Can we create a function that provably predicts the optimization power of intelligences? — AI Alignment Forum