• Moonrise2473@feddit.it
    link
    fedilink
    arrow-up
    1
    ·
    edit-2
    1 month ago

    I feel it’s just a side effect of them trying to block ai companies stealing large amounts of videos for training models. They see too many downloads from a datacenter IP address and require user login to continue

    Openai’s whisper often recognizes mangled words as “please like and subscribe” so they’re actively stealing videos and their subs (the manually created ones by companies like “caption+ by js”, which creators paid hundreds of dollars to make, not the free ones made by Google automatic transcriber or whisper itself) to improve their models so they can make profit

      • Moonrise2473@feddit.it
        link
        fedilink
        arrow-up
        2
        ·
        edit-2
        1 month ago

        Stealing, without the quotation marks. If you copy something and profit off it without crediting, compensating or asking permission to who paid for it, it’s stealing. We can’t downplay it as “but they just downloaded 700k hours of videos and 200k pirated books for training a simple model that they’re charging users $20 a month, what’s the issue”

        If you copy something for personal enjoyment without profiting from it, then it’s not stealing.

        • FozzyOsbourne@lemm.ee
          link
          fedilink
          arrow-up
          1
          ·
          1 month ago

          I get your point, it’s just hard to give a shit when one amoral megacorp takes some profit away from another. Google owns and profits from YouTube videos and occasionally throws a few pennies to the creators if they haven’t broken this week’s selection of ever-changing arbitrary rules.

          • Moonrise2473@feddit.it
            link
            fedilink
            arrow-up
            1
            ·
            1 month ago

            Probably Google just wants to block them not because they care about the creators but because it’s costing them bandwidth money.

            From the new agreement that they had with warner bros for creating closed captions, it looks like Google is also stealing the subs for training, they had direct access

            It just sucks that someone pays hundreds of dollars to have a human create subs for a show, then that is used without credit or permission for training a model (actually whisper accidentally credits moments of silence with the name of the subbing groups used for training)