It was most certainly the other way around. Marcus does not focus on differentiable programming in his paper. See my edit.