Mouse movements for the most part, if it was a robot it would choose the shortest path to activate the captcha from the cursor’s original location, us humans don’t have the same precision and we choose the paths at random. If its too fast and the system is not sure, it may also ask you to identify street numbers, plaques, cars, etc on random pictures.
Very few websites use java, you’re referring to javascript.
Some sites have said it looks at mouse movements, but more importantly it looks at your browser fingerprint, aligns that to your Google session and cookies, and can use that to determine if you’re likely to be a human or not.
Must surely be better than image recognition. Given that nvidia had chips that can do say 9000 image processing/recognition operations per second or more, and i can do maybe 3.
Street signs in these things are pretty hard for computers to recognise properly. I heard they started using them as a way for google maps/streetview to outsource the identification to meatsacks.
Google has always used captchas to offload work to the users. The “random” characters in previous captchas were actually sequences from books their character recognition algorithm failed to recognize. So they just had humans do the work for them - for free. The new captchas are pretty much the same, except this time they don’t train OCR algorithms but self driving cars.
Google really wants you to stay logged in at all times. The captcha works best when I am logged in. My guess is there is some kind of log in or IP based rate limiting which normal people browsing the web will never hit
Not all characters were actually checked. The captchas would often (alwyays? can’t remember) consist of multiple words and you only had to get one of them right. Google also gives the same characters multiple users. If the answers line up they are probably correct.