Automating ML/AI research seems completely tractable. Most of the other claims seem much less doable.