Sounds like what China really needs, then, is a new approach to testing that manages to distinguish deep understanding of the underlying concepts from mechanistic/formulaic memorization.

(Tangent: Has anyone yet tried to prompt GPT3 to generate "deep" word problems?)