7 pointsby speckx6 hours ago8 comments
  • f311a5 hours ago
    I think Google stopped using such data long time ago. They have been showing hydrants, buses, crosswalks for ages. It's a solved problem to detect them.
    • cassianoleal5 hours ago
      Pretty sure it's a small set of images of each type as well. It's probably easier for a dumb heuristics-based bot to beat it than it is for most people.
      • dlcarrier2 hours ago
        It's been proven (https://ics.uci.edu/2023/08/10/new-scientist-bots-are-better...). At this point, we should only respond to queries from users who have failed the captcha.
        • tosti2 hours ago
          Right. It says select tiles with cars, bikes or buses and I fail because I recognized one far away. Traffic lights and stairs, if a tile has just a tiny bit of the thing, does that count? I don't think they train on this, they've got their answers and they just want people to feel dumb, I guess.
  • Gualdrapo5 hours ago
    Got second hand anxiety the other day seeing someone logging into their fb account (I guess with a chrome private window?) and they had to solve like 3 round of reCAPTCHAs to finally be able to... get into a fb login page over again, do another 3 or so reCAPTCHAs and finally get into their fb. I don't think they were doing them wrong
  • 30minAdayHN5 hours ago
    Coincidentally today, I have formed a weird theory about reCaptcha. I created a fake facebook account recently. And every time I try to login, I'm getting verify you are human screen that is Google's recaptcha. And I was solving recaptcha for 8-9 times on some occasions. I think the number of times I have to submit in one session went up after couple of days of Facebook usage. My theory is that, Facebook tags many accounts to be fake internally and instead of banning, they use these accounts for data farming including collecting data through Google's recaptcha (but incentive is not aligned as it would help it's competitor unless there is some payback model). This is anecdotal and 99% I'm wrong. But if I'm on the other side with the incentive to use data, it is definitely one I would try to capitalize on.
  • Semaphor4 hours ago
    No, please do keep using them if you use any requiring puzzles (like hCaptcha) at all. It’s the easiest captcha to solve automatically, with many extensions available.
  • Saris5 hours ago
    I don't know about hCaptcha, I tried it out when testing stuff for a websites checkout flow (had card testing spam issues), and they are about the most difficult to solve captchas I've ever done.

    I guess there's a low friction mode only available on the $139/mo plan.

    • endemic5 hours ago
      > they are about the most difficult to solve captchas I've ever done

      Sounds like something a robot would say. Joking aside, I'd rather do a captcha that involves slightly more cognitive effort, rather than get trapped in a reCaptcha loop because I didn't select every last segment of a bicycle.

      • Saris5 hours ago
        True, I've been using Turnstile for now and it's nice because it generally doesn't even need the interactive challenge.
    • gruez5 hours ago
      >and they are about the most difficult to solve captchas I've ever done.

      hCaptcha? You clearly haven't seen the arkose lab captchas.

      • Saris5 hours ago
        I haven't, sounds rough if they're even harder!
  • 0x_rs5 hours ago
    >In Reality, Google's ReCaptcha is a Free LLM¹ Training Solution FOR Google to get Data from Millions of Users Constantly.

    Maybe over a decade ago. As it stands now, it's just a surveillance apparatus.

  • bloody-crow4 hours ago
    It reads like it's written by a person having a stroke in real time. Really weird formatting, punctuation, structure all over the place. Asinine claims.

    I hate the messenger despite loosely agreeing with the message, I guess.

    • nerevarthelame3 hours ago
      The author is German, so I'll cut them some slack on their English prose. And I think I prefer seeing their actual voice rather than pressing it through an LLM filter.

      But I agree the claims are kinda incoherent, and that's unforgivable.

  • ThrowawayTestr5 hours ago
    You could probably make a decent bot resistant captcha by just asking the user to type a slur