“We continue on to see hyperscaling of AI models resulting in greater performance, with seemingly no stop in sight,” a set of Microsoft scientists wrote in Oct in a website put up asserting the company’s massive Megatron-Turing NLG model, built in collaboration with Nvidia