• RedstoneValley@sh.itjust.works
    link
    fedilink
    arrow-up
    1
    ·
    2 hours ago

    I think a major problem is that it is difficult to prove which IP is in the model data. That’s why the AI companies argue that there isn’t a verbatim copy in the model, and therefore it’s not theft. The law in most countries is not equipped to deal with this scenario

    • Zephyr@sh.itjust.works
      link
      fedilink
      arrow-up
      1
      ·
      edit-2
      2 hours ago

      Seems easy enough to prove with a court order. Short of that though I’ve seen people get models to perfectly complete content inferring that information is in there somewhere or at minimum the model is willing to go fetch that information breaching copyright. I am still curious if this is an issue in AI labs elsewhere or if it’s primarily a US / UK issue.