Steven Byrnes comments on Garrabrant and Shah on human modeling in AGI