• The_Monocle_Debacle@lemmygrad.ml
      link
      fedilink
      arrow-up
      8
      ·
      edit-2
      3 years ago

      most of these so-called “AI” implementations are just self-optimizing algorithms trained with incomplete or biased data for a very specific problem. A lot of them can’t even do something in the same problem space that wasn’t part of their training data correctly.

      • Amicese@lemmy.ml
        link
        fedilink
        arrow-up
        3
        ·
        edit-2
        3 years ago

        Oh yeah I see what you mean. I struggle with discerning them though.

        I worry that the training data for deepfakes is suspiciously normative. (there seems to be no neurodiverse, queer, or (physically) disabled people in those training sets).

        • southerntofu@lemmy.ml
          link
          fedilink
          arrow-up
          3
          ·
          3 years ago

          Well first deepfakes need to die. It’s a dangerous tech that should not exist at all and does not need any more research.

          To be fair, i haven’t dug into deepfake models, but i assume you would train them on the specific person you’re trying to deepfake: i mean for basic video stuff going with a pre-trained model may be ok but for audio there’s no way you can get away with it ;)

          • poVoq@lemmy.ml
            link
            fedilink
            arrow-up
            2
            arrow-down
            2
            ·
            edit-2
            3 years ago

            There are also specific ML models for audio that sound pretty convincing in replicating a specific person’s voice.