Will a 100 trillion parameter deep learning model be trained before 2026?