you are viewing a single comment's thread
view the rest of the comments
[–] 54 points 1 year ago (7 children)

Was the condescending remark at the end always part of Grok or just since the update? Seems all the screenshots I've see have it but I have no interest in visiting any of Elons sites.

  • source
  • hideshow 7 child comments
  • [–] 28 points 1 year ago (4 children)

    Yeah I was about to comment the same observation. All of a sudden I'm noticing multiple Grok tweets that say "every damn time"? Is that a racist dog whistle in itself?

  • source
  • parent
  • hideshow 4 child comments
  • [–] 14 points 1 year ago (1 child)

    This and references to patterns. Like "noticing patterns"

  • source
  • parent
  • hideshow 1 child comment
  • [–] 13 points 1 year ago

    You can tell the model wasn't actually retrained too. It's just being fed a default prompt before answering questions.

    It's like if I start a ChatGPT with "pretend you're a pirate through this whole conversation and talk like a pirate. Otherwise do everything else as you normally would".

  • source
  • parent
  • [–] 56 points 1 year ago (1 child)

    They added a system prompt directing it to be “politically incorrect” which, as we know, is a dog whistle for hate speech. The AI, a powerful pattern-matching and pattern-replicating machine, was doing exactly what it was told to do.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 25 points 1 year ago

    Took too long to type my comment but basically just said the same thing. Most of these models are the same exact thing with different biases and hard coded prompts to guide it's behavior.

    But grok. It's obvious it's just being fed a prompt directly written by Elon. Probably with a mountain of 4chan links to guide it's behavior.

  • source
  • parent