Due heavy demagoguery censorship of .world irt Iowa Senate picks pro 🏳️‍⚧️genocide, we urge this /c/ to emigrate to !politics@lemmy.dbzer0.com

I am tired of being treated like congregational rug. Either treat us with respect, or sulk in the deathcamps.

  • TheTechnician27@lemmy.world
    link
    fedilink
    English
    arrow-up
    3
    ·
    edit-2
    2 days ago

    Only tangentially related: someone in that thread posted this link to an exposé by The Needle meticulously detailing and proving beyond any doubt that Erin Reed of EITM has been having an LLM write huge portions (read: the majority) of her prose, even moreso this year after she got hooked on Claude Opus 4.6.

    It’s so fucking disappointing, but Erin is abusing LLMs to fuck and back and has been for years. She’s lost any credibility, although I’ll be interested now to check out The Needle’s work.


    Edit: It’s a long piece, so if you’re strictly interested in proof positive of her extensive abuse of LLMs, check out sections 6.1, 6.2, 6.3, and 6.4. If that’s too much, Ctrl+F for “The Rate of Straight Marks Spikes After Release of Claude Opus 4.6” and look at the graph as just an example of how dire it is. (Notably, the context is she was using ChatGPT before this, which at the time used curly marks. Then she switched to Claude Opus 4.6, which uses straight marks.)


    Edit 2: I forgot to give a special mention to Section 3 where Reed deadnames a transmasc author by using an LLM. (His name is Nate; Erin’s piece calls him “Natalie”.) The authors of the exposé note they can’t tell if this is his real deadname or an LLM hallucination, but both are egregious.

    • Walters, Fern@bookwyr.meOP
      link
      fedilink
      English
      arrow-up
      0
      arrow-down
      1
      ·
      1 day ago

      Jane Migliara Brigham & Artemis T. Douglas: We then used machine learning and stylometrics

      …

      • TheTechnician27@lemmy.world
        link
        fedilink
        English
        arrow-up
        2
        ·
        1 day ago

        … Yes? Things like a topic model, for example.

        I don’t know if this is supposed to be a “gotcha” or if there’s just some genuine fundamental misunderstanding here of what machine learning is, but things like stylometrics on nearly a million words is a well-studied, well-proven classical machine learning task.

        Machine learning at its core is statistics on large amounts of data, and they explain their methodology at each step.

        And they of course cite peer-reviewed studies for the stylometric methods they use; they don’t just copy–paste her articles and ask: “Chat[GPT], is this real?”

  • njm1314@lemmy.world
    link
    fedilink
    arrow-up
    1
    ·
    2 days ago

    I wouldn’t know. Been banned there for years because Jordan doesn’t like if you say mean things about nazis.

  • TropicalDingdong@lemmy.world
    link
    fedilink
    arrow-up
    1
    ·
    2 days ago

    Not the place for this, but I wholeheartedly support your point and second your sentiment.

    It’s a soft biasing of the forum.