Apple's latest machine learning research could make creating models for Apple Intelligence faster, by coming up with a technique to almost triple the rate of generating tokens when using Nvidia GPUs.