Comment on Codeberg: army of AI crawlers are extremely slowing us; AI crawlers learned how to solve the Anubis challenges.

mfed1122@discuss.tchncs.de ⁨15⁩ ⁨hours⁩ ago

• iopq@lemmy.world ⁨54⁩ ⁨minutes⁩ ago

  AI these days reads text from images better than humans can

  source
• lapping6596@lemmy.world ⁨15⁩ ⁨hours⁩ ago

  Accessibility gets throw out the window?

  source

  • mfed1122@discuss.tchncs.de ⁨14⁩ ⁨hours⁩ ago

    I wasn’t being totally serious, but also, I do think that while accessibility concerns come from a good place, there is some practical limitation that must be accepted when building fringe and counter-cultural things. Like, my hidden rebel base can’t have a wheelchair accessible ramp at the entrance, because then my base isn’t hidden anymore. It sucks that some solutions can’t work for everyone, but if we just throw them out because it won’t work for 5% of people, we end up with nothing. I’d rather have a solution that works for 95% of people than no solution at all. I’m not saying that people who use screen readers are second-class citizens. If crawlers were vision-based then I might suggest matching text to background colors so that only screen readers work to understand the site. Because something that works for 5% of people is also better than no solution at all. We need to tolerate having imperfect first attempts and understand that more sophisticated infrastructure comes later.

    But yes my image map idea is pretty much a joke nonetheless

    source
• echodot@feddit.uk ⁨14⁩ ⁨hours⁩ ago

  AI is pretty good at OCR now. I think that would just make it worse for humans while making very little difference to the AI.

  source

  • mfed1122@discuss.tchncs.de ⁨14⁩ ⁨hours⁩ ago

    The crawlers are likely not AI though, but yes OCR could be done effectively without AI anyways. This idea ultimately boils down to the same hope Anubis had of making the processing costs large enough to not be worth it.

    source

    • nymnympseudonym@lemmy.world ⁨12⁩ ⁨hours⁩ ago

      OCR could be done effectively without AI

      OCR has been neural nets even before convolutional networks emerged in the 2010s

      source

      • mfed1122@discuss.tchncs.de ⁨10⁩ ⁨hours⁩ ago

        Yeah you’re right, I was using AI in the colloquial modern sense. My mistake. It actually drives me nuts when people do that. I should have said “without compute-heavy AI”.

        source

        • -> View More Comments
    • interdimensionalmeme@lemmy.ml ⁨13⁩ ⁨hours⁩ ago

      Do you know how trivial it is to screenshot a website and push it through an OCR ?

      This battle is completely unwinnable, just put a full dumb.zip of the public data on the front door and nobody will waste their time with a scrapper.

      Is the data public or is it not ? At this point all that you’re doing anyway is entrench the power of openai, google and facebook while starving any possible alternative.

      Anubis will never work, no version of anubis will ever be anything more than a temporary speed bump.

      source

      • mfed1122@discuss.tchncs.de ⁨12⁩ ⁨hours⁩ ago

        Yeah, I do. I’m just grasping at straws. But you’re right, the only real solution, ironically, is to have non-open sites where you need accounts to view content. I wouldn’t mind seeing some private phpbb forums though.

        source
• NeilBru@lemmy.world ⁨8⁩ ⁨hours⁩ ago

  Computer vision models can read/parse pixel geometry.

  source
• Baleine@jlai.lu ⁨14⁩ ⁨hours⁩ ago

  Humans that don’t see:

  source