• JohnDClay@sh.itjust.works
      link
      fedilink
      arrow-up
      9
      ·
      1 year ago

      What’s the data set from? It will at least be skewed towards people most posting to the social media networks scraped for training data. I wouldn’t be surprised if foreign language social media was substantially underrepresented in the data set because the programmers putting it together weren’t as familiar with it.

  • philthi@lemmy.world
    link
    fedilink
    arrow-up
    17
    ·
    1 year ago

    I suppose in reality it’s the average of people who post pictures online and are willing to display it publicly, and probably weighted slightly by how many pictures of a person are uploaded (e.g. if someone has uploaded tens of thousands of pics of themselves - a celebrity or whatever - they will throw the average slightly towards their own face). Assuming the sample set is something like: all of social media photos of a person’s face.

  • Merwyn@sh.itjust.works
    link
    fedilink
    arrow-up
    17
    arrow-down
    1
    ·
    1 year ago

    People in the comment seems to not understand that it doesn’t mean average on the “scale of beauty/attractiveness”. But averaged features. Like if you merge all nose shapes of a million person you get this nose, ect.

    It was tested already several years ago that people tend to like faces made by merging a lot of faces together and “averaging” them. Most of the time rating them more attractive than the individual faces used.

    I don’t find the source anymore… I’ll check better later.

    • Lojcs@lemm.ee
      link
      fedilink
      arrow-up
      8
      arrow-down
      1
      ·
      1 year ago

      People in the comment seems to not understand that it doesn’t mean average on the “scale of beauty/attractiveness”. But averaged features. Like if you merge all nose shapes of a million person you get this nose, ect.

      It more likely means ‘like faces that were called average in the dataset’ rather than an actual average.

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

    What if the ai is lying? What if it’s lazy and like “fuck it, this’ll do to keep the meat bag happy.”

    What if?!

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

      If you use humans to fine tune and judge the quality of output, then in some sense, that’s pretty much all the AI can possibly do.

      Everyone can see the output, “I don’t know.” and mark it zero (dude). But the meat bags will definitely end up rewarding the model if it instead generates some plausible nonsense.