• dubyakay@lemmy.ca
    link
    fedilink
    English
    arrow-up
    11
    arrow-down
    5
    ·
    16 hours ago

    I have an acquaintance who is a lead Dev at an Indie studio where he is developing and training an NPC behaviour engine with thousands of responses and actions. Think fallout or mass effect response wheel, where 2-4 dialogue choices have 2-4 outcomes, but instead you can tell the NPC anything and it will have a different response. Or it will do different things whether you hand it a book, give it book, throw a potion at it or cast a healing spell on it or hug it. It could also change tactics if you tried to snipe it vs if you went at it melee. All of these are trained and accounted for and made in a way where it can be built into any game using a certain engine. And this is just aimed at generic npcs, not companions.

    So if this is what disclosure of the use of generative AI means, I’m not against it. I think there is nuance to what can be done with it. Using final art assets? It’s theft. Writing? Theft. NPC behaviour? Definitely not.

    • shoo@lemmy.world
      link
      fedilink
      English
      arrow-up
      5
      arrow-down
      2
      ·
      12 hours ago

      Strange to not qualify the last one as theft. If it’s out putting code, it’s from the same kind of training set. If it’s out putting character responses, they’re from that same literary training data.

      • Kay Ohtie@pawb.social
        link
        fedilink
        English
        arrow-up
        3
        arrow-down
        4
        ·
        11 hours ago

        Open-source training texts intended for pairing with your intended style of output have been around for far longer than OpenAI has been grifting data from the entire Internet and collected book works. It came across like that’s what they’re using, not some shit off HuffingFarce that was built off of AO3 and Harry Potter.