AI Allegations Thread

[EmilyMiller]

There is no best or worst. You'll get different results from them, but if they all scream AI, your in serious trouble.

Here are a few for our perusal.

https://www.scribbr.com/ai-detector/
https://www.scribbr.com/ai-detector/
https://copyleaks.com/ai-content-detector
https://writer.com/ai-content-detector/

I tried the last two, plus

https://www.zerogpt.com/

And another I can’t recall

Em

 

[nice90sguy]

That's a baseless presumption. But yes, I thought they read just fine.

It's not baseless, it's giving the software the benefit of the doubt. Admitedly the arms race between Ai and AI detectors is just warming up -- and the competition among AI fakery detectors is hot, and the stakes are high. Lots of money to be made if you can get a reliable one -- you could become the Norton Antivirus or Kaspersky of tomorrow!

 

[AwkwardlySet]

I am getting a 3%-8% AI score, but it is interesting to see which sentences the tool flags. Some of them make sense because of the blandness, but others...

 

[alohadave]

Are you the fucking reporter? Jesus Christ.

Seems likely.

It's not baseless, it's giving the software the benefit of the doubt. Admitedly the arms race between Ai and AI detectors is just warming up -- and the competition among AI fakery detectors is hot, and the stakes are high. Lots of money to be made if you can get a reliable one -- you could become the Norton Antivirus or Kaspersky of tomorrow!

OpenAI has pulled their AI detector because it doesn't work. They created ChatGPT.

 

[NoTalentHack]

Based on your suggestion, I'm taking a look at their catalogue just now. One of those at the top of their list is After the Future is Gone.

A section of that story has been flagged as 81.1% likely to have been generated by AI.

View attachment 2292292

I've barely even begun looking into their back catalogue, but entire sections are being flagged as strongly suspicious.

This is just using the example of the author you provided in response to me, completely unprompted, but I'll send a report off to Literotica and ask them to do a similar sweep of this author's work in the same manner as the OP.

Every author's in the same boat though. Those who didn't cheat by using software or tools to help create their stories have absolutely nothing to worry about. Also, Literotica is only going to delete stories that meet their criteria for failing the scans at their end. Scores of >50%, >70%? It's entirely up to them.

81.1% is a very high score for an uninterrupted block of text.

Fucking LOL. Bring it on.

Those who didn't cheat by using software or tools to help create their stories have absolutely nothing to worry about.

Okay, HUAC.

 

[nice90sguy]

OpenAI has pulled their AI detector because it doesn't work. They created ChatGPT.

Obviously there could be more than one reason for that.

 

[EmilyMiller]

I've tried one myself and yeah, it has 25k characters limitation. It probably gets better if you register?

One was only 1500 characters, which you’d not think enough to be statistically significant

Em

 

[NoTalentHack]

Seems likely.

OpenAI has pulled their AI detector because it doesn't work. They created ChatGPT.

I was about to say this. In the AI tools thread, we were playing around with ChatGPT's "critique" capabilities, and it's absolute nonsense. If AI could even semi-infallibly detect AI generated text, it wouldn't create obviously AI generated text in the first place.

 

[Red_Herring4]

OpenAI has pulled their AI detector because it doesn't work. They created ChatGPT.

Yes, and Vanderbilt University has decided to disable Turnitin’s AI detection tool for the foreseeable future. The linked article states:

"Based on this, we do not believe that AI detection software is an effective tool that should be used."

 

[nice90sguy]

I use github copilot in my daily work. To me, it's even more eerily smart than ChatGPT. Many times a day I think to myself, "how the FUCK did it know that's what I wanted to type?"

 

[AwkwardlySet]

I was about to say this. In the AI tools thread, we were playing around with ChatGPT's "critique" capabilities, and it's absolute nonsense. If AI could even semi-infallibly detect AI generated text, it wouldn't create obviously AI generated text in the first place.

Check post #256

 

[EmilyMiller]

George / Eric would be so proud…

https://imgur.com/a/VqYQw6k

 

[MillieDynamite]

I don't know how many I have bookmarked, but I think it's five. I don't have a paid version, so you're limited to between 1500 and 5000 characters on most. One only allows so many checks a day. But I'll be damned if I'm paying for it when I write every friggin word with only three of us, Me, Myself, and I, doing the writing.

 

[nice90sguy]

If AI could even semi-infallibly detect AI generated text, it wouldn't create obviously AI generated text in the first place.

I don't get your reasoning there. Are you suggesting they used adversarial training? I'm not sure they did

 

[MillieDynamite]

AI detects AI written material as a professional courtesy to itself, saying, "Yeah, I write like that."

I was about to say this. In the AI tools thread, we were playing around with ChatGPT's "critique" capabilities, and it's absolute nonsense. If AI could even semi-infallibly detect AI generated text, it wouldn't create obviously AI generated text in the first place.

 

[OverconfidentSarcasm]

Okay, HUAC.

What does the House Un-American Activities Committee have to do with this?

 

[Kelliezgirl]

Based on your suggestion, I'm taking a look at their catalogue just now. One of those at the top of their list is After the Future is Gone.

A section of that story has been flagged as 81.1% likely to have been generated by AI.

View attachment 2292292

I've barely even begun looking into their back catalogue, but entire sections are being flagged as strongly suspicious.

This is just using the example of the author you provided in response to me, completely unprompted, but I'll send a report off to Literotica and ask them to do a similar sweep of this author's work in the same manner as the OP.

Every author's in the same boat though. Those who didn't cheat by using software or tools to help create their stories have absolutely nothing to worry about. Also, Literotica is only going to delete stories that meet their criteria for failing the scans at their end. Scores of >50%, >70%? It's entirely up to them.

81.1% is a very high score for an uninterrupted block of text.

Might I suggest running some of your own work through the tool?
That way we get a comparison of something that you can guarantee is not AI.

 

[ronde]

You missed out the word "warped"

I prefer to think of myself as "uniquely different".

 

[TheArsonist]

I also heard back from Laurel. This was the main bulk of her message:

I rechecked your work and do still see parts that appear to be "reworded" by generative AI. Please be aware that Pro-Writing Aid does use generative AI. Generative AI does NOT "create" a story - it mashes together words from a library of (stolen) works, and the result is bland and toneless. If everyone on Lit chose to use generative AI, every single story on the site would sound exactly the same. Readers hate generative AI text, and they (and us as well) would rather you submitted grammatically wonky work in your own voice than submit generative AI.

We believe that writing is thinking. When you're putting your thoughts to paper (or a .doc file ) - be it a grocery list or a novel - you're organizing and sharpening your thoughts. Every single person who can read can craft a story of some kind. Some of my favorite writers (on Lit or elsewhere) would be flagged by Grammarly. Grammar software is NOT a calculator. While math is fairly objective, grammar is not. Relying on a grammar corrector will not make you a better writer.

Obviously Laurel can choose any policy she likes for her site, but this strikes me as an absolutely terrible one.

She doesn't specify whether her determination is based on an AI text detector or just her subjective gut feeling, but the detection tools are notoriously unreliable and her gut neither available for consultation nor apparently infallible. The number of authors who have had stories flagged and credibly insist that they are wrongfully accused shows—in my opinion—that it's a crap shoot even if you're not using AI assistance.

So what it amounts to is that she'll arbitrarily delete your stories if she happens to dislike your writing style, based on no clear criteria. (The claim that "readers hate" it is disproven by the "H" ratings.) Am I wrong, or is that a radical shift in policy? — And not a very author-friendly one, either.

And that's not to mention that the policy will soon become hopelessly out of touch with reality. Applications like MS Word have long offered grammar and style checking (itself based on very primitive "AI"), and is there any doubt that in the near future these features will incorporate LLM-based suggestions? (They might already be in the latest versions, for all I know.) It's going to be as ubiquitous as predictive text for typing on phones. A blanket ban is not going to be workable, and will either be ignored or render the site obsolete.

 

[Devinter]

As I am sadly now included in the list of victims to the false claims, I decided to run some experiments.

I took a snippet of text - just under 5000 characters - from a story of mine that was falsely flagged earlier today and ran it through six different websites, and each and everyone of them said it was human although some of them highlighted a few sentences as potentially suspicious. The amusing part is that every single website highlighted completely different sentences. The worst score I got was from 'content detector', seen in the following image:



When "Honey, you're okay now" - the most basic of sentences - shows up as potentially AI written, I think that speaks volumes for how ineffective these tools are. I'm quite disappointed that some of my work is being falsely flagged as I can only see it as my English not being good enough due to it being my second language, and that the detection tools thinks I 'write weird'. Kind of hurts my feelings.

 

[NoTalentHack]

You know what's really funny? I remember chatting with a couple of friends on Discord earlier this year about the steps of AI art becoming, if not accepted, then inevitable. One comment I made:

This is the tipping point, I think. When artist start mistakenly thinking that human-created art is AI art, to the point where they’re accusing each other and it turns into the Spanish Inquisition, that’s the point of no return.

And we’re starting to see it.It’s very easy to go from “fuck you, I did this myself and was under deadline pressure, and that’s why it’s not great” to “fuck, I’m under deadline pressure, and the Adobe AI art tool is RIGHT THERE.

One of the criteria for determining AI-written text, as I understand it, is that it looks for things we'd consider "amateurish." Simple sentence structure, sentence length that doesn't vary, lots of telling instead of showing, a lack of immediacy, overuse of passive voice, etc. So the things that a detector is looking for are the same things that, say, a brand new writer on an amateur site is likely to be guilty of.

I looked at the part of mine that got flagged and... yup. That's very definitely the case there. That passage is literally one of the first things I wrote since I got back in the game, so I can see now where it's... fine. Not great, not even really good. Fine. Which is what detectors are likely to hit.

Of course, since none of these folks actually publish their algorithms (partially because they're paid services and partially because, as with most ML-driven tools, they only sort of know how it works), this is just a guess. But if I get called out again, I'm going to have words with that kid in China I'm paying to write my stuff.

 

[EmilyMiller]

As I am sadly now included in the list of victims to the false claims, I decided to run some experiments.

I took a snippet of text - just under 5000 characters - from a story of mine that was falsely flagged earlier today and ran it through six different websites, and each and everyone of them said it was human although some of them highlighted a few sentences as potentially suspicious. The amusing part is that every single website highlighted completely different sentences. The worst score I got was from 'content detector', seen in the following image:

View attachment 2292319

When "Honey, you're okay now" - the most basic of sentences - shows up as potentially AI written, I think that speaks volumes for how ineffective these tools are. I'm quite disappointed that some of my work is being falsely flagged as I can only see it as my English not being good enough due to it being my second language, and that the detection tools thinks I 'write weird'. Kind of hurts my feelings.

Sorry, hun.



Em

 

[NoTalentHack]

What does the House Un-American Activities Committee have to do with this?

"If you're not a communist, you have nothing to fear from us. So why are you scared? Is it because you're a communist?"

"I hold in my hand this list of 273 so-called writers that are using AI to generate text. I cannot give you my sources, just know that they, like me, are men that can be trusted."

 

[NoTalentHack]

Might I suggest running some of your own work through the tool?
That way we get a comparison of something that you can guarantee is not AI.

Oh, you mean the person with no favorites, follows, or stories and no comments on the forums, that created their account over a year ago and sat on it until now?

My guess is that they're being at least somewhat disingenuous in their attacks to begin with.

 

[MillieDynamite]

I'd say freedom would be the theme: freedom of expression and the right to choose how to express yourself. However, I don't support the use of AI for actual nuts and bolts writing.

What does the House Un-American Activities Committee have to do with this?