Comment by rolux

Will there be a point in that series of ARC-AGI tests where AI can design the next test, or is designing the next text always going to be a problem that can be solved by humans and not AI?