• MissJinx@lemmy.world
    link
    fedilink
    English
    arrow-up
    11
    arrow-down
    1
    ·
    14 hours ago

    If you use ai as a tool in your daily life one tip is to ALWAYS ask “what is wrong with the answer you gave above” 9 out of 10 times it will correct something

    Ps: I know, I know “AI bad can’t use”. I’m just leaving a tip for those that do use it as a tool. e.g. I use sometimes to map frameworks for example, very dumb work.

    • Buddahriffic@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      37 minutes ago

      It depends on the model.

      I’ve had some interactions where you can make it flip flop by repeatedly asking “are you sure about that?”

      Or another time, I was thinking about some math thing, figured it was probably already a theorem if it was true, asked one of the GPT5 models on duck duck go, and then argued with it for longer than I should have when it gave a response that was obviouly wrong. Asked another model, it gave the correct response plus told me the name of the theorem.

      So I asked the first one about that theorem, and yes, it was familiar, but it didnt apply in my case for some bs reason (my specific case was trivially reduced to exactly what the theorem was about). I did eventually get it to admit the truth, but it just wouldn’t let it go for the longest time.

      So it doesn’t hurt to ask, but a) it might be wrong when it corrects something that was right, and b) it might argue it is right when it is wrong.