“Ultimately it is my fault for not insisting on a better QA process for this work and pushing our team hard to hit a deadline,” Figma’s CEO said.
    • Kissaki@beehaw.orgEnglish
      5·
      5 months ago

      I’d actually like that if it follows good practicses. I feel like the good ones will continue to be good ones anyway, ignoring unification.

  • perishthethought@lemm.eeEnglish
    10·
    5 months ago

    I’m curious, what exactly is the problem with this? Does Apple have copyrights on the whole design or each individual visual element? Where would figma get in trouble if they left it working that way?

      • perishthethought@lemm.eeEnglish
        3·
        5 months ago

        Yah. ALl of this makes me think it’s crazy for a company like Figma to even try this. If the designs don’t steal from well known brand, people will say they suck. If they do steal, they get booed as well. Losing proposition, it seems to me.

    • jarfil@beehaw.org
      3·
      5 months ago

      Apple has copyrights on the looks of every element, and patents on the way they interact. Aldo trademarks, but I doubt they apply in this case.

      Some patents may have expired (good thing they don’t last as long as copyrights), some they may not bother to defend (litigate) against small users.

      In this case, Figma’s AI seems to repeatedly follow Aple’s design too closely.

    • jarfil@beehaw.org
      1·
      5 months ago

      Some apps have multiple pages, some have a map, some show different timespans of weather predictions with different data, some show photos of places, some allow selecting multiple places but show them differently, some show maritime weather, and so on.

      There is a limited amount of creativity, but a properly “creative” AI, should not keep repeating the same design pattern over and over.

        • jarfil@beehaw.org
          1·
          5 months ago

          I’m guessing it should’ve been trained on more pattens, and increase that temperature a bit.

          To be somewhat creative, it should have a large training set, and an iterative approach to the output. From the looks of it, this one is just a single step LLM, picking design elements one by one.