LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling Paper • 2605.08083 • Published May 8 • 69