Two-Phase Pretraining: Unlocking LLM Scalability & Precision (NVIDIA, Stanford) | code_your_own_AI | Podwise