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.
alt Hacker News
It looks like they are recursively gradient descending into their own local minima at increasing code velocity.