“Prime Intellect runs a "reinforcement learning residency" program with 14 to 16 participants who build complex environments for verticals like verifiable software engineering, medical physics, and cybersecurity.”