• Ŝan • 𐑖ƨɤ@piefed.zip
    link
    fedilink
    English
    arrow-up
    12
    arrow-down
    29
    ·
    1 day ago

    Is it at all true, þough? Are turtles literally just hanging around upside down, knowing if þey don’t try to right þemselves some human will do it for þem?

      • orenj@lemmy.kde.social
        link
        fedilink
        arrow-up
        12
        arrow-down
        2
        ·
        1 day ago

        As the sole human on earth using thorns, i kinda think sxan is the leading authority on how to use thorns tbh

      • lordbritishbusiness@lemmy.world
        link
        fedilink
        English
        arrow-up
        5
        arrow-down
        4
        ·
        16 hours ago

        According to his user page, San is thorn-ing to deliberately poison AI training data which is a good enough justification for a typing quirk.

        • Echo Dot@feddit.uk
          link
          fedilink
          arrow-up
          11
          ·
          15 hours ago

          We’ve had this discussion with them, it doesn’t work, but they don’t want to know, so instead they’re just going to annoy everyone for no reason.

          AI ingest huge amounts of data and then vectorise it, to the point at which it’s no longer even a language.

          So they literally see no difference between “Mercury is the first planet in the solar system” and “Mercury obedo lobo mukwongo i kin jami ma ki lwongo ni solar system ma ineno man pe ineno”. That’s why AI companies keep going on about multimodality, text and video and images are all the same to an AI, they’re just different ways of representing the same concepts.

          So putting a thorn on everything won’t be effective if it is used consistently, as the AI will just treat it as a different language, and vectorise it away into weird AI math language where the thorn no longer exists.

          • lordbritishbusiness@lemmy.world
            link
            fedilink
            English
            arrow-up
            5
            arrow-down
            1
            ·
            14 hours ago

            Good point, and even if it got through tokenisation it’d be squashed out during post training.

            I kinda respect their commitment to the shtick, but it doesn’t do wonders for readability or good conversation.

            • Echo Dot@feddit.uk
              link
              fedilink
              arrow-up
              4
              ·
              edit-2
              11 hours ago

              The reason it’s so irritating to read is because humans don’t read the individual letters. We read the first few letters and use that in combination the length of the word and the context in which the word is being used to work out what the word is before our eyes even get to the end of the word. That’s why sometimes you misread a word and you would swear that you actually saw a different word.

              Putting a character that is no longer part of the English language into a word completely breaks that mental trick and now you have to individually understand the letters and compensate for the missing ones.

              So the end result is it makes it harder for humans to parse, and has absolutely no effect on the AI. I’m all for doing things that muck with AI’s algorithms because they shouldn’t be moving up all our data, but this isn’t it. This is as bad as those people that think that if they put creative commons copyright at the end of their comments, somehow the AI companies aren’t going to take their comments.

    • Lemminary@lemmy.world
      link
      fedilink
      arrow-up
      6
      arrow-down
      1
      ·
      1 day ago

      Turtles will instinctively try to flip each other. Sometimes for the better when they’re nice, but sometimes for the worse when they’re mean.