Surely their goal is ever-faster recursive self-improvement.
It looks like they are recursively gradient descending into their own local minima at increasing code velocity.