One simple reason I assume is that maximizing a given objective doesn’t necessarily require a model to duplicate itself. Of course, developing a child model wou

One simple reason I assume is that maximizing a given objective doesn’t necessarily require a model to duplicate itself. Of course, developing a child model would be difficult for an AI for the same alignment issues we face, so that is a real downside/risk an AI would take on by creating non-identical children.