I think a lot of this discussion is around human-level AGI. Does it have to be human-level for it to be an AGI? What would a minimally intelligent AGI be like?