• jsomae@lemmy.ml
    link
    fedilink
    English
    arrow-up
    46
    ·
    1 day ago

    Can the rest of us please use copyrighted material without permission?

      • nodiratime@lemmy.world
        link
        fedilink
        English
        arrow-up
        6
        arrow-down
        2
        ·
        24 hours ago

        The AI just gives you a 1:1 copy of it’s training data, which is the material. Viola.

    • FreedomAdvocate@lemmy.net.au
      link
      fedilink
      English
      arrow-up
      5
      arrow-down
      18
      ·
      edit-2
      22 hours ago

      You already likely do. Every book you read and learned from is copyrighted material. Every video you watch on YouTube and learned from is copyrighted material.

      The “without permission” is not correct. You’ve got permission to watch/listen/learn from it by them releasing it and you paying any applicable subscription etc costs. AI does the same.

      • jsomae@lemmy.ml
        link
        fedilink
        English
        arrow-up
        11
        ·
        edit-2
        20 hours ago

        By “use” I actually meant “reproduce portions of” and “make derivative works of”

          • jsomae@lemmy.ml
            link
            fedilink
            English
            arrow-up
            1
            arrow-down
            1
            ·
            4 hours ago

            For sure, AI can reproduce wholesale verbatim copies of text from miscellaneous sources. It can also create images that are so close to random deviantartists’ images that it’s undeniably plagiaristic. I expect this bug will be worked out eventually, but it is currently quite capable of doing this. In other words, you could say the weights contain a lossy encoding of many artists’ works, and those works can (lossily) be eked out of the model with some coercion.