Semiconductor researcher Vikram Sekar on why inference is the Wild West of chips right now: "Training seems to have one answer. You throw a bunch of GPUs at it, and you train. That's it. Nobody really talks about anything else. You could argue TPUs do it differently. But apart from those, inference has so many recipes and so many ways of doing things that nobody really knows what's going on." "Which topology is going to become the dominant way of doing inference? Is it just ultimately GPUs,…