Thread by @cute_sayako

here's something chinese mathematicians (just me) come up with fighting censorship and the use of ai thereof: they find a image A that they will definitely censor, such as one affronting xi jinping, and then find some image B that definitely doesn't,

(1/n)

https://twitter.com/cute_sayako/status/1388692276042362880

and they interpolate the color points pixelwise, then equidistantly sample it, and now you post all of them to some chinese social media. here's what's gonna happen: they will either delete some of them, in which case you come to an algorithm to mangle your future images

(2/n)

(the interpolation process is reversible to a human readable form with bruteforce algorithm); or they delete everything and probably train their ai to do so, in which case you just use this method to train their ai to delete every image and thus render their ai useless,

(3/n)

also in an extreme case this will result in a frivolous scenario in which they ban the use of image on all chinese social networks, in which case the general population of china will rebel against censorship. either way, red hat hackers win, chinese communist party loses.

(n/n)

@threadreaderapp unroll, s'il vous plait.