• XiJinpingStanAccount@lemmy.ml
    link
    fedilink
    arrow-up
    10
    ·
    3 days ago

    Honestly that’s a really good point. Where the fuck did the training data come from? If AI can only make what its trained on. I hadn’t really thought about that in relation to this. How isn’t that a bigger talking point?

    • YourMomsTrashman@lemmy.world
      link
      fedilink
      arrow-up
      4
      ·
      2 days ago

      On the one hand, diffusion models can merge concepts together pretty well (think of those bird-cat chimeras).

      On the other hand, yes, there’s CSAM in training data. A while ago researchers scanned and validated thousands of CSAM images in the training data of just one dataset. This one in particular the backbone of the popular Stable Diffusion.

        • anhopki1@lemmy.ml
          link
          fedilink
          English
          arrow-up
          1
          ·
          2 days ago

          They probably wanted some ai generated CSAM for themselves too. I honestly wouldn’t be surprised.

          • XiJinpingStanAccount@lemmy.ml
            link
            fedilink
            arrow-up
            2
            ·
            2 days ago

            I mean its either gross negligence or active criminality. If PornHub had CSAM on it and just failed to take it down they’d get into serious trouble. An AI company including it in their data sets and then letting their models be used to make more of it is arguably much more blatant than that. Data sets don’t just appear out of thin air. Someone has to make them. We know for a fact they curate them too.

    • shalafi@lemmy.world
      link
      fedilink
      English
      arrow-up
      4
      ·
      3 days ago

      If the AI knows what a naked adult looks like and knows what clothed child looks like, I don’t see why it couldn’t mash that up. I’d also assume it could train on medical reports as well.

      I’m also quite certain the model reached into plenty of dark areas of the internet and picked up CSAM.