It’s almost like they are making fun of the concept.
Meh, the one I saw was pretty easy–just multicolored text on top of colored backgrounds. Far worse is when they distort the text itself. And don’t bother to tell you how to differentiate between 0, o, and O. Some also mess up 1, I, and l, or even 2 an Z, 5 and S. Gah. Whoever designs them needs to remember that you have to balance making them too difficult for computers and easy enough for humans.
Eventually, we’re going to have to come up with alternatives, anyways. I think natural language tests might be useful. A different direction is having some sort of universal login for sites that aren’t security risks. (You’d still have a separate login for your bank account, for instance.) Those could use longer, more easily accurate Turing tests, as each person would only need to do them once in a long while. OF course, that’s assuming people use strong passwords…
You are lucky that you didn’t get this:
or this
Can anyone solve those?
I had to try 4 or 5 times right here on SDMB today to get one that I could read. They weren’t as bad as some others I’ve seen, but still, the swirly backgrounds are getting out of hand. I can’t tell which is which sometimes when the letters are distorted badly enough. All of us with color and depth perception issues need to stage a revolt.
Serious question: Why do people make these things so difficult? Sure, most of 'em are easy enough, but why do they need to be anythinbg other than completely legible?
MOreover, why do they need to be graphics at all? Why can’t they just be a random sum like what is 6x3 or something? Or why can’t they be a multiple choice like "Click the odd one out: Table, Chair, Couch, Giraffe’? As long as they were randomly generated, surely a bot couldn’t solve them, right?
Well the problem with the second one is that you have a 1/4 chance of randomly choosing the correct one. So with a bot’s infinite attempts it would choose the correct one…an infinite number of times (fun with numbers!)
The first one is CDNSS2. The second one is an abomination.
Interestingly enough, I have on occasion typed in the wrong letters/numbers and it still let me in on some websites.
The Facebook ones are like that. As long as you’re close, it gives you a pass. They’re also much, much easier to read.
Because it’s an arms race; as the bot AI / pattern recognition improves, the CAPTCHAs have to become more and more obfuscated to prevent the bots from deciphering them. Unfortunately, human pattern recognition ability is constant, so we are beginning to get left behind.
How is it that the bots manage to decipher them? Do they run some kind of colour or pixel analysis or do they bruteforce it? And either way, isn’t it an awful lot of effort to go through just to get a free Youtube account so you can post some spam that everyone’s gonna just ignore or report anyway?
Ah, yep. Fair enough. How about if the question said “Type the word ‘tomato’ in the box below” or something like that? You’d have to manually enter the entire word into the box, and the word would be different each time. Could the bots get around that?
Do the bots even know they’re on YouTube specifically? Do they know the difference between spamming ann active forum or blog where their crap will be gone within the hour, and some long-abandoned and forgotten-about guestbook where it will remain?
Turing tests suck.
Voight-Kampff tests, on the other hand, rock.
BMPEGa
Good God, Sir, this is a family board, there are innocent children around! I mean, I’m no prude, Heaven knows! Goat felching, the Venus Butterfly, Ron Jeremy’s selective elephantiasis, that’s all very well, but BMPEGa! There are limits, sir, limits!
I love reCaptcha. That one seems forgiving enough that if you’re in the ballpark it’ll count it; also it’s for a reasonably good cause.
Awhile ago I read about an intriguing idea for a captcha. They’d show you 10 pictures and ask you to click on the ones that were dogs. The pictures would be pulled at random from Petfinder’s website, which has literally hundreds of thousands (if not millions by now) of pictures of dogs, cats, and other animals, taken from all different angles and distances. The pictures would be linked in the captcha’s database to their species, but the picture file wouldn’t contain that information. It’d be an extremely hard task for an AI to accomplish, but relatively easy for a human. I dunno why that technique hasn’t appeared.
It would be very hard to make that accessible to blind users.
Audio files of animal sounds would work.
'cause letters-on-swirlygig is ? :dubious:
That’s a C, not a G.
Isn’t it? 