IQ is, indeed, not a good measure of intelligence. In fact, intelligence isn't a unitary thing, but a bunch of separate capabilities, at least one of which handles organizing and communicating with the other parts.
That said, if we're going to talk informally about intelligence, IQ is a reasonable stand-in. It means something pretty reasonable in the area between 80-120, possibly 75-125. I'll grant that in no area is is a really good definition, but it's easily quantified.
Note that the very concept of an IQ of 1000 doesn't make any sense. So accept it as a figure of speech. Accepting it as a figure of speech, I still think he's wrong, because I believe that for every task there is an optimum level of intelligence. If he's approximately correct, then there will be a very few extremely intelligent AIs, but it sure won't be your sneakers. The claim that it *could* be in my sneakers is interesting, and a bit unbelievable. And I've got large feet. (Well, he didn't claim that the super-AI would be in my sneakers, just that it would have more computing power than I did, which is also a bit unbelievable unless you start doing strange things with word definitions. I could manage definitions that would make that a reasonable claim, but they sure aren't the standard ones.)
OTOH, my projection to a human equivalent AI is still around 2030, which is sooner than he is talking about. But I'm not expecting that thing to be mobile or portable. And when I say "human equivalent" I'm not talking about all characteristics. I'm not talking about motivational structure. I'm not talking about built-in sense organs. I'm not talking about computations/watt. Etc. I'm talking mainly about ability to reason about situations with incomplete data of uncertain reliability...which, admittedly, covers a lot of what we do.