someone using AI to generate art and calling themselves an artist. 

[randomanon]

On 3/12/2025 at 4:30 PM, Cute_Kitten said:

Then there are the "hybrid" attempts I've seen where a person will take an AI written story from a prompt they fed the AI, tweak or "edit" the story and claim they 100% wrote the story.  I've seen people who will feed AI prompts scene by scene, cobble those AI written scenes together and call that a story they wrote and call themselves an author. 🤷‍♀️.

I'm curious: would you count my latest story as feeling like AI? I used a base LLM model, not a chatbot. What I do is I write part of the story, then see how the AI would continue it. After that, I do one of the following:

1) take some with minor editing, write a bit more myself to steer the next part, repeat. Rarely do I take more than a sentence or two at a time.

2) try sampling again with a different random seed

3) look at the output probability distributions and "sample" manually to get something I like (kinda like writing but with a list of a likely set of words that'd follow what's written so far)

4) Reword or rework the previous writing and try again

It's definitely not the same as normal writing, but it's also not the LLM chatbot prompting that's more like commissioning a work than actually doing it yourself (and with all the Chat-GPTisms like above that come with that)

My first story that I started doing this with I mentioned the AI use, but as I got better at using the workflow (and the models improved) I didn't bother mentioning it as the output didn't "feel" AI the way the first one did - and the various AI editing tools got integrated into mainstream products, which I felt was more similar to what I was doing.

[BabySofia]

4 hours ago, randomanon said:

I'm curious: would you count my latest story as feeling like AI? I used a base LLM model, not a chatbot. What I do is I write part of the story, then see how the AI would continue it. After that, I do one of the following:

4) Reword or rework the previous writing and try again

It's definitely not the same as normal writing, but it's also not the LLM chatbot prompting that's more like commissioning a work than actually doing it yourself (and with all the Chat-GPTisms like above that come with that)

I agree that's better than pure AI writing, but I think that's what's required to make any of that make sense from the little bit I've played with it. (I tested using one to 'finish' a story one time out of curiosity that an author had abandoned) Personally, I would still think that worthy of having an AI tag, and academically, I think there should at least be a citation of the use of the specific language model/generator. 

That's just my .02 on that.

[Babypants]

5 hours ago, randomanon said:

It's definitely not the same as normal writing,

 

1 hour ago, BabySofia said:

Personally, I would still think that worthy of having an AI tag, and academically, I think there should at least be a citation of the use of the specific language model/generator. 

Baby Sofia is quite right.  In an academic setting using the work of others without attribution, regardless of source, is almost universally considered to be plagiarism.  In my classroom, this would have resulted in a failing grade, and a hearing before the academic disciplinary committee.  On this site, given what you have written above,  I would hope that you would tag your work as AI.

[oznl]

1 hour ago, BabySofia said:

Personally, I would still think that worthy of having an AI tag, and academically, I think there should at least be a citation of the use of the specific language model/generator.

This is an interesting thread and I absolutely agree.  At a very minimum, I believe AI generated content (including partial-generation) needs to be labelled as such using a consistent and unambiguous tag.

This is not only for the "benefit" of the reader (who is already being short-changed in the literary sense just by being fed LLM content) but also to deal with another emerging issue:  the problem of Large Language Models (LLM) ingesting their own output content as training set data with the associated accuracy issues that causes.

I do say that I find the emergence of this issue to be sad.  Although on society's scrapheap, my brain still works and fascinated by ChatGPT when it emerged, I studied how LLMs work (I can recommend Stephen Wolfram's writings).

Here is my REAL issue with LLM-generated creative writing.

Without boring you on the mechanics, LLM is a shallow computational model applied to a massive data-set to build content word-by-word based on statistical likelihoods of certain words occurring adjacently.  It isn't intelligence.  It is an emulation of intelligence reflected from everything it read on the internet.  As such, it is biased towards the "average" of what it has seen.

The average human IQ is 100.  Just saying...

At the end of the day, although it is remarkable and doubtless in some ways extremely useful, there is no ghost in the LLM machine.  For "creative" writing, any beauty or insight we see in its output is a product of that output being subjectively novel to us or our own synthesis being applied.

I suppose if one is sub-standard at prose and uninterested in originality, an LLM can drag them up into average fairly easily.  I'm sure Barbara Cartland could have written quadruple the dross she did with such a tool but I don't see how that makes the world a richer place.

I’m not sure you can call yourself an “author” if you’re using an LLM to generate your material.  You have become a word technician operating a machine.  An author is an artist who paints in words.

 

[randomanon]

1 hour ago, Babypants said:

Baby Sofia is quite right.  In an academic setting using the work of others without attribution, regardless of source, is almost universally considered to be plagiarism.  In my classroom, this would have resulted in a failing grade, and a hearing before the academic disciplinary committee.  On this site, given what you have written above,  I would hope that you would tag your work as AI.

But it's not the work of other people. An LLM is not a person. Unless you're claiming that I'm indirectly plagiarizing the training data, but that doesn't really work either unless you would count someone vaguely remembering reading something a decade ago and letting it influence them as "plagiarizing" it. LLMs are a fancy auto-complete, better than a Markov chain (the probability of each possible next word given the previous N words for small N) but certainly not anything approaching a person.

It's not the same as normal writing, but nor is writing on a computer the same as writing on a typewriter. Is tapping the next word suggestion on my phone's keyboard plagiarism?

(Also, in an academic setting, how would you know if the person didn't admit it? It's definitely a real problem, especially if the writing is meant to test understanding and the person is using the LLM to get around that. Punishing the people who are willing to explain what they did seems like a very bad incentive structure...)

1 hour ago, oznl said:

Without boring you on the mechanics, LLM is a shallow computational model applied to a massive data-set to build content word-by-word based on statistical likelihoods of certain words occurring adjacently.  It isn't intelligence.  It is an emulation of intelligence reflected from everything it read on the internet.  As such, it is biased towards the "average" of what it has seen.

Mostly agreed on this, but I disagree with your overall point. An LLM is a fancy auto-complete, a compression algorithm on all of human language. Turns out that compression and the pattern matching involved are surprisingly close to intelligence.

Any artistic endeavor can be thought of as a search through the state-space of possible works. Each letter is 8 bits of data (assuming a basic ASCII encoding). But the Library of Babel is useless, as the vast vast majority of text is incomprehensible gibberish. Shuffled English text, so just going by pure word-by-word frequency, is something like 10 bits per word. Adding some basic grammar information brings it down to 6ish bits. An LLM gets that down to something like 4 bits of entropy per word on average.

Those 4 bits are what really matters, the space of "somewhat reasonable English text". Sometimes I can randomly sample from that probability distribution and get something I like - but even then I'm applying the "do I like that" 1 bit of filtering per generation. Quite often I choose words that the LLM finds extremely unlikely, in the 0.1% probability but still in the top 100 possible next words. Occasionally what I want is low enough that it's not in that list at all.

If you just prompt it at the start and let it rip, that's what, a few hundred bits of information in the prompt? Not much. Strong agree that AI written text like that is basically not written at all by the prompter. But I'm putting in something like 2 bits per word. Less than a pure human text at 4 bits per word, but still a lot - and the remaining 2 bits is just me going "it doesn't really matter" and letting randomness take hold. Does Jackson Pollock's use of randomness from paint splatters make him not an artist?

*For correctness's sake I should be using "token", not "word" here, but the two are close enough for general purposes.

1 hour ago, oznl said:

I’m not sure you can call yourself an “author” if you’re using an LLM to generate your material.  You have become a word technician operating a machine.  An author is an artist who paints in words.

Is the person making 3D models that then get rendered into 3D animation a technician as well? If there's some animated water, are the people who worked on that part of the movie not artists because they didn't hand-model every wave and instead ran a water simulation?

[oznl]

25 minutes ago, randomanon said:

Turns out that compression and the pattern matching involved are surprisingly close to intelligence.

Probably outside the scope of this thread but whilst I will give you that it produces something that looks like the product of intelligence (indirectly, it is), it has no agency and is not of itself therefore, intelligent.

27 minutes ago, randomanon said:

*For correctness's sake I should be using "token", not "word" here, but the two are close enough for general purposes.

Yep, I used "word" instead of "token" to avoid alienating the 5% of the readers of this thread that I haven't already alienated 🤣  I didn't want to drag in another concept.

28 minutes ago, randomanon said:

Is the person making 3D models that then get rendered into 3D animation a technician as well? If there's some animated water, are the people who worked on that part of the movie not artists because they didn't hand-model every wave and instead ran a water simulation?

Interesting point.  I guess I was aiming my commentary at those who feed in some initial data and press "play", waiting for fap-worthy material to spew forth 🤣

I accept that using an AI tool to in-fill the dumb stuff is a productivity booster and doesn't endanger the artistic merit of the venture:  "Shrek" didn't happen by somebody painting 32 frames for every second.

I'm not sure what the right answer is for "AI-augmented" content as opposed to "AI-generated".  The purist in me still wants to see this tagged somehow.