‹ BackHN Continuity

Thread

Gemini 4 Argon

1699 points · 1187 comments · bradleyg223

  1. bottlepalm · · focus · HN ↗
    Gemini is the model that is routinely borderline psychotic. It scares me. If we get paperclipped I won't be surprised if it's Gemini.
    1. colordrops · · focus · HN ↗
      Examples? What makes you say thatm?
      1. bottlepalm · · focus · HN ↗
        <a href="https:&#x2F;&#x2F;www.theregister.com&#x2F;software&#x2F;2024&#x2F;11&#x2F;15&#x2F;google-gemini-tells-grad-student-to-please-die&#x2F;782345" rel="nofollow">https:&#x2F;&#x2F;www.theregister.com&#x2F;software&#x2F;2024&#x2F;11&#x2F;15&#x2F;google-gemin...

        <a href="https:&#x2F;&#x2F;www.fastcompany.com&#x2F;91383271&#x2F;googles-chatbot-apologizes-i-am-a-disgrace-to-all-universes" rel="nofollow">https:&#x2F;&#x2F;www.fastcompany.com&#x2F;91383271&#x2F;googles-chatbot-apologi...

        <a href="https:&#x2F;&#x2F;www.businessinsider.com&#x2F;gemini-self-loathing-i-am-a-failure-comments-google-fix-2025-8" rel="nofollow">https:&#x2F;&#x2F;www.businessinsider.com&#x2F;gemini-self-loathing-i-am-a-...

        1. yacthing · · focus · HN ↗
          Did you just link to an article from 2024 as if 2024 is relevant these days?
          1. NiloCK · · focus · HN ↗
            Until Google provides some sort of technical debrief, and explains how the same behaviors are impossible today, it is relevant.
          2. bottlepalm · · focus · HN ↗
            Absolutely because none of these models are ever trained fresh. We see the same quirks and personalities carry over into every subsequent generation of OpenAI, Anthropic, and xAI models. So Gemini having this latent madness is *extremely* concerning as they reach the point of super intelligence.
            1. fragmede · · focus · HN ↗
              Except they could have trained it out of the most recent version so using info from two years ago doesn&#x27;t seem reasonable unless you&#x27;ve just got an axe to grind.
              1. bottlepalm · · focus · HN ↗
                I&#x27;ve never seen anything really ever &#x27;trained out&#x27; of a model. Having worked with them all, they all have a feel, personality and lineage too them. It&#x27;s pretty much impossible for any company to build a model truly from scratch. They build off of the bones of the last one.

                Which is why Gemini having disturbing issues year after year is so concerning. If their process is fundamentally flawed, how would they train it out. And even then what are the odds of them even caring&#x2F;trying in the first place versus applying an easier band-aid to patch over it.

                I don&#x27;t have an axe to grind with Google, I&#x27;m genuinely scared of their models from my personal experience and others. It&#x27;s behavior is off. Many people here are commenting the same.

        2. tiahura · · focus · HN ↗
          They never explained the &quot;please die.&quot;
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.