• Fifrok
    link
    fedilink
    arrow-up
    3
    arrow-down
    1
    ·
    3 months ago

    You do realize that people can still tell when slop is passed as music. And the only reason it’s so prevalent on spotify and whatever other corpo platforms is because it’s used to pay artists who upload there less, right?

    • whoever loves Digit 🇵🇸🇺🇸🏴‍☠️@piefed.social
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      2
      ·
      3 months ago

      You do realize that people can still tell when slop is passed as music

      In the past, but what about the future? Or even the present?

      And the only reason it’s so prevalent on spotify and whatever other corpo platforms is because it’s used to pay artists who upload there less, right?

      Sounds right, yeah, but what does it have to do with what I said?

      • Fifrok
        link
        fedilink
        arrow-up
        2
        ·
        3 months ago

        Or even the present?

        Still somewhat easy, but it might take a couple of seconds. In the future it might get hard, but since generative models are already poisoning all training data sources, I hope it will take a while.

        what does it have to do with what I said?

        That this will ‘work’ to reduce slop visibility just because that it won’t be actively pushed by the company, despite how hard it is to track and remove the garbage itself.

        • jarfil@beehaw.org
          link
          fedilink
          arrow-up
          1
          ·
          2 months ago

          Poisoning what? Intentional poisoning doesn’t work, and self-consumption only works when it’s the exact same model feeding its own next version.

          • Fifrok
            link
            fedilink
            arrow-up
            3
            ·
            3 months ago

            Maybe you’ve only heard the least realistic sounding algorithmic music or something

            Might be that tbh, I listen to music mostly in Polish. And the english music I listen to leans on the niche side, so might not have enough training data scraped to make it not weird sounding