• erwan@lemmy.ml
      link
      fedilink
      English
      arrow-up
      3
      ·
      6 months ago

      That’s the crux of the problem: they’ll make it so you can’t (as for any other celebrity) but there is a high risk that the safeguards can be circumvented.

      • Adalast@lemmy.world
        link
        fedilink
        English
        arrow-up
        3
        ·
        6 months ago

        Just train a couple LoRAs and you can add in the vectors needed for doing whatever you want. “Circumventing” “safeguards” in this case means using a feature built into the system already. I have been making AI porn from the start. The day in installed their local framework for using it and got a Rick Astly image instead of anime tits I went and found their censorship function and neutered it. It is just python code, super easy to do.

    • General_Effort@lemmy.world
      link
      fedilink
      English
      arrow-up
      1
      ·
      6 months ago

      Good question. That is (almost certainly) political speech and as such especially protected by law. It’s also quite controversial and so companies will try to prevent their services being used for it.

  • Corroded@leminal.space
    link
    fedilink
    English
    arrow-up
    3
    ·
    6 months ago

    “We want to ensure that people have maximum control to the extent that it doesn’t violate the law or other peoples’ rights,” Joanne Jang, a member of the product team at OpenAI, told NPR. “There are creative cases in which content involving sexuality or nudity is important to our users.”

    The other problem in my mind is the fallibility of current safeguards. OpenAI and rivals have been refining their filtering and moderation tools for years. But users constantly discover workarounds that enable them to abuse the companies’ AI models, apps and platforms.

    Some highlights from the article.


    It seems like AI porn is inevitable and OpenAI has safeguards in mind for exploitative content so it doesn’t seem like a horrendous idea.

    • DarkThoughts@fedia.io
      link
      fedilink
      arrow-up
      1
      ·
      6 months ago

      to the extent that it doesn’t violate the law or other peoples’ rights

      Am I the only one who finds this so weird when we talk about LLMs? If someone makes a bot that resembles some specific person, that person’s rights aren’t really violated, and since they’re all fictional content, it is very hard to break actual laws through its content. At that point we would have to also ban people’s weird fan fiction, no?

      Not arguing about whatever they want or don’t want on their platform, but the legal & alleged moral questions / arguments always weird me out a bit, because there’s no one actually getting hurt in any sort of way by weirdos having weird chats with computers.

      The bigger issue is the enforcement. Either you monitor an absurd amount of content, which is worse for privacy, or you straight up censor the models, which makes them typically restrictive even in valid cases / scenarios being played out (other platforms went through this, with a consequential loss of users).

      • Corroded@leminal.space
        link
        fedilink
        English
        arrow-up
        0
        ·
        edit-2
        6 months ago

        Am I the only one who finds this so weird when we talk about LLMs? If someone makes a bot that resembles some specific person, that person’s rights aren’t really violated, and since they’re all fictional content, it is very hard to break actual laws through its content. At that point we would have to also ban people’s weird fan fiction, no?

        Not arguing about whatever they want or don’t want on their platform, but the legal & alleged moral questions / arguments always weird me out a bit, because there’s no one actually getting hurt in any sort of way by weirdos having weird chats with computers.

        I could see some people making the argument that it could be considered defamatory especially in cases where it is being peddled as real. Politicians might even try to link it in with revenge porn or other non-consensual pornography laws.

        It would sure get messy in a hurry though. Imagine someone trying to make lewd photos of Tomb Raider’s Laura Croft for example and accidentally generates images resembling Alicia Vikander or Angelina Jolie from the Tomb Raider movie.

        • DarkThoughts@fedia.io
          link
          fedilink
          arrow-up
          0
          ·
          6 months ago

          I could see some people making the argument that it could be considered defamatory especially in cases where it is being peddled as real.

          Hard sell overall imo. But in any sort of malicious case we should punish the people behind it, not the software used to make it.

          • 8ender@lemmy.world
            link
            fedilink
            English
            arrow-up
            1
            ·
            6 months ago

            That’s tough though. Do you punish “the artist” or the person who commissioned them? Or both?

            • DarkThoughts@fedia.io
              link
              fedilink
              arrow-up
              1
              ·
              6 months ago

              What? We’re talking about LLM created content, so there’s no artist or person commissioning anything. But if you’re asking for the hypothetical case of someone commissioning blackmail material at an artist (without telling them the purpose), then obviously the person who ends up doing the blackmail. I don’t see the how the artist would’ve made themselves liable unless it was very obvious that it was intended to be used for illegal purposes.

              • 8ender@lemmy.world
                link
                fedilink
                English
                arrow-up
                1
                ·
                6 months ago

                By artist I mean the LLM. Do you punish the LLM (or company running it) for generating it, or the person who asked it to?

                • DarkThoughts@fedia.io
                  link
                  fedilink
                  arrow-up
                  1
                  ·
                  6 months ago

                  So you’re asking me a question that is literally already answered within the comment you were replying to.

          • Corroded@leminal.space
            link
            fedilink
            English
            arrow-up
            0
            ·
            edit-2
            6 months ago

            I feel like it’s going to be a challenge to find a definition of malicious most people agree on.

            Someone might think it’s fine to make nudes of Captain Marvel for example because she’s a character. They don’t really care about the Brie Larson aspect.

            I suppose there is the option to eliminate any kind of name based suggestions.

            • DarkThoughts@fedia.io
              link
              fedilink
              arrow-up
              1
              ·
              6 months ago

              I personally don’t see that much of an issue of people making “nudes” of others since they’re fake anyway. I see an issue when they’re used for things like bullying, blackmail, etc. That is technically already illegal, just not well enforced for any sort of digital topic and hasn’t been for over a couple of decades now. Hence why I find the attention the LLM stuff gets exceptionally hypocritical and overblown, because non of them really cared when someone simply got cyberbullied, or blackmailed through classically edited images - let alone screamed for the outlawing of editing software or social media.