Tag
Google describes using long-running agentic loops to recursively evaluate and refine its models, which, while not runaway recursive self-improvement, represents a closely related direction worth monitoring.