• NeatNit@discuss.tchncs.de
    link
    fedilink
    arrow-up
    144
    ·
    2 days ago

    Love this webcomic. Original source: https://www.peppercarrot.com/en/miniFantasyTheater/014.html

    It’s an open-source comic. The Krita project with all the original layers is free to download, and the SVG and text for the speech bubbles is hosted on git so it can be translated to many other languages. I’ve never seen a webcomic do this before, it’s really neat. And of course, the artwork is just beautiful!

    It’s an awesome comic.

    • NeatNit@discuss.tchncs.de
      link
      fedilink
      arrow-up
      63
      arrow-down
      1
      ·
      2 days ago

      Well, sort of. They’re not for secutiry, that’s for sure. They were originally about making it harder for automated bot requests to go through and overload the server. ReCAPTCHA then started turning it around to make OCR better using machine learning, which is commonly agreed to be a Good Thing since it helped digitize old books and things like that. But of course, this in turn made it possible for bots to get past the CAPTCHA, and everything spiraled from there.

      At some point everyone kind of forgot the real point of a CAPTCHA, and it’s now much more of a free training data generator and much less of an obstacle for bots. But it still can prevent complete rookies from making thousands of requests per second with a simple python script, so it does serve a little bit of that original purpose.

      • ulterno@programming.dev
        link
        fedilink
        English
        arrow-up
        25
        arrow-down
        1
        ·
        2 days ago

        commonly agreed to be a Good Thing

        So, did they open access their trained model weights?

        it still can prevent complete rookies from making thousands of requests per second with a simple python script

        So now you can know that if you are getting DOS’ed, it is actually malicious.

        • edinbruh@feddit.it
          link
          fedilink
          English
          arrow-up
          5
          ·
          11 hours ago

          did they open access their model weights?

          In that instance it wasn’t really training, it was crowdsourcing the transcription. Rechapta would pull out a word from their book archive that the OCR failed to recognise, and if many people identified it as the same word, it would be archived. Now that rechapta has been purchased by Google, the archive and the transcriptions are available on Google books.

          They stopped doing this once ai became more effective than rechapta for book transcriptions.

          Modern chapta actually is about training models. But old, classic rechapta was really just about book transcriptions, and those are available.

    • WanderingThoughts@europe.pub
      link
      fedilink
      arrow-up
      15
      ·
      2 days ago

      The last one is popping up as age verification. And it’s done with a camera now because otherwise anyone just sends in an AI generated picture of a face, and people don’t trust companies with government id.

      • Buddahriffic@lemmy.world
        link
        fedilink
        arrow-up
        1
        ·
        3 hours ago

        I’d say countdown to programs that pretend to be webcams and display an AI video of the requested action has started but I bet at least someone has already done it. And then the arms race between actions to be requested and what AI can do will start until eventually passing the test will be a fail because the actions requested are either too difficult for humans to understand or too difficult for humans to perform, at which point AIs will be trained on knowing the physical limitations of humans.

        This will come in handy for when they get tired of our shit.

        • T156@lemmy.world
          link
          fedilink
          English
          arrow-up
          2
          ·
          3 hours ago

          They’ve existed for quite a long time at this point.

          That’s how virtual puppetry/V-Tubing works. The camera tracks your face, and then moves part of a corresponding model, and unlike face posing inside of Garry’s Mod, or something like that, since it’s bound to a real face, it would move more or less like a human face.

          eventually passing the test will be a fail because the actions requested are either too difficult for humans to understand or too difficult for humans to perform, at which point AIs will be trained on knowing the physical limitations of humans.

          This also exists for some forms of captcha, which track how you complete a puzzle, or something along those lines. A bot would either be completely stumped, complete it far more quickly than a human would, or do it by snapping their cursor to the relevant parts, instead of moving it.

          • Buddahriffic@lemmy.world
            link
            fedilink
            arrow-up
            1
            ·
            12 minutes ago

            On yeah, the little mouse puzzles. I always figured it wouldn’t be that hard to give cursor movement a more natural curve, just give it an interpolation that clamps the first 3 derivatives of position and adds jitter and a little overshoot and correction or clamps the derivatives even harder at the end to mimic slowing down for precision.