“Andrei Karpathy holds the view that humans do not primarily use reinforcement learning for high-level intelligence tasks like problem-solving.”