we've spent decades optimizing cpu architecture and now everyone's just dumping their problems on the gpu because "machine learning" and calling it innovation