[dealuromanet]

I wanted to make a Twitter / X account but the sign-up flow requires you to solve 10 of those puzzles where you have to rotate the animal with the matching symbol to be oriented in the direction that the hand with that symbol is pointing in. After putting up with this next-gen captcha to train X's algorithm for free, it asked me to do another 10 because supposedly it couldn't tell if I'm a human or not. So I decided not to bother with making an account since it's clearly trying to monetize by obtaining free training data for its ML. I also don't trust that they're not attempting to associate performance on these types of tasks with someone's identity to infer things about them and sell that data further.

Is this what it now takes for a sign-up to be credibly done by a human? Or is it just a cheap attempt to obtain free training sets from people who aren't being paid to label data?

[tzs]

I could understand this if it were about Google's old CAPTCHA that used scans from books and could use the user input to improve an OCR system, or Cloudflare's CAPTCHA that asks you to click the pictures that contained traffic lights or bikes or whatever where the user input can be used to improve an image classification system.

Better OCR and better image classifies are things you can make money with.

It is not clear to me how one makes money from a "what way is this animal pointing?" classifier (especially since building that CAPTCHA requires them to have a photo of the animal where they already know it is pointing close to 0, 60, 90, 150, 210, or 270 degrees).

This seems to be a task that likely only be of interest as part of a CAPTCHA and so the most that your solving it for will do is improve the CAPTCHA. It's not like OCR or image recognition where you might be improving some separate money making product of theirs.