AI bots are so good at mimicking the human brain and vision that CAPTCHAs are useless - The bots’ accuracy is up to 15% higher than that of humans

L4sBot@lemmy.worldmod to Technology@lemmy.world – 181 points –
AI bots are making CAPTCHA tech useless
qz.com

AI bots are so good at mimicking the human brain and vision that CAPTCHAs are useless - The bots’ accuracy is up to 15% higher than that of humans::The bots’ accuracy is up to 15% higher than that of humans

27

Honestly sick of captchas. Think because I'm using Firefox with adblockers. I get them anytime I try to visit a site. It's basically making using the Internet a slog.

I'd start creating bots to do them for me. Fucking life

Are you sing Tor or a VPN? Sharing the same IP with thousands of other people is something that would lead to getting captchas every time you visit a site. Most sites use Cloudflare or other CDNs, and they see the same IP making tons of requests every second, so they flag it as a potential bot IP and issue the captcha challenge.

Vpn. Proton. Yeah that makes sense. It asks me on certain sites. Cloud flare stuff

Kinda ironic. I use CloudFlare’s own VPN (WARP/1.1.1.1) and I still get CAPTCHAs, arguably more than when I used proton.

Weird, I use the same and rarely see them

1 more...

AI bots are not good at mimicking the human brain and vision, they're just good at object recognition

Which is why they're so good at captchas. Letters and numbers are the objects they're trained to recognize here.

Wasn't that the entire point of the captcha? It was never intended as a permanent solution. They help train AIs to identify objects and patterns.

That was definitely not the original point of the captcha. It used to be every site implementing their own version, some better than others, but all were just solving something that should be easy for humans and almost impossible for bots. But then Google got the idea of using it for crowdsourcing to help with digitizing books and they launched reCAPTCHA. This was easy to embed on any web page and served a noble purpose, so it quickly became the de facto captcha over much of the web. Fast forward several years and now we're all providing free work in training AI object recognition, a problem AI is already very good at, which means the images we're being given are increasingly more difficult for humans to interpret.

Early forms of captcha were helping to transcribe books.

Those image captchas are useless. “Prove you’re a human by identifying bicycles exactly the same as our AI identified them.” Of course an AI is better than a human at identifying them the same way as an AI.

It's silly, because you can just cycle between captchas until you find one you can actually do

I'm sick of trying to guess random things about American culture.

"Click in all the school bus" it says. How? If it's blue is it a school bus, or is it only the yellow ones that are the school buses, I don't know.

Nice to know that I'm not the only one that finds that annoying.

So can an AI tell me if a traffic light pole is part of the traffic light or not. Because I never really was sure.

Completely unsurprising given that the millions of captchas people have been doing for the past like 15 years have been being used to train the AIs...

Recently I've seen captchas with AI-generated images. Like "click pictures with a panda in a flower pot".

If the image is generated by AI, it must be easier for AI to identify right?

H̶͉͙̔̂o̶̱͗͝w̶͍̒̍ ̴̳͚͗̄d̵͚̀o̵̧̿ ̵̳̜̔y̷̰̅o̴̗͛͜ù̸̢̘̿ ̶̟̎̐v̴̠͊e̵͕̅̒r̷̝̓̄i̴̮̕f̷̤͉̉ŷ̷̡̐ ̶̗͚̿a̴̡̫̔ ̵̩̉͝p̶̺͝è̶̼͍̈r̴̜͗̚š̴̡͇o̷̘̜͆n̶̫͛ ̶̢̒͘i̶̢̖͆s̸͖̕ ̴͓̇͜r̵̬̞̒̂ẹ̷̡̐̑a̶͓̗͋l̵̹̓̿,̵̫͛͝ ̶̞̝̍̃w̸̞̙̄̕h̵̡̖̄͛a̵̖͖͆̈́t̴̲̐̎ ̷̳̀į̶̳̍s̶̝͈̓̌ ̵̘͍̓̎r̸͋͒ͅe̴͕̰͑͋a̷͍̮͊l̵͕̄