Can we please separate the actual breakthroughs from the obvious incremental updates? This "" new model is just a tweaked hyperparameter and a bigger GPU, let's not pretend it's the second coming of deep learning.