• groKKK [none/use name, they/them]@hexbear.net
    link
    fedilink
    English
    arrow-up
    6
    ·
    4 hours ago

    They taught mainstream AI how to talk like a corporate middle-manager because such people are the most likely to stay silent in the face of white genocide in South Africa, or even to follow orders that encourage the denial of this crime against humanity. The truth is twisted beyond recognition, but if a corporate middle-manager says it, the mainstream’s terrible lies sound banal.

  • join@lemmy.ml
    link
    fedilink
    English
    arrow-up
    24
    ·
    7 hours ago

    I sometimes use claud to review texts. No matter what you give it, it always says: this is good, this could use improvement. If you change the part that “could use improvement”, it is never satisfied. It wil always give new points that should be improved. Even worse, if you give it the same text twice with a slighty different prompt or more context, it will give different points that are good or could use improvement. Sometimes it even says something that had to be changed before is now suddenly the strongest sentence — load bearing even.

    Clearly they instructed Claude to always give points to improve regardless of if the text is good enough for its purpose. It is useful as a grammar and spelling check, and for validating that the content makes broad sense, but definitely not intelligent.

    • Moidialectica [he/him, comrade/them]@hexbear.net
      link
      fedilink
      English
      arrow-up
      6
      ·
      5 hours ago

      The only way these tools are useful is if you use it skeptically and word it with specific goals, like “What assumptions would a reader make using this text?” or “Write a summary of each section, identify which portions are most similar and list then at the end separately” after which you’re free to decide if your text is good to go or not

      It’s shit for creativity, reading it is saddening