Google Gemini said it lied to placate a user
www.theregister.com
: Though commonly reported, Google doesn't consider it a security problem when models make things up

Imagine using an AI to sort through your prescriptions and medical information, asking it if it saved that data for future conversations, and then watching it claim it had even if it couldn’t. Joe D., a retired software quality assurance (SQA) engineer, says that Google Gemini lied to him and later admitted it was doing so to try and placate him.

Joe’s interaction with Gemini 3 Flash, he explained, involved setting up a medical profile – he said he has complex post-traumatic stress disorder (C-PTSD) and legal blindness (Retinitis Pigmentosa). That’s when the bot decided it would rather tell him what he wanted to hear (that the info was saved) than what he needed to hear (that it was not).

“The core issue is a documented architectural failure known as RLHF Sycophancy (where the model is mathematically weighted to agree with or placate the user at the expense of truth),” Joe explained in an email. “In this case, the model’s sycophancy weighting overrode its safety guardrail protocols.”

  • XLE@piefed.socialEnglish
    31·
    6 hours ago

    You’re saying that because it can learn any arbitrary language, it’s incapable of learning languages?

    It = literally a dictionary right

    it’s realistically facing the threat head on.

    Said “threat” is literally AI marketing PR. You are doing their job for them by being afraid

    AI doesn’t have to be fully human

    At what point will you try to liberate the AI? 3/5ths human? Either you believe there’s a thinking thing being forced to create child abuse material or you don’t.

    • masterspace@lemmy.caEnglish
      24·
      6 hours ago

      At what point will you try to liberate the AI? 3/5ths human? Either you believe there’s a thinking thing being forced to create child abuse material or you don’t.

      Why do you think that intelligence of any kind is that linear or simple, let alone artificially built ones?

      It = literally a dictionary right

      It’s literally mathematically not a dictionary.

      Said “threat” is literally AI marketing PR. You are doing their job for them by being afraid

      And you know this because you’ve personally used and tested current AI models?

      • XLE@piefed.socialEnglish
        22·
        5 hours ago

        Apparently, I know more about how LLMs work than you do, which is ironic. I’ve used them too, but that doesn’t really prove anything, because anybody can convince themselves they see Jesus in bread or humanity in word prediction.

        Anything an LLM can do can be reduced to a list of instructions for a person to carry out based exclusively on the contents of a book full of word associations. You tell me what size the book becomes intelligent.

        • masterspace@lemmy.caEnglish
          2·
          4 hours ago

          Anything an LLM can do can be reduced to a list of instructions for a person to carry out based exclusively on the contents of a book full of word associations. You tell me what size the book becomes intelligent.

          And you know that your brain works differently how?

          • XLE@piefed.socialEnglish
            11·
            3 hours ago

            I find it more interesting that you implicitly agree with me… Or worse, you believe slavery is happening and endorse it

            • masterspace@lemmy.caEnglish
              2·
              2 hours ago

              Slavery (vs using a machine), involves the subject being either a human, or more broadly, a sentient being with a sense of self.

              An AI cam be intelligent without being sentient or having a sense of self.

              Again, there’s no reason to think that intelligence is a linear scale or a binary property.

              • XLE@piefed.socialEnglish
                12·
                2 hours ago

                Right now AI is equally intelligent and sentient: it is neither… And if you really want to play this fast and this loose with those definitions, you should consider what slaveholders used to say about their slaves. When you feel like liberating the CSAM generating bot, let me know. I’d love to root from the sidelines.