“We carry on to view hyperscaling of AI models bringing about better efficiency, with seemingly no conclusion in sight,” a set of Microsoft scientists wrote in October within a weblog submit saying the company’s significant Megatron-Turing NLG model, built in collaboration