“We carry on to find out hyperscaling of AI models resulting in greater general performance, with seemingly no end in sight,” a pair of Microsoft researchers wrote in Oct in a blog post announcing the company’s massive Megatron-Turing NLG model, built in collaboration with