CEV is not about getting humans to be their best, it is about having a transhuman AI turn out for the best. Humans don’t have to be anywhere near perfect, they just need to be sufficiently representative of human nature that the automated extrapolation process can infer the appropriate optimization criteria from them. If you think the computational part of CEV requires, or is assisted by, human attempts at self-improvement (whether that is purification of the will or just improvement of self-knowledge), then that is not CEV as I understand it.
CEV is not about getting humans to be their best, it is about having a transhuman AI turn out for the best. Humans don’t have to be anywhere near perfect, they just need to be sufficiently representative of human nature that the automated extrapolation process can infer the appropriate optimization criteria from them. If you think the computational part of CEV requires, or is assisted by, human attempts at self-improvement (whether that is purification of the will or just improvement of self-knowledge), then that is not CEV as I understand it.