• ✨Abigail Watson✨@lemmy.blahaj.zone
    link
    fedilink
    arrow-up
    15
    ·
    1 year ago

    this hyper-attractive girl

    I swear, it’s actually really difficult to make ai women that AREN’T yassified. Is all the training data for what “real” women look like from Instagram?

    • scarabic@lemmy.world
      link
      fedilink
      arrow-up
      5
      ·
      edit-2
      1 year ago

      Decades before the internet we knew that popular media does not depict actual women and tends to advance distorted ideals of beauty, often converging toward just one.

    • cley_faye@lemmy.world
      link
      fedilink
      arrow-up
      8
      ·
      1 year ago

      No, just internet in general. There are a lot of sites/services that put the stereotypical “hot” women in the light and don’t even consider anything else. The same way text content is driven by the most vocal people, video are cut/tailored towards hyperactive squirrels, etc.

      AI just averages all this out and asks you “do you want it tipped a bit this way or that way”. (I know it’s a gross overgeneralization, but it still stands)

      Curating the internet as a “sane, balanced” data source raise two main issues : what does “sane / balanced” mean, and how much human power is needed to seep through all the garbage. This is the main reason I’m not too worried about “AI”. We will bend the thing to be a useful tool for a lot of usecases by using local training, but unless these two questions gets an appreciable upgrade, having one big “ubiquitous” AI model that will answer every need is not on the horizon.

    • birdcat@lemmy.mlOP
      link
      fedilink
      arrow-up
      2
      ·
      edit-2
      1 year ago

      Lol totally! I started with subtle wrinkles around here eyes and gradually tuned it up to unattractive, old, ugly, sad, which is what you see in the pic I posted 🤣