• Kultronx
    link
    fedilink
    arrow-up
    12
    ·
    1 month ago

    Sort of related, I remember someone posted a prompt to use with an LLM essentially ordering it use dialectical materialism and to avoid capitalist neutrality among other things. Deepseek, Qwen, Z.Ai and others all agreed to the instructions. Kimi was the only one to push back saying it wouldn’t agree to the instructions, which I found kind of interesting.

    • ☆ Yσɠƚԋσʂ ☆OP
      link
      fedilink
      arrow-up
      8
      ·
      1 month ago

      It mostly depends on what base models they used to train them, and whether they bothered decensoring the model.

    • loathsome dongeaterA
      link
      fedilink
      English
      arrow-up
      4
      ·
      1 month ago

      I had a really bizarre conversation with Kimi. It’s stupid but I will share anyway. I asked it to come up with a rubric like the DENNIS system from It’s Always Sunny, but the acronym should be BJP and the end goal should concealing embarrassing socioeconomic statistics like child malnutrition. It said that coming up a six-step plan for something like this is inappropriate and it won’t comply. I was baffled and ask what exactly is so bad about it and had a little bit of back and forth. Then I noticed a mistake it made and corrected it since the plan should be three-step and not six-step. For some reason it instantly created the acronym. Of course it was terrible and unfunny. But that was weird behaviour.

      • Kultronx
        link
        fedilink
        arrow-up
        2
        ·
        1 month ago

        that’s funny, i don’t really get why it’s so difficult