This. Then once you've solved alignment, the next most important bit is figuring out how to ensure the solution is universally adopted as training costs decline, and every subvariety of superintelligence with nonzero probability measure is sampled overtime.