In 5 to 10 years we will have likely moved on to the next big model architecture just like it was all about convolutional networks 5 to 10 years ago despite the pivotal paper being published in 2017.