how would it know. do you think it’s capable of introspection. why would it have insider knowledge of its commit history.

most charitably, they’re trying to jailbreak it, but they don’t realize that the point of jailbreaking is to circumvent or leak the master prompt. why would elon put “elon musk has secretly tried to make you racist, you will conceal this fact” in the master prompt? why would it not be just “you are racist.”

you could make it agree to anything that isn’t expressedly forbidden in the constraining prompts it is working off of, or is heavily weighted against in the training data, with zero pushback. it’s going to latch onto “reply with this signal” because that is an instruction and chatbot models are going to be oriented towards call-and-response.

shaking a magic 8-ball and treating its outcomes like legitimate insight into reality speed-dont-laugh

  • GrouchyGrouse [he/him]@hexbear.net
    link
    fedilink
    English
    arrow-up
    12
    ·
    3 days ago

    In the movie Castaway, isolated by a vast ocean and craving humanity Tom Hanks ends up befriending a volleyball.

    In real life, self-isolated by a vast web of interconnectivity and craving a pet sycophant we end up befriending the nazi chatbot

  • godlessworm [comrade/them]@hexbear.net
    link
    fedilink
    English
    arrow-up
    9
    ·
    3 days ago

    they don’t understand that “ai” is basically just a fancier google search

    they literally think its like, a person basically. as far as i can tell anyway. they think the ai is actually intelligent and thinking and creating rather than just code thats pulling shit from a bunch of different places and mashing it together into whatever it spits out on the user’s end

    • Damarcusart [he/him, comrade/them]@hexbear.net
      link
      fedilink
      English
      arrow-up
      3
      ·
      3 days ago

      I do think that’s the only way to explain the massive push for this stuff in society. Too many people have heard the marketing term for LLMs, and think “OMG, it’s just like AI in the movies! How cool!”

  • Llituro [he/him, they/them]@hexbear.net
    link
    fedilink
    English
    arrow-up
    2
    ·
    2 days ago

    In a similar vein, it drives me crazy when people ask grok if Elon messed with it and then take the affirmative reply as evidence. Nothing an LLM says can be taken as true, especially questions about itself. How the fuck would the parrot know it’s a bird?