Comment on Japan determines copyright doesn't apply to LLM/ML training data

• camelbeard@lemmy.world ⁨11⁩ ⁨months⁩ ago

  So if AI companies pay for a book or music (like a consumer) it’s no problem? Because I don’t think this is about paying for content, it’s that content holders refuse to work with AI companies.

  source

  • Mustard@lemmy.blahaj.zone ⁨11⁩ ⁨months⁩ ago

    Unironically yes, if AI companies paid for training data everyone would be much happier.

    I sincerely doubt that NOBODY is willing to sell data to them. It’s far more likely that they have not offered anyone a fair price yet, which makes sense because that would set a precedent.

    Even then, if people don’t want to sell them their copyrighted work then tough. You can’t compel people to take customers they don’t want.

    source

    • ArmokGoB@lemmy.dbzer0.com ⁨11⁩ ⁨months⁩ ago

      So if I go on a free website that hosts art (ArtStation, DeviantArt, etc.) and get training data that I could have legally accessed for free…

      source

      • Mustard@lemmy.blahaj.zone ⁨11⁩ ⁨months⁩ ago

        They’ve all already done that haha. You could argue that a human has only one life in which to remix that art but an AI is theoretically immortal, so it’s a different category of customer.

        At any rate, it’s clear that AI should not have free access to copyrighted works, like news articles, academic papers, stock images, and various kinds of non deviantart art.

        source