Post

I'm just gonna say: it somewhat baffles me that people who are understandably skeptical of opaque LLMs as legitimate sources of information will suddenly go full on "the computer KNOWS" when they are using opaque tools to "detect" AI. The computer does not know.

Susanna Kearsley@susannakearsley.bsky.social

❤️ Gillian Green, publisher at Tor Bramble, told The Bookseller: “HM Wolfe has confirmed that Daggermouth was written entirely without AI. We stand behind our author, and we do not recognise third-party AI detection tools as a fair or reliable measure of authenticity when used on their own.”

332 reposts1.2K likes

The data scientist who developed the "detector" was crashing out against people on the skyline the other day. Because he of course is an expert and everyone else is a rube.

I think you're mistaking two different people—the data scientist who works with those results and crashes out at people on bluesky a lot (I think his goals are very noble, he is generally trying to produce material to be used against AI companies in copyright cases; I question his methods)...

...and the developers of Pangram. Data scientist is Tuhin Chakrabarty, Pangram guy (on bluesky, though rarely) is Max Spero. For clarity/accuracy!

Also, wouldn't an ai detector end up being a perfect loss function to train an ai to be undetectable?

It's "strange" that this is only happening to a certain demographic of author and that the people who proudly tell everyone that they do use it still wind up getting book deals and don't have their careers ruined.

And to be clear by strange I mean reeks of racism and sexism. Cause I only hear of Black women's books being accused of gen AI use and then I see white authors (mostly men but also some women) who proudly talk about their use of it and nothing happens (outside of my putting them on my DNR list)

I mean, the “detection” tools were trained the same way as the creation tools; why in the world would they be trustworthy?? “Oh the machine that makes shit up says this is made up” is not something anyone should take as other than a bad joke, and at their expense probably.

Hell yeah! I'm a college instructor who submits ~70 AI cheating claims a year. Nobody ethical uses "detectors". I am firmly convinced the best detection method is a real conversation with another human being about their ideas and writing process.

Also, how DEEPLY STRANGE AND MYSTERIOUS that it was *this* particular author who was targeted for it. I can't imagine why she of all romantasy authors would be hypercriticised and suspected. What possible characteristic might she have that woud lead people to justify their own wariness. a MYSTERY.

Oh god, of course. When the CFO at the place I quit started looking around for jobs he couldn’t replace with AI, guess who held the jobs he thought could be done by AI?

I haven't read Daggermouth and I haven't even read the weird article about it because I find strange investigations based on ~analyzing~ the text to be... Questionable. But I don't like the fact that the self-appointed AI police are proceeding with such reckless ignorance and unhinged confidence.

I think current genAI is, for the most part, an anti-social attack on culture and society. And I'm gonna need y'all to stop making that position look ridiculous by turning yourselves into an over-correcting anti-social attack on culture & society.

Yeah I had to go outside to weed but I think this is also a very important aspect of this. It's always the racism. bsky.app/profile/haun...

👻evenmorehauntedtoasters👻@hauntedtoasters.bsky.social

Also, how DEEPLY STRANGE AND MYSTERIOUS that it was *this* particular author who was targeted for it. I can't imagine why she of all romantasy authors would be hypercriticised and suspected. What possible characteristic might she have that woud lead people to justify their own wariness. a MYSTERY.

Sadly, that was the first thing I thought of when I saw the author's photo. A lot of our society are seriously determined to persecute black women. ☹️

The other day I put into a checker a paragraph from a document that is a decade old at this point, and it said that 2/3 of that paragraph was AI generated. The detection tools are garbage.

One of the most exhausting things about the "AI" cycle is that every skill, every trick, every tool and every piece of published research is out of date by the time we master them. Even the most solid research is generations out of date by the time it's published. It's why the hype cycle dominates.

AI detectors are BS, but also it is sometimes possible with actual critical analysis to tell that prose did not come from a mind. Not often enough to catch every instance, but often enough that it should act as a deterrent if it's properly developed as a methodology.

People are just looking for attention because that’s what we place value on. It’s much easier, louder, and more effective to have a hate fueled zero nuance dogpile than have a reasoned opinion about something. I’d bet a lot of the loudest AI opponents use it. They’re just riding the outrage train.

I don't use it unless it's forced on me at gunpoint and most of the people I know are the same and we're all very vocal haters. I just think that the people being ridiculous are still being ridiculous even if I agree with them. 😂

I use AI as a software dev resource where I feel it’s less problematic, but never to generate content for me and I can’t even fathom using it for writing or art. I think that’s a very valid stance to not use it at all. But I also think people using AI art never wanted art to begin with.

This is a hard topic to boil down to 300 characters, but art is fundamentally about emotional connection and that kind of requires a person on both sides. I know a lot of artists make their living off people that don’t care about that. And AI + Capitalism is a mess.

To be clear, that’s not me saying “everyone uses AI” or “AI writing/art isn’t problematic.” I don’t think there’s any real space for LLM generated art. It’s not art as far as I’m concerned, it’s just meaningless content.

But it’s easier to jump to a conclusion that makes people mad. I just really don’t understand... The path to getting your work shared is the same as it’s always been: Getting people who appreciate it to talk about it. Are there people out there appreciating LLM generated books? They’re trash.

For all the problems with AI, the way it works is to produce text that is statistically similar to human created text, there's just no way to make a machine to accurately separate those out from one another.

I think that gives AI and text-prediction far too much credit. I have huge skepticism about the claims of quality/99.9% accurate /whatever of AI detectors; they are provably not random or completely unable to distinguish. I'd love to see more independent studies.

Seriously though, try running a hundred human pages and a hundred AI pages through Pangram and see what happens. I wouldn't recommend this with any other detector because they all steal the inputs; Pangram doesn't. See what happens. I doubt it will be perfect! But I think it will be informative.

Just as I think people saying "it's absolutely impossible for a person to EVER claim they can tell a piece of writing is AI" are giving LLMS way too much credit and human readers (and writers) not enough, and that if they read a few they'd know what people are talking about.