People are using AI but not disclosing it. How do you detect it ?

[NumberSix]

See, the thing is, I don't really think you were trying to be funny.

Click to shrink...

No, I think you just don't appreciate proper sarcasm.

 

[The Radiance's Follower]

See, the thing is, I don't really think you were trying to be funny.

Click to shrink...

Indeed he was not. He is one of the primary reasons of why I want to lock the thread. He talks in bad faith, lies, distort, and is all around an unpleasant person to interact with.

 

[xihute]

A.I. generated text detection does not work.

And even worse, it penalizes non-native English writers, like myself.

Click to shrink...

I agree that it's why "punishment for not tagging" cannot work, because there's no reliable method of detection without fail-positives.

On the individual level, if someone would use such AI detection for filtering, it can work. It only has to be reliable enough.

I get the fear over potential for witch hunts, and slippery slope - I just read about a case where a story was retracted from a magazine over AI use accusations.

The story got into the magazine, passing whatever quality control they have, to meet accusations of AI-use and was retracted due to backlash. No proof, just suspicion. There's no way to prove your innocence, the accusation can be the verdict.

LurkingInTheDeceit gave good reasons why even the things that I thought as harmless, can have unintended harmful consequences - the change of attitude among users what is allowed and what's not.

I would think that the best solution, at the moment, would be some AI detection at the level of users, that would tag things as AI for them. The best solution ignoring the technical difficulties, of course. (It would also necessity use of AI, so I can see why people that don't want consume AI due to their ethical reasons wouldn't want also use AI for AI detection)

 

[NumberSix]

Indeed he was not. He is one of the primary reasons of why I want to lock the thread. He talks in bad faith, lies, distort, and is all around an unpleasant person to interact with.

Click to shrink...

I am sorry that people are not jumping on your band wagon and your bad idea is getting push back.
Actually no, I am not. That was the only lie that came out of me in this thread.

 

[Arc-on-Fire]

I don't think that would work as well as you suspect.

Click to shrink...

Really? Give some examples, here on the CrW.

If you pull the AI card and then go ''Wait, no, I was joking guys, this isn't AI generated I swear. For reals.'' Quite a lot of people will not, in fact, believe you even if your writing style is radically different. All you'll get by saying your prose isn't Ai adjacent is people going ''Oh, so you used a custom model then.''

Click to shrink...

A. Beyond my initial statement of saying I haven't used AI. I don't have to manually prove anything; I can just point to my writing:

Spoiler: The beginning chunk of a new fic I'm drafting

THIRTY EIGHT HOURS WAS ALL I THOUGHT WHEN --

The cracked brass bells sounded in the early morning. The sound rumbled through the cobbled streets of Daedalus. My heel ceased its tapping and outside: I saw grumbling men packing wagons alongside women with bruised eyes jerking steaming loaves into baskets. From a thinner ally, a trio of dirtied barefoot orphans stumbled into the street, mewling at the people for breakfast. Inside, when I turned to bed, a little lump began to quiver underneath the thick quilt.

I bit into a chunk of stale brownbread smeared with cheese. The stale crumbs sliced into my gums even with the sour brie dulling the grains.

I'd been awake since five this morning, courtesy of Bell's innerclock, and the truth was this: thirty-eight hours of work inside the Dungeon to make a semblance of rent.

"All day and tomorrow," I murmured as I licked at my scraped gums.

I repeated the number again like it was a prayer but it didn't get any smaller.

The ratty stool groaned as I stood. As the bells rang, the bed beside me squeaked and a blackhaired head popped out from underneath the thick quilt. Bright blue eyes bleared open and stared for a moment; they watered with sleepiness.

For a moment, the day ahead seemed to fizzle dead as I met her quivering stare. Man… why was Hestia so cute?

"Hnnng!"

She leaned up and stretched and the bed pomfed as she fell back.

"Bell… stay this morning," Hestia said, her voice scratchy as she extended one hand toward me.

I chuckled before giving her a cup of water that sat on the end table.

"Sorry, don't think I can," I said, giving her the ceramic.

She wrinkled her nose for a moment before she supped at the cup with loud gulps. Then she gave it back to me.[[ Arc, 7/4/2026 5:03 AM']

[Reword, as Somebody suggested: the action description here would be better used to convey Hestia's character here instead of just having it be logical flow.]]

"Just… stay for this morning. Okay?"

"What do you wanna do this morning, then?" I said.[ See about expanding more friction. Bell here just flat out agrees with her instead of raising his concerns over their current financial situation. This expanded friction would smooth out the interrupted flow when he shows her the flyer.]

Hestia dragged herself into my lap, her gaze half-lidded.

"I dunno." She muttered. Hestia rose a lazy hand and poked at my white fringe.

I sniffed at that, "Are you going to stay inside all day again? Go back to that bookshop, maybe?"

Click to shrink...

Notice the smallcaps opener and the manually inputted em-dash. The irregular compound nouns and -- the biggest killer of AI Generated accusations; "pomfed"

That word does not exist in the English language at all. Its an onomatopoeia translated into verb form, captured from the infamous meme of the same name. Do you honestly think an AI would willingly convert an onomatopoeia, one associated with loli hentai nonetheless, and use in technical and formal writing? I don't think so. There's also more examples that I could give: double entendres, references to copyrighted material, the protagonist even graffitiing a slogan with a racist slur on the door to the Guild's headquarters after a certain thing which I won't remark on.

At the end of the day, when you basically say that people will default to the accusation of "Okay, you're using a custom model then." -- then what is the point of this thread? How do you detect AI writing? Can you even detect it in the first place?

You're probably also overestimating the amount of people who would show up to bitch in your thread.

Click to shrink...

Are you sure about that? (Btw, searching for "AI" just in that thread alone, brings up more than six pages of arguments and then some. Obviously, there's jokes in there and shit but still.)

Is the outcome actually really worth writing a whole story the hard way with no feedback or even knowledge of whether it will be received well until it's too late to change it?

Click to shrink...

>release first chapter, get to work on the second. See what'll happen innathread in the mean time
Yeah lmfao. I won't have to worry much about feedback because Writers Write about Writing exists. Get feedback and help there WAY more better than the ever infamous:
>no pfp
>lurker
>ESL
Slopbattler Johnny!

 

[BronzePlaceWriter]

Really? Give some examples, here on the CrW.

Click to shrink...

Give examples of a hypothetical, subjective idea, based on a combination of my own opinion and a thing you have not actually done yet not working...?

A. Beyond my initial statement of saying I haven't used AI. I don't have to manually prove anything; I can just point to my writing:

Click to shrink...

Pointing to your writing isn't going to work. People already have a hard time telling Ai and not Ai writing apart. So if you tag it as Ai and then later go ''Woops, I was writing it by hand all along! It's not Ai at all!'' I would hazard to say that a signifigant amount of people will not in fact go ''Oh, you got us! You totally fooled us!''

No, they will say ''Well, if this isn't Ai, why did you flag it as AI? What would be the point of that? And now it's suddenly not? I don't actually believe this isn't Ai, I think you're just trying to widen your audience.''

You can then point to your writing, but if your audience is already expecting it to be Ai, then sorry, but it's going to read like Ai to them. The human brain is great at finding patterns it expects to already be there. You primed them to see it as Ai, so they will.

Notice the smallcaps opener and the manually inputted em-dash. The irregular compound nouns and -- the biggest killer of AI Generated accusations; "pomfed"

That word does not exist in the English language at all. Its an onomatopoeia translated into verb form, captured from the infamous meme of the same name. Do you honestly think an AI would willingly convert an onomatopoeia, one associated with loli hentai nonetheless, and use in technical and formal writing? I don't think so. There's also more examples that I could give: double entendres, references to copyrighted material, the protagonist even graffitiing a slogan with a racist slur on the door to the Guild's headquarters after a certain thing which I won't remark on.

Click to shrink...

I can literally see you saying this exact thing on the thread and then someone pointing out a custom model could do all of that.

At the end of the day, when you basically say that people will default to the accusation of "Okay, you're using a custom model then." -- then what is the point of this thread? How do you detect AI writing? Can you even detect it in the first place?

Click to shrink...

I wasn't talking about detecting Ai writing in this case so why are you bringing that up with me or replying to me about it?

People are likely to assume that you're using a custom model specifically because your original plan was to tag it as AI then later take that back. Since most people try to hide their AI or stay undisclosed, there is no reason for your viewers not to believe you when you say it's ai. So when you later take it back, it looks like you're trying to hide that fact now it's getting big.

People are far likely to decide it's a custom model because that sounds much more reasonable to most than ''I intentionally tagged my non-Ai stuff as Ai''

Are you sure about that? (Btw, searching for "AI" just in that thread alone, brings up more than six pages of arguments and then some. Obviously, there's jokes in there and shit but still.)

Click to shrink...

And how many threads does that happen on, and how many people do it, and how concentrated were those posts? Just because one thread has a specific derail doesn't mean that you can use that for your thread.

>release first chapter, get to work on the second. See what'll happen innathread in the mean time
Yeah lmfao. I won't have to worry much about feedback because Writers Write about Writing exists. Get feedback and help there WAY more better than the ever infamous:
>no pfp
>lurker
>ESL
Slopbattler Johnny!

Click to shrink...

wasn't your original idea that you were going to do the whole story first and then post it? Because if you don't do that, you'll get buried by the actual Ai stories that can produce chapters on demand. If you try to release chapter by chapter here, you're going to fall way behind. As a (pretend) Ai story, you've cut off a portion of your audience that wants to read real story. As a (hidden) real story, you can't generate or post nearly as fast as the true AI stories.

It's the worst of both worlds.