Earlier quoted context omitted.
Rapidly (and with its help) get it to a point where it can self-improve beyond human intelligence.
That doesn't make any sense. Just because a bot can pass for a human doesn't make it capable of improving itself into a superintelligence. Evidence: 6+ billion humans who are capable of improving themselves but are not a superintelligence yet.
The gulf between the smartest human being ever and the dumbest human being ever is not really that wide. It's just a small blip in the long line of intelligence progression. There is no reason why this point should be the limiting point of AI intelligence. It is very likely that the stable point of AI intelligence is far beyond human intelligence, from just the sheer quality of processing hardware that exists that is better than, or could become better than human hardware.