• sdf_iain@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    They just have to pay royalties to all of the source material (it all contributes to the weights in the model). Isn’t it just like sampling music?

    I imagine with these large models it should be a few thousand percent of any revenue generated. Any output should be commercially non-viable.

    • taleo@alien.topB
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      That would be like trying to sue any songwriter who learned I IV V I is a decent chord progression, or any artist that uses perspective.

      • sdf_iain@alien.topB
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 year ago

        Would make you think before using EVERYTHING as training data, wouldn’t it?

        There are no royalties on the public domain