• credo@lemmy.world
    link
    fedilink
    English
    arrow-up
    14
    ·
    3 months ago

    I see where this is going. FSD 2.0 will use human in the loop, saving money on cameras and LIDAR.

    • Thorny_Insight@lemm.ee
      link
      fedilink
      English
      arrow-up
      1
      ·
      edit-2
      3 months ago

      The current version of FSD Beta is 12 and it’s fully AI powered with no human intervention in the loop. It studies videos of good human drivers and tries to mimic that and currently seems to be doing so with great success. Much better than competition despite not having lidar.

      • CileTheSane@lemmy.ca
        link
        fedilink
        English
        arrow-up
        2
        ·
        3 months ago

        Current AI art studies good human art and tries to mimic that. It’s good with broad ideas but terrible at fine details or understanding specifics. “A painting of five apples” will understand “painting” and “multiple apples” but not the number 5.

        These specific fine details are very important for safe driving.

        • Thorny_Insight@lemm.ee
          link
          fedilink
          English
          arrow-up
          1
          ·
          3 months ago

          Up untill V12, FSD beta was broadly good but sucked with fine details. That is not so much the case anymore. The jump from V11 to V12 was a massive one. It’s still not perfect but it’s really good. So good infact that some people have been questioning wether there’s a Tesla employee remotely driving it.

          • CileTheSane@lemmy.ca
            link
            fedilink
            English
            arrow-up
            2
            ·
            3 months ago

            Musk has been promising FSD “next year” every year for the past decade. I’m not holding my breath that there has suddenly been a miracle.