“AI models struggle with tasks like 10-year strategic decisions because the feedback loop for reinforcement learning is too long.”