4 ms·
> It was a huge deal internally when "Custom" was added as a gender option in the sign-up flow, and even then, there's a nudge towards the binary to retain bett
by feoren 4y ago
> It was a huge deal internally when "Custom" was added as a gender option in the sign-up flow, and even then, there's a nudge towards the binary to retain better ad targeting.
It's odd that advertisers would dislike that change for targeting -- surely the people who pick "custom" are sending a major signal that correlates with other preferences. It's a third grouping they can specifically target! If they split "male" into "HYPER-MASCULINE ALPHA MALE: CARS, GUNS, AND FOOTBALL FUCK YEAH!" and "I have a penis but I try not to let it make decisions for me", advertisers should presumably love that distinction too. Of course the reality is that their decades of research deciding what constitutes "male" and "female" has gotten them stuck in a shitty local minimum and they can no longer abandon this binary concept that they cling desperately to without admitting that they've never actually known what the fuck they're doing.
- zeroonetwothree 4y agoIf there was a 300 dimensional vector expressing your nuanced gender identity that would be even better, but advertisers would have no idea how to target it (at least at first).
- csa 4y ago> advertisers would have no idea how to target it (at least at first). It would take very little time to reap massive efficiencies in targeting, and the rest would be long-tail that would be discovered as budget allowed.
- anankaie 4y agoThe easiest way to do this would be to train a bandit model to be a bidder, using this vector as part of the input features. You would spend some time and money doing randomized exploration (maybe something more sophisticated with active learning, but just uniform random could work), then train a basic bidder to optimize whatever you are currently interested in. The process of manually deciding which tags fit which ads is not long for this world.