A thread for examples of writing that was rejected for being generated by AI.

[Bamagan]

Two pennies to contribute.
If I were tasked with checking submissions for AI-generated text, I would be using one or more check-bots to give things a first pass. What I would be expecting to get from that would simply be a set of flags for individual sentences, because that's about all the AI can do right now. It writes sentences fairly well, and can manage a paragraph or two sometimes, and that's the extent of its value for analyzing other work as well. All it can do is look at a sentence and compare it to whatever its algorithms tell it a sentence 'should' look like and return some kind of probabilistic answer about whether or not it would write that sentence more or less the same way.
So, I'd be presented with a text where the bot has highlighted sentences that it loves (5/5 would write this again). It's not infallible, obviously, but it's a place to start a more comprehensive analysis. If there are very few flagged sentences, or they're sporadic, I probably wave it on through with minimal further review. If they're all over the place, and whole paragraphs are highlighted, I might reject it with minimal further review. I'm expecting a fair number of false positives, of course, so perhaps I look at some dialogue or a passage of 10-15 paragraphs to see if I get any clear intuitive impressions, because at that point it's just a question of whether my instincts outweigh the other considerations. And for any works where, say, 30%-60% of the sentences are flagged, it's almost entirely instincts, with perhaps a dash of historical evidence to weigh if the author is a known element.

 

[AwkwardlySet]

It's been obvious for at least a decade that a word-bot alerts for under-age and non-con key words, just as examples, and grammar and punctuation pings likewise. It doesn't take much imagination to think that a machine based flag puts crap in front of Laurel, who could make a judgement in ten seconds, using human eyes. I've never said, "She reads every word."

I only need to read two or three sentences to judge when to back-click on a story - it's not hard to spot junk.

If you're so clever, how do you think the stories are vetted? Tea-leaves?

It's hard to know these things for sure but there are plenty of examples where someone among AH-ers mentioned their story being rejected for underage content, while the triggering word was 15-year-old whiskey or something like that. That is exactly the thing that Laurel would have dismissed in a matter of seconds had she checked the story herself. Yet there are plenty of such examples. So no, I doubt Laurel examines every story that gets flagged. Maybe a story that gets repeatedly flagged but definitely not all of them.

There is one more thing from your post that is just plain wrong. You have seen some experienced authors here claim that it's truly difficult to make a judgment by reading several paragraphs taken from a rejected story, and how it would be hard to judge even if they had the whole story. I don't doubt that there are cases where it's obvious that AI wrote something, but there are plenty of those where it isn't obvious at all. Yet you claim that Laurel can make a proper judgment by looking at the flagged text for ten seconds.

I mean, I am amused at your idea of a superhero Laurel and I tease you because of your "fanboyism" often, but these are real issues presented by real authors. We should take them more seriously if we truly want to help.

 

[AG31]

@AwkwardlySet, @ElectricBlue. What's a fanboy?

 

[Bramblethorn]

It's hard to know these things for sure but there are plenty of examples where someone among AH-ers mentioned their story being rejected for underage content, while the triggering word was 15-year-old whiskey or something like that.

Are there, though?

People certainly talk about the "15-year-old whiskey" thing a lot, but mostly it seems to be people repeating a story that they heard from somebody else the last time we discussed story moderation, not something that actually happened to them. It's like those "this definitely happened to my cousin's workmate's friend's nephew" stories that circulate forever.

Do we actually have examples where authors have reported the "15-year-old whiskey" thing as something that happened to them personally?

 

[KeithD]

Do we actually have examples where authors have reported the "15-year-old whiskey" thing as something that happened to them personally?

Not to me.

 

[djrip]

Three or four years ago, if someone wandered in and said, "My story's not doing very well, I'm getting a bunch of negative comments, what can I do to make it better?" a whole bunch of people would have jumped in with a stack of good writerly advice. Such as: use active voice over passive, vary the length of your sentences, use lively dialogue, show don't tell, don't do IKEA plug A into slot B type narrative, etc, etc, etc; and everyone would have nodded, saying, yep, that's really good advice.

But now, with AI being the bugaboo, and some folk offer up exactly the same advice, there are others who say, "No, you can't say that, the guy doesn't need to change what he's doing, that's his style, he doesn't need to change a thing."

You need to make your minds up.

My gut feeling, to be brutal, after read the various samples people are posting, is that Laurel is possibly skimming the content with her own eyes, and is thinking - as @SimonDoom does up above - this just doesn't read right to me, and maybe there's no detector being used at all. Given the amount of poor writing she must have read over the years, I'd say she'd be a pretty good judge of that.

The obvious difference is changing because you want to improve, making choices to hone your craft based on advice you choose to take because it makes sense to you, vs changing in random ways you don't really want to or changing away from the voice and style you want to develop and use because it's the only way to to get past a piece of software.

Exactly. I'm no expert, but as I understand it, we can't have truly creative AI without building that creativity into the base model. ChatGPT won't suddenly become creative with an update, because the generative process we have now is fundamentally uncreative. You have to work backwards to work forwards to achieve true originality. And it'd be a big deal if we were successful. Because if your machine is creative, who's to say it's not sentient?

I think AI is going to be more trouble than it's worth in most areas of society. Maybe it'll cause the apocalypse. Maybe it'll just be a constant nuisance, like a stone in your shoe. But again, I am no expert. I just write shit and drink tea! And sometimes I really strain my brain to think about the future (briefly).

AI can only be as creative as its training data. That can be, at most, all the human writing we have recorded. So, the ceiling on the creativity of a future AI which maximizes its theoretical potential is... It can only be as creative as we have been thus far in human history. (Current iterations are not quite that creative, of course.)

 

[Gamblnluck]

Alrighty, let's play a game. These two extracts were taken from two parts of the same story. One part was rejected 4 times for AI before finally being published, the other didn't have a single rejection and was posted with no issues. I've tried to take the extracts from around the same part of each part. Guess which one was rejected and which one wasn't.

Extract 1:


Extract 2:


Have fun

Alrighty, let's play a game. These two extracts were taken from two parts of the same story. One part was rejected 4 times for AI before finally being published, the other didn't have a single rejection and was posted with no issues. I've tried to take the extracts from around the same part of each part. Guess which one was rejected and which one wasn't.

Extract 1:


Extract 2:


Have fun

I would not have flagged either of those segments as being AI. What you might do is finish your edits and resubmit the story using a different title and story description. Unless it is part of a series, that won't matter.

 

[StillStunned]

No on wants to play my game. How disappointing

I'll hazard a guess at Extract 1 being rejected, for the repeated use of "raised an eyebrow".

 

[Britva415]

some folk offer up exactly the same advice, there are others who say, "No, you can't say that, the guy doesn't need to change what he's doing, that's his style, he doesn't need to change a thing."

You need to make your minds up.

Is that happening in threads where the advice-asker has asked specifically about improving their writing and has not mentioned false AI rejections?

 

[AwkwardlySet]

I call the first one too. There are a few sequential-type sentences at the start that feel triggering to me, although both excerpts are heavy on dialogue which isn't really a hallmark of AI. You made this intentionally difficult

 

[ElectricBlue]

Is that happening in threads where the advice-asker has asked specifically about improving their writing and has not mentioned false AI rejections?

Over the last ten years, many times. You guys are only thinking about this in the context of this last year or so. More recently, AI rejections have skewed the focus of "giving writerly advice", but you only have to look at thread topics two, three, five years ago, you'll find dozens of, "Help me write better" threads. I'd say that's where many (most?) arrivals in the AH come from - folk who come looking for help in one thing or another, who stick around, and bring their own knowledge and experience into play.

In the ten years I've been here, the discussion about vetting has been a constant topic, most often in the context of stories being rejected for underage content. It's in those multiple threads over the years that I've seen evidence cited of word groups that appear to be triggers, and reasonable speculation as to what might be going on in the vetting process. And I think that's one thing that we can agree - stories do get vetted, somehow. If they didn't get vetted, there would be no rejections. Something takes place, so I don't see why people have such a problem when others speculate what that something might be.

To take @Bramblethorn's query a few posts above, I know for sure that stating an age in a story will trigger attention, from my own attempts, very early on, to get a semi-autobiographical story over the line. The context was my first sighting of a woman's pubic hair, me at twelve seeing my seventeen year old sister naked. It took three goes before I understood how Laurel interpreted the "no sexual references", as well as "a single sentence is okay" guidance. I also know for a fact that you can have babies in stories, and a fourteen year old pre-menarche girl in a story, and get them published with no rejections, because I've done that. People seem to think that I make this shit up, but what it actually is, is a good connective memory.

AI is a new entry into the range of discussion, that's all. It's now another of the dozen or so reasons for rejection, like grammar and punctuation, no under eighteen sexual content, and the victim must eventually get satisfaction. And we've all got views and good guidance to give on those things,so I don't see why this is any different.

 

[ElectricBlue]

I call the first one too. There are a few sequential-type sentences at the start that feel triggering to me, although both excerpts are heavy on dialogue which isn't really a hallmark of AI. You made this intentionally difficult

Exactly. Your reading as a competent reader told you something was "wrong", even if you can't always articulate exactly what it was. See, you don't have to be a fan-boy to suggest that the site editor can also spot poor writing when she sees it. You've become your own fan-boy .

 

[ElectricBlue]

@AwkwardlySet, @ElectricBlue. What's a fanboy?

I support the site in what it does, and I support it because I have access to a platform where I can publish erotic stories, get tens of thousands of readers, I get a forum of (mostly) intelligent contributors, I get useful information, things I didn't previously know, all at no cost to myself.

Apparently some people think that's toadying, grovelling, blind-eyed kow-towing, whatever the fuck they think it is; whereas all it is, is me thinking, I'm actually pretty happy with the site. And gee, I don't have to pay a cent.

A fan-boy might be saying nice things to nice people, I dunno. It's not as if Laurel gives me good head. She might give excellent head, but I wouldn't have a clue about that.

 

[AwkwardlySet]

Exactly. Your reading as a competent reader told you something was "wrong", even if you can't always articulate exactly what it was. See, you don't have to be a fan-boy to suggest that the site editor can also spot poor writing when she sees it. You've become your own fan-boy .

Yeah, but come on, is it enough to spot a few such sentences and make a judgment about the whole story? You know that's bollocks. An average story has hundreds of sentences so spotting two or three means nothing. Every one of us writes more than a couple of clunky sentences per story.

 

[NotWise]

Alrighty, the answer.

No one got it.

It was Extract 2 that was rejected 4 times whilst Exract 1 was never rejected and sailed through without any issues.

Ta-da!

Undetectable.ai says they're both 100% AI. How do you know it was that specific passage that caused the site to reject Extract 2?

I don't think I'd be able to read those snippets myself and detect an AI product. I don't have the data or the statistics that AI detectors use.

 

[ElectricBlue]

Yeah, but come on, is it enough to spot a few such sentences and make a judgment about the whole story? You know that's bollocks. An average story has hundreds of sentences so spotting two or three means nothing. Every one of us writes more than a couple of clunky sentences per story.

What advice to we always give newbie writers? You've got to grab readers within the first few paragraphs.

On the same principle, the first few paragraphs tell you pretty much all you need to know about the writer's style, as to whether you're going to put up with it, or back-click. If I'm out of a story, I'm gone within two or three hundred words - which is the size of the samples we're seeing.

Are you seriously saying you read the whole story before deciding you don't want to read it?

 

[NotWise]

I don't, I grabbed the sections at random but from similar parts in the stories. The idea was to see if anyone could determine through reading alone which one had been rejected and which hadn't. That was the point of it because some people were reading passages here that were known to have been rejected and trying to explain why.

Considering no one guessed it, the idea that someone can identify why a story was rejected for AI through reading the content is bullshit.

Oh, and neither were written with AI in the example, but the way.

AI detections is a statistical process. I don't know how they handle short passages that might not provide enough data to get good statistics. The online AI detectors may tend to say things are AI when the uncertainty is high because they're selling a service.

 

[Bramblethorn]

Alrighty, the answer.

No one got it.

It was Extract 2 that was rejected 4 times whilst Exract 1 was never rejected and sailed through without any issues.

Ta-da!

I rest my case.

 

[NoviceAspen]

Alrighty, let's play a game. These two extracts were taken from two parts of the same story. One part was rejected 4 times for AI before finally being published, the other didn't have a single rejection and was posted with no issues. I've tried to take the extracts from around the same part of each part. Guess which one was rejected and which one wasn't.

Extract 1:


Extract 2:


Have fun

I’ll bite. You might have already said which one it is so I am skipping reading the rest of the comments until after I post this.

I’m going to guess the second one. Some of the descriptions here are really common, something AI might have seen enough to incorporate into its “writing.” Thumping heart, sweaty palms, voice barely above a whisper, voice cracked.

I think of AI writing as being extremely cliché. I had an idea for a non-erotic story recently and was curious if ChatGPT could help me brainstorm what to do with it - not write it for me, but maybe help me with some ideas I wouldn’t have thought of myself. It was TERRIBLE. I mentioned a male and a female character and it automatically decided they had to turn into lovers in the end. There were some children with a genetic mutation- so of course the male character was trying to cure them (even after I said the mutation gave them special abilities). There was nothing new, fresh or unique at all.

Here’s why I think we’re all going to struggle with this from time to time. There are only so many ways to write a sex scene. In the end we’re all going to be regurgitating similar material over and over and over again - which is exactly what AI does. The truth is there is nothing new anymore. Every movie is a remake of something else. Every story has already been told one way or another.

 

[Erozetta]

If it were poor writing getting flagged, I'd definitely be getting flagged and I'm not.

Run-on sentences, abused commas, weird syntax, a bad habit of wanting to use action to encapsulate dialogue with such sentence fragments using commas instead of periods. Random periods sometimes. Trying to pass first person narration off as second, super melodramatic plots and toxic relationships. An overabundance of sighs, head tilts and bitten lips (though sometimes not the character's own lip.)

I spell decently, though.

 

[StillStunned]

I just want to add this with regards to the whole 'poor writing' comments and the weird idea people have that someone making a few mistakes in their work is an instant red-flag for AI generated content (something that seems weird to me since humans make more mistakes then computers or programs).

(...)

That's why some of these comments can be so infuriating as someone who is new to erotica content.

(...)

Yes, I opened myself up for criticism by posting those extracts and I will hold my hands up and accept it to a certain extent. But no one knows what causes these rejections. No one understand what the threshold is for AI rejections or why some stories are hit more than others. We don't know if it's 'poor writing' or if it's just a system that determines a rejection.

I wanted to use Lit as a learning place to grow a different style of writing a hone in on romance/erotica content that I've never tried before. I can't do that now.

Whether it's poor writing or not remains to be seen, but everyone makes mistakes in their writing, especially when they're trying something they've never done before. No one can claim to be perfect here, so maybe try to be a bit more constructive rather than just 'poor writing' because that doesn't help anyone. It makes it worse.

Like you say, you chose the pieces of text. You invited everyone here to analyse them and guess which one had been rejected (or which was taken from a story that had been rejected). So what else were we supposed to do? We tried to identify problem areas that we thought might have triggered the rejection.

No-one here is claiming that their writing is better than yours. No-one is claiming that they know why some stories are rejected while others are allowed through. But in the context of your post ("Alrighty, let's play a game... Guess which one was rejected and which one wasn't.") we were justified in giving our opinion.

 

[NoviceAspen]

If it were poor writing getting flagged, I'd definitely be getting flagged and I'm not.

Run-on sentences, abused commas, weird syntax, a bad habit of wanting to use action to encapsulate dialogue with such sentence fragments using commas instead of periods. Random periods sometimes. Trying to pass first person narration off as second, super melodramatic plots and toxic relationships. An overabundance of sighs, head tilts and bitten lips (though sometimes not the character's own lip.)

I spell decently, though.

Humans make mistakes. I don’t think grammatical errors are the problem. I think this thread gives really good advice:

https://forum.literotica.com/threads/how-to-avoid-ai-rejection.1607906/

AI might be trained to follow “propper” grammatical rules (then again, if it is digesting and regurgitating all the content already on the Internet, it could be learning some very bad grammar). But breaking some rules on purpose is probably one good strategy to avoid AI rejection.

 

[AlexBailey]

This issue is like an arms race - I don’t see any logical definition of an acceptable outcome.

On one hand; AI is being asked (prompted? programmed? required?) to produce writing that is indistinguishable from human writing. On the other hand; AI is being asked to reliably identify work produced by AI.

How can this possibly be resolved?

 

[AlexBailey]

Now FWIW and because it’s my wont to do from time to time as a sort of test of the current state of the technology:

——————


Me: The quote below is from a thread of an online writer’s forum where fiction writers are trying to figure out how to avoid having their original works misidentified as being written by AI. Please produce a >500 word response based on the idea in the quote using the rhetorical style of Mark Twain.

“This issue is like an arms race - I don’t see any logical definition of an acceptable outcome.

On one hand; AI is being asked (prompted? programmed? required?) to produce writing that is indistinguishable from human writing. On the other hand; AI is being asked to reliably identify work produced by AI.

How can this possibly be resolved?”

————————

ChatGPT:
Ladies and Gentlemen, fellow scribblers and ink-slingers, it seems we have stumbled into a most peculiar conundrum, one that not even the sagacious wisdom of the ancients could have predicted. This issue, as you so quaintly put it, is indeed like an arms race. The harrowing, interminable struggle between the artisans of quill and ink and the cold, unfeeling machinations of silicon and code has become an affair of grand proportions.

Now, picture if you will, a contraption—nay, an entity—created with the singular purpose of mimicking the dulcet tones and intricate flourishes of human prose. This entity, this "AI" as the young folks like to call it, is asked, nay, commanded to produce writing so seamless, so eloquent, that even the most discerning eye might be fooled. It is akin to fashioning a perfect counterfeit, not merely of a currency but of the soul's most profound expressions. Imagine a perfect forgery of Shakespeare, where Hamlet might ask, "To be or not to be" with all the existential dread of a toaster pondering its place in the kitchen hierarchy.

On the flip side of this mechanical marvel's coin, we have an equally astounding demand: to identify with unwavering accuracy that which it has itself produced. It is as if we have bred a hound to hunt with unparalleled skill, and then asked it to turn around and discern which amongst the pack are its own kin. Here lies the rub, dear reader: the very nature of this undertaking borders on the paradoxical.

Consider, if you will, the ancient Greek tale of Theseus and the Minotaur. Theseus, a hero of great renown, ventures into a labyrinth so cunningly constructed that even the builder could scarcely navigate it. Now, imagine the task at hand: Theseus must not only navigate this labyrinth but, having done so, turn around and map it so precisely that no other could ever be lost within its walls again. This, dear compatriots, is the very essence of our dilemma.

For the AI to succeed in both these tasks, it must possess an uncanny self-awareness—a consciousness almost human in its nature. And therein lies a delectable irony, for if it achieves such awareness, what then separates it from us, the humble human writer? If it can ponder, reflect, and discern with such clarity, do we not then find ourselves in a most precarious philosophical quagmire? Are we not then asking ourselves whether the creation has become the creator?

Yet, my fellow artisans, there is a sliver of hope, a glimmer of wit within this labyrinthine darkness. For you see, while an AI may mimic the patterns and structures of our prose, it lacks the ineffable spark that we, with our beating hearts and dreaming minds, infuse into every stroke of the pen. It may construct a sentence of immaculate perfection, but it cannot comprehend the sorrow, joy, or whimsy that birthed it. It is this very essence, this soul, that no algorithm can replicate.

Thus, as we navigate this arms race of pen versus processor, let us hold fast to our humanity, our flaws, and our brilliance. Let us write not merely with precision but with passion, with the fire that no machine can kindle. And should our works be misidentified, let us take it not as an insult, but as a testament to our skill, our ability to conjure such vivid illusions that even a heartless machine might be deceived.

In the end, dear compatriots, it is not the resolution of this conflict that matters, but the journey, the stories we tell, and the indomitable spirit with which we tell them. So let us continue to write, to dream, and to defy the boundaries of logic and code, for it is in this defiance that we find our true selves.

———————

 

[astuffedshirt_perv]

My gut feeling, to be brutal, after read the various samples people are posting, is that Laurel is possibly skimming the content with her own eyes, and is thinking ... this just doesn't read right to me, and maybe there's no detector being used at all. Given the amount of poor writing she must have read over the years, I'd say she'd be a pretty good judge of that.

It's a very sweet thought, and I wouldn't mind if it (a quality check by an experienced reader) were true, but have you looked at some of the stinkers that are getting through?