This sounds both exciting and worrying, this is definitely one of the tipping points of the singularity, when machines can improve themselves it happens faster, without AGI regulations to require limitations on its abilities it may be capable of more then we think of are ready for. Give the order to make humanity happy, but some people need time not being happy, and you got AI trying it’s best to solve unsolvable problems. Granted everything truly dangerous can be stopped with just a few commands built in, law of robotic equivalents, but still. This is potentially a revolution in ai here.