ChatGPT suddenly starts sending alarming messages to users
www.independent.co.uk
AI system has started speaking nonsense, talking Spanglish without prompting, and worrying users by suggesting it is in the room with them

ChatGPT has meltdown and starts sending alarming messages to users::AI system has started speaking nonsense, talking Spanglish without prompting, and worrying users by suggesting it is in the room with them

  • stevedidwhat_infosec@infosec.pubEnglish
    1020·
    4 months ago

    Here’s an idea let’s all panic and make wild ass assumptions with 0 data lmao.

    This article doesn’t even list a claim of what their settings were nor try to recreate anything.

    Whole fucking article is a he said she said bullshit.

    If I set the top_p setting to 0.2 I too can make the model say wild psychotic shit.

    If I set the temp to a high setting I too can make the model seem delusional but still understandable.

    With a system level prompt I too can make the model act and speak however I want (for the most part)

    More bullshit articles designed to keep regular people away from newly formed power. Not gonna let these people try and scare y’all away. Stay curious.

    • Buffalox@lemmy.worldEnglish
      81·
      4 months ago

      Here’s an idea let’s all panic and make wild ass assumptions with 0 data lmao.

      Where did that come from?

        • stevedidwhat_infosec@infosec.pubEnglish
          25·
          4 months ago

          Literally the top comment for me (and maybe not you depending on which instance you’re registered with, because some instances block another) says that this is because they’re training their modes off user input lmfao.

          But go off with your douchey assumptions.

      • stevedidwhat_infosec@infosec.pubEnglish
        46·
        4 months ago

        Bare in mind depending on your instance, you won’t see the same comments as others do.

        With that said, top comment here for me is talking about how this was because they’re training their models on user input.

        As if the leaders in fucking AI development don’t know what they’re doing, especially for a concept that’s covered in every intro level AI course in college. 🙄

        Then again not everyone went to college I guess and would rather make arm chair assumptions and pray at the alter of google despite complaining about how AI is ruining everything and google being one of the first people to do shit like this with their search engine for “better results” (not directed at you of course, thanks for being respectful and just asking a simple question rather than making assumptions)

    • crazyCat@sh.itjust.worksEnglish
      5·
      4 months ago

      I mean, OpenAI themselves acknowledged there was an issue and said they were working on it,

      “We are investigating reports of unexpected responses from ChatGPT,” an update read, before another soon after announced that the “issue has been identified”. “We’re continuing to monitor the situation,” the latest update read.