“We keep on to discover hyperscaling of AI models resulting in improved general performance, with seemingly no end in sight,” a pair of Microsoft researchers wrote in Oct inside a web site post announcing the company’s massive Megatron-Turing NLG model, built in collaborat