That's terrible...which ones seem to be the best at solving these captchas. Is there a link to a hugging face model with 13 billion parameters or less? Or maybe a dataset used in training? I need to know what to avoid when Google asks me to train their AIs.

Some of these are impossible to solve though. I still dont know if the correct answer to ‘is this a traffic light’ includes the pole the lights hang off or just the lights part and sometimes like 1 pixel is in one of the squares and Im not sure if that counts as if you should click the square or not. I never get like a ‘you got it right’ confirmation after so I have never learned the correct answer. Unclear how the machine learned it when I cant and its supposed to be learning from me.

Well so much for that line of defense.

Soon we’ll start seeing Bot-tests that require you to pick out a picture with the correct number of fingers.