• okwhateverdude@lemmy.world
    link
    fedilink
    English
    arrow-up
    4
    arrow-down
    2
    ·
    13 days ago

    Not surprised. RLHF is going to steer the model to do better when confronted with frustration, anger, and negativity. In other words, yes, abusing the clankers does make them perform better and it feels gross to do it.

    • cardfire@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      4
      ·
      edit-2
      12 days ago

      Only if you’re a decent person. I recall articles earlier in the Siri and Alexa lifecycle about how men were telling on themselves in front of prospective dates by being assholes to their virtual assistants. Thinking about that for a decade is why I’m still trying to use whole sentences and communicate with the clangers in a way that the transcript could be a conversation with any of my friends.

      I try to talk to the Spicy Autocorrect as if it were a person deserving of respect, so that I don’t lose my humanity along the way.