you are viewing a single comment's thread
view the rest of the comments
[–] 11 points 2 months ago (29 children)

It was posted like some kind of gotcha but....this book was in the dataset.

I'm aware that LLMs don't keep the dataset in memory but it "knows" that this is an existing work but these sites aren't doing any wild calculations to figure out if it was AI-generated. They basically just check to see if the sentences exist elsewhere and they do. In the original dataset.

So it was wrong to say it's AI-generated but it correctly identified that it's not original.

  • source
  • parent
  • hideshow 29 child comments
  • [–] 93 points 2 months ago (9 children)

    They analyze statistical patterns, they don't cross reference the training data.

    It's picking up the text as generated because it's mostly guess work and constantly spits out false positives, especially with non native speakers.

  • source
  • parent
  • hideshow 9 child comments
  • [+] -6 points 2 months ago (8 children)

    Honestly I didn’t expect them to flag non-native speakers. LLMs don’t often make gramatical mistakes (or at least not ones your average joe could identify)

  • source
  • parent
  • hideshow 8 child comments
  • [–] 48 points 2 months ago (6 children)

    Foreign speakers often know English better than most Americans

  • source
  • parent
  • hideshow 6 child comments
  • [–] 6 points 2 months ago* (last edited 2 months ago)

    Oooh yeah, the most common dumb mistakes are (by my own checking) done by Americans in the VAST majority cases. Stuff like not knowing how to use your/you're, there/their/they're, cloths/clothes, definitely/defiantly, where/were/we're, its/it's etc. correctly and the many other failings of basic grammar like "would/could of" and so much else.
    I don't think there's another country in the world where the native speakers are so fucking bad at even the most basic shit.
    The most common response I've gotten and seen when others correct stuff like that is "calm down, English probability isn't their main language" yet when checking it's been Americans who made the mistakes 99% of the time.

  • source
  • parent
  • [–] 3 points 2 months ago (3 children)

    The decline I've seen in my time on the internet is genuinely shocking

  • source
  • parent
  • hideshow 3 child comments
  • [–] 3 points 2 months ago

    Weirdly enough i saw a massive increase in little typoes on reddit this year. More typoes and such in a few months than i saw in years.

    It feels like LLMs started making mistakes on purpose.

  • source
  • parent
  • [–] 32 points 2 months ago (2 children)

    That is not in the least bit how a tool like this works.

    All AI detection is extremely unreliable, but they operate on principles which, if the assumptions supporting them were true, would be sound. The way you imagine they work is different: you're describing a "novel text detector" which is not at all the same thing as an "AI text detector".

    AI detection works, at a very high level, by throwing a bunch of examples of AI text, and a bunch of examples of non-AI text, into a machine-learning model, and training it until it is able to recognise the AI examples as such. It doesn't work by throwing in all existing text including novels written before AI, because that would never do what you want.

    (The reason, if you're interested, why this ends up not working is because the features such a model identifies as indicative of AI text are very sensitive: if you train it on Claude and ChatGPT, it will not work on Gemini output. If you train it on Gemini, when Gemini updates it will get worse. If someone generates text with a weird prompt, it may slip by. If someone writes in a weird way, it may get flagged. And if any AI company wanted to defeat AI detectors, it could trivially feed its output through one during the training process and give that output as examples to avoid in the training, so that the model learns to create output which doesn't "look like" AI output to those detectors.)

  • source
  • parent
  • hideshow 2 child comments
  • [–] 2 points 2 months ago (1 child)

    I was being overly simplistic - I meant more that the patterns it's trying to detect were created by an LLM trained on the data they're inputting.

    It's like how reddit comments from 2016 look generated. If you stick one of those into an AI detector, it'll give a false positive for the same reason

  • source
  • parent
  • hideshow 1 child comment
  • [–] 4 points 2 months ago (11 children)

    I'm sorry, what? It's correctly identified that it's what, now?

  • source
  • parent
  • hideshow 11 child comments
  • [–] -2 points 2 months ago (10 children)

    It's original in the book. What the user entered is an exact copy.

    "Original" is not the correct work but you know what I mean.

  • source
  • parent
  • hideshow 10 child comments
  • [–] 5 points 2 months ago (8 children)

    The little scale thing doesn't say it is 100% not original, it says it is AI/LLM generated.

  • source
  • parent
  • hideshow 8 child comments
  • [–] -2 points 2 months ago (7 children)

    Yes...because the pasted text has some of the exact patterns that are part of the dataset that the website trained on...because that dataset contains text generated from a dataset containing the exact paragraph...

    I feel like you're being deliberately obtuse.

  • source
  • parent
  • hideshow 7 child comments
  • [–] 6 points 2 months ago* (5 children)

    I feel like you don't understand the difference between 'AI generated' and 'something that existed over 200 years ago'.

  • source
  • parent
  • hideshow 5 child comments
  • [–] -2 points 2 months ago (4 children)

    I feel like you don't understand that LLMs train on text that existed 200 years ago...

  • source
  • parent
  • hideshow 4 child comments
  • [–] 1 point 2 months ago (3 children)

    So you confirm that you don't understand the difference. Ok.

  • source
  • parent
  • hideshow 3 child comments
  • [–] -2 points 2 months ago (2 children)

    Sorry for not answering your idiotic question. Also sorry that you're dead set on deliberately misunderstanding then doubling down on it. Hope it makes you feels smart!

  • source
  • parent
  • hideshow 2 child comments
  • [–] 0 points 2 months ago (2 children)

    Everything you said was hallucinated. Nothing you said was even slightly coherent or connected with reality. Absolutely nothing was correct. If you let your cat dance on the keyboard more sense would come out.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 3 points 2 months ago (1 child)

    That must have sounded really clever and powerful in your head for you to hit send, huh? Couldn't decide which zinger to go with so you sent all 4?

    I'd have given you more credit if you just posted the quote from Billy Madison. Try harder lol

  • source
  • parent
  • hideshow 1 child comment