-
i hate that there's a widely used dataset for machine learning that is pictures of celebrity faces
-
(why? because we are training foundational models of "what a human face looks like" on faces that are a) photoshopped b) taken in specific contexts and c) of people who generally conform to very normative standards of beauty, d) also there's a ton of age & race & gender stuff)
-
we're fighting the battle to see clearly in the here-and-now, and meanwhile computer scientists are busy building our current problems into the foundations of the future. fuck the CelebA dataset.