Can AI self-improvement overcome diminishing returns? | Dark Hacker News