Microblog Memes

AI is like: "i made this"

Damn, GPTZero also flags it as 57% AI.

I postulated exactly these issues with detecting LLM code in software projects, making "no LLM code"-type policies unenforceable. It doesn't work most of the time now, and it will only get worse in the future.

Before anyone goes on a wild goose chase, this excerpt is on the opening of Chapter 4 instead of 5 in the first edition.

Anyone surprised to find that public domain art is determined to be AI generated does not realize that feeding an LLM requires a lot of art, and public domain is the easiest one to come across. AI detectors, as far as I know, check AI signs, and (most) AI signs came from their use by public domain artworks.

I made a machine with the purpose of literally creating output that plausibly sounds like a human wrote it.

Now that it creates things that sound plausibly like they came from a real person it's a problem. Now quick make something else that pretends to know it's not human when the whole point of the damn machine is to sound human.

Tjciiktn it's ridiculous. HMsie?$+'izhav!&(:8_! This world is so fucked

confidently incorrect

Yeah yeah AI system has false positives. Like a human, really. In fact humans are pretty bad at detecting AI text: https://arxiv.org/html/2211.13087

Among AI models, Blenderbot stood out: in AI-AI conversations, it was judged human 67% of the times - more often than actual human-human conversations. In human-AI conversations, human participants were labeled as humans 68% of the time, and AIs were classified as AI 56% of the time.

(FYI, a detection rate of 50% is the same as random chance, so 56% is pretty poor)

Shit like this happens if you train your AI on every god damn fucking text on earth.

I've been commenting on the open, public internet in some form or fashion since the 1900s. These models have been trained to sound like me, not the other way around. Fortunately, the bullies at the back of the class who don't pay attention have mostly caught on to this fact, after having it repeated to them hundreds of times. The other day I was thinking about it, and realized it's been a bit since anybody's accused me of using AI to write a comment. I think it's because I stopped proofreading what I write. Leave in those mistakes and the assholes at the back of the room who don't pay attention are too naive to think that machines can't make mistakes, I guess.

Intelligence is knowing that Frankenstein is the doctor. Wisdom is knowing that Shelley vibe coded the entire story.

It's a 100% match on what the model was trained on.

false positives exist, and the real monster is that it also happens with "faces" of criminals.

To be fair, the monster was artificial and intelligent.

We're using AI to determine what is AI, even though AI doesn't know what AI is.

Perfectly reasonable, no careers will be destroyed, surely

Does it think it's AI because it was trained on it so it recognizes it as being part of AI now?

And this is the "unseen" loss of value. Can you imagine writing a book (by hand the old-fashioned way -- maybe with just a text editor) and submitting it to publish and the publisher rejects it as "written by AI?"

Those detector sites are really evil. They simply do not work. They can't work. But they are advertising and people are using them to "check" texts. Which then totally fucks students when their idiotic teacher uses such a site and accuses them of using AI tools.

I realized my assignment went through an originality checker, which I guess is safer than calling it a plagiarism checker.

Still felt odd that it flagged a quote I used from the source book, in the section asking for a quotation. I would've thought it'd at least disregard things between quotation marks.

What's even dumber is that it flagged a single instance of the word "the", just the word not the sentence it was in. I don't know why, but I guess that specific "the" was unoriginal.