“Patrick suggests that based on an observed eight-year cycle of major architectural shifts in deep learning, a new fundamental architecture is overdue and could emerge around 2025.”