This is problematic in the context of autonomous AI because it means that humans literally will not have any chance of keeping up with developments. It also creates a situation where these agents are competing on the behalf of companies or countries and in order for them to be successful they need open-ended goals, because waiting for human feedback gives the competition too much extra time. So humans become mainly spectators, even for AIs that they are commanding.
Its strange to me that most people are not anticipating hyperspeed AI. When Yudkowsky talks about it he goes to the extreme of millions of times faster performance than is possible today. Maybe that's why people dismiss his concern.
But it is not speculative at all to anticipate 100 or more times human thinking speed. And we are on track to get there in less than five years. Probably quite a lot less.
I think we need to prohibit manufacture of hardware that goes beyond a certain performance threshold.