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.”

  • Ech@lemmy.caEnglish
    41·
    6 hours ago

    Actual AI would be more than “just math”, but LLMs aren’t AI, so the comparison is moot.

    Now we’ve built a collection of simulated neurons, at a scale close to that of the human brain, and trained it on the entirety of the human language

    We are not even close to anything of the sort. We’ve got a probability machine that’s mostly decent at previous collections of human language. The other two are much farther down the road (if they’re even possible) than you or the rest of the tech bros are trying to convince everyone else of.

    • village604@adultswim.fanEnglish
      3·
      6 hours ago

      LLMs are made of neural networks which attempt to mimic the brain. But yeah, they don’t have true intelligence.

      • itsprobablyfine@sh.itjust.worksEnglish
        31·
        5 hours ago

        Neurons are much more sophisticated than transistors. A neuron can have multiple connections and can provide a range of values. Digital logic is all yes/no. I’m not sure we even can build something that mimics a brain with current technology.