• Otter@lemmy.ca
    link
    fedilink
    English
    arrow-up
    7
    ·
    4 months ago

    sarcasm is already hard to understand online, even harder for generative AI

    I know sometimes I would take a peek at the person’s comment history to see if they were well informed / a shill for the product. The AI can’t do that

    • sugar_in_your_tea@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      7
      ·
      4 months ago

      Generative AI doesn’t understand anything, it just adds it to it’s model. If more people are being sarcastic than genuine in the data set, that’ll be more represented in the generated text.

      AI could categorize users by competency (i.e. how often they discuss specific topics and agree with some corpus), but I doubt it does that. It’s probably just taking posts at face value.

      • Balder@lemmy.world
        link
        fedilink
        English
        arrow-up
        3
        ·
        4 months ago

        Doing that would require significantly more compute power, so there’s little economic incentive.

      • Alphane Moon@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        4 months ago

        AI could categorize users by competency (i.e. how often they discuss specific topics and agree with some corpus), but I doubt it does that. It’s probably just taking posts at face value.

        This is not being done though right? I haven’t heard anything about content ranking with connections outside of Google seemingly using authors name is articles from large news sources.