For years, it seemed obvious that the best way to scale up artificial intelligence models was to throw more upfront computing resources at them. The theory was that performance improvements are ...