When OpenAI trained GPT-3, they didn't roll their own orchestration layer. They used Ray, an open source Python framework born out of the same Berkeley research lab lineage that gave us Apache Spark. And here's the twist: Ray was originally built for reinforcement learning research, then quietly faded as RL hit a wall. Until ChatGPT showed up. Suddenly reinforcement learning was back, as the…