A Buttery Pastry posted:

You can actually click the little menu next to a recommended video and say you're not interested, and it cleans up the problem entirely if you hit one or two good ones. Like, I think I did that with a Shapiro video and some racist Swede (not that one), and I haven't gotten a single recommendation along those lines since then. I think the algorithm has basically decided it's an all-or-nothing thing. If you like one racist video you'll watch the rest, if you don't you'll hate them. Which is probably fair, the only exceptions is gonna be when they recommend a racist video from your nation's traditional enemies.

It is emblematic of similar approaches used in other types of algorithms, like ones that define non-whites as inherently more criminal and "scientifically" bake in racist attitudes and creating a feedback loop that reinforced the original ideology. Basically, if they can't even make this work, and don't even test whether some facial recognition software recognizes non-whites in fun stuff, what are the chances they do due diligence on more serious issues where it's even harder to check if it's working correctly - in fact, it appears to be working perfectly, mirroring the entirely "neutral" data you put in.