Good question. Self-play / self-evolution is usually where these systems either shine or collapse. Curious if you saw convergence or mode collapse when evolving agents on their own generated tasks.