Sorry if these are dumb or obvious and answered questions in the field, but could a MoE’s experts be internally optimized over time by the AI, itself? And to minimize ethical risks, could there be an “ethical” division of labor within the system?