Everyone assumes the next generations of chips that run AI are going to be horrifically complex and require superhuman levels of talent to optimize. What if th

Everyone assumes the next generations of chips that run AI are going to be horrifically complex and require superhuman levels of talent to optimize. What if they're wrong, and there's a much, much simpler architecture that could be found, that could scale almost infinitely, and make better use of Dennard scaling?