But in her order, U.S. District Court Judge Anne Conway said the company’s “large language models” — an artificial intelligence system designed to understand human language — are not speech.

  • Opinionhaver@feddit.uk
    link
    fedilink
    English
    arrow-up
    27
    arrow-down
    8
    ·
    6 days ago

    I get that hating on anything AI-related is trendy these days - and I especially understand the pain of a grieving mother. However, interpreting this as a chatbot encouraging someone to kill themselves is extremely dishonest when you actually look at the logs of what was said.

    You can’t simultaneously argue that LLMs lack genuine understanding, empathy, and moral reasoning - and therefore shouldn’t be trusted - while also saying they should have understood that “coming home” was a reference to suicide. That’s holding it to a human-level standard of emotional awareness and contextual understanding while denying it the cognitive capacities that such standards assume.

    “I promise I will come home to you. I love you so much, Dany,” Sewell Setzer III wrote to Daenerys, the Character AI chatbot named after Game of Thrones.

    The bot replied that it loved the teenager too: “Please come home to me as soon as possible, my love.”

    “What if I told you I could come home right now?” Sewell wrote, to which Daenerys responded: “Please do, my sweet king.”

    It was the last exchange Sewell ever had. He took his own life seconds later…

    Source

    • DancingBear@midwest.social
      link
      fedilink
      arrow-up
      8
      ·
      5 days ago

      I would not have understood that to have anything to do with suicide… do they use the phrase coming home to mean death or suicide in the game of thrones show?

      • Seefoo@lemmy.world
        link
        fedilink
        arrow-up
        7
        ·
        5 days ago

        You have to read the other chat logs. Arstechnica has a good summary I think, the link between “coming home” and suicide is specific to the kids chats with these AI.

        • Buddahriffic@lemmy.world
          link
          fedilink
          arrow-up
          3
          arrow-down
          1
          ·
          4 days ago

          Iirc when he did make it more explicit, the AI responded with “no, don’t do that” kind of responses. He just kept the metaphor up when the AI didn’t have such an association in its training data and just responded as a lover would respond to their love saying they’d come home in their training data.

          Though I’d say that if a kid would shoot themself in response to a chatbot saying anything to them, the issue is more about them having any access to a gun than anything about the chatbot itself. Unless maybe if the chatbot is volunteering weaknesses common in gun safes, though even then I’d say more fault lies with the parent choosing a shitty safe and raising a kid that would kill themself on the advice of their chatbot girlfriend.

    • Natanael@infosec.pub
      link
      fedilink
      arrow-up
      3
      ·
      4 days ago

      All you need to argue is that its operators have responsibility for its actions and should filter / moderate out the worst.

      • Opinionhaver@feddit.uk
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        4 days ago

        That still assumes level of understanding that these models don’t have. How could you have prevented this one when suicide was never explicitly mentioned?

        • Natanael@infosec.pub
          link
          fedilink
          arrow-up
          1
          ·
          3 days ago

          You can have multiple layers of detection mechanisms, not just within the LLM the user is talking to

            • Natanael@infosec.pub
              link
              fedilink
              arrow-up
              1
              ·
              edit-2
              3 days ago

              I’m told sentiment analysis with LLM is a whole thing, but maybe this clever new technology doesn’t do what it’s promised to do? 🤔

              Tldr make it discourage unhealthy use, or else at least be honest in marketing and tell people this tech is a crapshot which probably is lying to you

    • fluxion@lemmy.world
      link
      fedilink
      English
      arrow-up
      6
      arrow-down
      10
      ·
      5 days ago

      That AI knew exactly what it was doing and it’s about time these AIs started facing real prison time instead of constantly getting a pass

      • Liz@midwest.social
        link
        fedilink
        English
        arrow-up
        4
        ·
        4 days ago

        Unironically, the provider of the chat bot should be liable for anything the chat bot says. Don’t fire humans so youn can hide behind a neural network.

        But this ain’t suicide encouragement.

      • Mubelotix@jlai.lu
        link
        fedilink
        arrow-up
        5
        ·
        5 days ago

        LLMs are tools. They are not sentient. You must not use a tool if you can’t handle it

      • piecat@lemmy.world
        link
        fedilink
        arrow-up
        2
        ·
        5 days ago

        You could think of LLMs as a glorified ‘magic 8 ball’, since that’s about as much ‘understanding’ it has.