Prime Intellect's Noah Ziems argues teams should use Prime Lab hosted training instead of self-managed GPU infrastructure for reinforcement learning · Digg