The primary methods for scaling AI models are pre-training, synthetic data, and reinforcement learning