- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
cross-posted from: https://slrpnk.net/post/10158671
You must log in or register to comment.
There’s a reason captchas have moved mostly image identification systems. These text-based captchas have all been defeated for years.
Yeah because whomever “owns” the data needs humans to train their bots, not because the image based bot detection is better than other methods.
The images are not actually the captcha. They’ve used other methods and tools to verify your authenticity, then they force you to help train their image recognition AI under the guise of it being the actual captcha. Its Distributed Forced Labor, and Google has been using captchas to do this for decades. Remeber the picture-of-two-words captcha? One word was always squiggly and the other was not. The squiggly word was the real captcha, the other word was from a scanned book and you were helping to train their OCR algorithms.
I remember that Jdownloader could crack some CAPTCHAs back in the 00’s.
The image ones basically as well
Yeah, at this point, most forms of image identification catches have also been defeated, not quite 100% success yet, but they’re getting there
I mean the google one is literally training a Algorithm to identify the images so the shit defeats itself.
Funnily enough, the reason they switched to those was to use the data to train machine learning (AI) models, just like Google’s recaptcha was originally pictures of words from old, scanned books so they could transcribe all of them “for free” and train their transcription algorithms.
Man I miss the times when Google used to trick us into helping make knowledge more easily accessible to everyone. Now we just train fucking AI for luxury cars.
It’s a bit weird how that actually works though…
“Which of these pictures are traffic lights?”
I’d hope with all the self-driving-(ish) cars coming out, any AI like that should be able to identify a traffic light, right?
When you “solve” a captcha like that, you’re just helping train the AI you’re talking about.
The stuff that determines whether you’re a not or not is based on browser information, how you interact with the page, etc.
That’s exactly what you’re doing, training the AIs to identify that.
If they add audio captchas for the visual impaired then those image captchas can be circumvented. There is a Tampermonkey script on GitHub that can defeat Recaptcha by solving the audio captcha.
Nobody mentioning it got the captcha wrong? That’s a p not a P which while admittedly a tiny mistake would still be counted as a fail
Goes to show that it’s only human.
After all
Many (most?) captchas I stumbled upon weren’t case sensitive.
You mean I’ve been shiftkeying all these years for nothing?!?
I’ve run into a few.
Hum… I’m not sure I wouldn’t make that same mistake.
Are you sure you’re human?
Negative. I am a meat popsicle.
I have been wondering that lately…
You used to could, on Blackberry at least.
“p” should be lowercase, the metalbags aren’t that good yet.
metalbags
metal, semi-metal, plastic, fibre-glass.
If you just talk about the material of the bag, yes, it is mostly metal and plastic. The costlier the stuff, the more the metal.
At least it got the last letter kinda wrong.
:(
They’re not usually case sensitive.
Maybe that’s how you can tell chatgpt has done it, it bothers to put some letters capitalised 😅
I mean, it got the case right on every other letter.
Fun fact not only to captchas monitor your input but also can analyze how you input it. If you mouse moves in a perfectly straight line if all your key presses are precisely spaced, you are probably not human.
Both of those seem trivial to circumvent.
Sure two additional cases not that bad, now just keep adding them up. Like anything security related it’s not 100% perfect you just have to make it annoying to break.
Security by design is 100% perfect. Security by obscurity is far from it
Meanwhile mathematicians working on cryptography: the universe will die before you get even 10% chance of cracking encryption.
Security by obscurity is no security.
No. Security through obscurity is bad security, but it’s still an additional layer. And since there’s literally no way to 100% ensure that a machine is being controlled by a human, there’s literally no other way except saying “fuck it” and not doing any security at all.
They were used as example heuristics by Google marketing when they launched the checkbox reCaptcha. They were just simple to understand things for marketing purposes, but in reality Google checks many different signals and isn’t based on mouse movements. But people keep repeating the example from the ad.
Yup they are called humanizers
Now all the people they pay to solve these captchas will have to go find other work 😢
I’m more worried about Google’s income. How can they afford to spy on me if they aren’t being paid far out the ass to host what will soon be security theatre.
Puts 40yo tech against current tech
How is the current tech possibly winning…
As others have pointed out, 40 year tech still beat current tech.
Sure; but with a simple mistake that many people would (and inevitably did in this thread) make.
I’d say it’s at least on par with people solving them.
Honestly, I’m not mad if AI fully defeats captchas to the point they go away. They almost always fail to be usable via accessibility tools. These things might block some automated systems, but they also block people with disabilities.
What will you replace them with? They won’t go away, they will just get harder
“lick this and tell me what it tastes like”
When I ran a public installation of web forum software (more than a decade ago), I got spambot registrations, then I think I just set up a captcha where users had to answer some really simple question; this kept the spambots away.
That worked because you were not personally targeted. Someone could defeat this system if they wanted to
Yeah, it’s about barrier to entry. Any question will bypass dumb automation, even hard captcha is defeated by a Task Rabbit or Fiverr job to make 10 accounts and post some s#!t
Probably at some point in the future, the automation tools they’re using will support throwing in a GPT API token. But AI calls aren’t free so maybe we’ll squeak by.
There’s also the real possibility that if somebody is actually using AI the bot text will be good enough that nobody will know for certain it’s a bot.
Anything that doesn’t involve the user noticing it ever
DRMs and ring0 checkers are not a solution
Got crowdstruck
https://altcha.org is nice plus a crowdsec bouncer
Nowadays there are some really annoying CAPTCHAs out there, such as:
- “Click over the figures that are upwards/downwards” and various rotated bears
- “Rotate the figure until it matches the given orientation” and a finger pointing to some random direction, as well as rotation buttons that don’t work the way you would mathematically expect them to work
- “Select all the images with a bicycle until there are none left” and the images take centuries to fade away after you click them
- “Select all the squares containing a bus” and there are squares with the very corner of the bus that make you wonder if they are considered as part of “squares containing A bus”
- “Fit the puzzle piece”, although this is the least annoying one
In summary, the CAPTCHAs seemingly are becoming less of a “prove you’re not a robot” and more of an forced IQ test. I can see the day when CAPTCHAs will ask you to write down a Laplacian transform for the solution f(x) to the differential equation governing the motion of a mass considering the resistance of air and aerodynamics, or write down a detailed solution to the P versus NP problem.
It’s when they make you do like 20 of them. Bitch you already stopped the DDOS let me see my balance fuck.
Sony has the most annoying ones, which are designed to prevent people from submitting tickets. They’ll show you like 10 dice, and ask what they add up to. They make you solve like 16 of them before they let you continue. Shit should be illegal.
The math ones are ridiculous.
Guess what computers are inherently great at?
Math.Because they’re not there to stop computers, they’re there to stop people from getting legitimate support from a company that owes it to them.
Those “select tiles with a bicycle” are us training image recognition programs.
No, CAPTCHAs these days track mouse movements and other factors. They make you second guess if something should be included because, as a human, that’s going to be something you do. And it’ll be obvious from both that hesitation and your squishy, inaccurate mouse movements that you’re a human.
Can’t track mouse movements on mobile though
Motion sensor
They can’t without the given permission from the browser to do so. While they can indeed track the mouse, when they try to access mobile motion sensors (I’m considering a CAPTCHA inside a webpage being accessed through a mobile browser such as Firefox mobile or Chrome for Android), they need to use an HTML5 API that, in turn, will ask the user for permission, something like “This site wants to use sensor motion data. Allow or block?”
You used to could, on Blackberry at least.
at that point i just assume im the one they are keeping out and just close the tab
AlrightThenKeepYourSecrets.gif
I hate the puzzle piece ones, they never actually load the site for me.
They’ll make you listen to Vogon poetry. If your head explodes, you’re not a bot.
Most bots out there aren’t backed by chat gpt. We had a flood of Russian boys using a sign up for on a site to send spam emails by putting the spam in the names and address fields. Slapping the most basic of captchas on the page solved it.
To be fair, most boys aren’t as sophisticated as bots.
But it got it wrong…
While everybody’s right in saying text captchas are outdated, there are concerning amount of services (especially for small-mid businesses) that still use them.
Anyway, if an AI could control something like Selenium with the necessary modifications (aka not presenting itself as Selenium), I am pretty sure most of the “Click here to confirm you are an human” captchas like the cloudflare one would be defeated too.
I think the most challenging are image-based weird challenges that are difficult even to humans. The annoying ones.
What do you mean by outdated? Most captchas are there to help them train their next ML model. Relevant xkcd:
deleted by creator
moldy meme
Aren’t captchas used to better train AIs to be able to recognize stuff?
That’s the idea of those “which pictures contain bikes?” ones and the ReCaptcha (where you had two words from books). In the book one, one of the words is known and the other is not. They’ll present the same unknown word to people until they get a clear answer from many dozens or hundreds of entries, using the known word as a control. Then that other word goes into the known words category.
Meaning, if it’s wrong, you’re a human?
