“Jeff Dean's work on neural networks at Google, starting in late 2011, was driven by the goal of scaling up model size using large-scale parallel computation.”