It's different to our ideas of being good at maths. On the one hand, as you mention it can sometimes resolve large individual calculations, but I have seen it be wrong for very simple world problems or even explicit equations with only a couple operations which I myself can immediately catch. It's not a fundamental flaw: I expect it to get better at this as time goes on. But its other of those fascinating little quirks it has that are so noteworthy only because of how competent it appears to be in other areas