I think the problem is the formulation "If so, AGI can't be far behind". I think that if a model were advanced enough such that it could do Einstein's job, that's it; that's AGI. Would it be ASI? Not necessarily, but that's another matter.