That turned out to be only real right until about 2012, when we eventually had sufficient compute (in variety of GPUs specially, many thanks NVIDIA) and more than enough facts (many thanks ImageNet) to basically scale these designs, as was initial shown when Alex Krizhevsky, Ilya Sutskever and Geoff Hinton received the 2012 ImageNet problem (assume: The World Cup of Computer Vision), crushing their competitors (16.4% mistake vs.