no one's talking about the actual elephant in the room: deep learning has been stagnating for 5 years and we're still relying on 2016-era architectures to get the job done