The advancements made in deep learning and vision have led to improvements in NLP and reinforcement learning applications, eliminating the need for different architectures for each individual problem domain.