“We carry on to check out hyperscaling of AI models resulting in much better performance, with seemingly no conclusion in sight,” a pair of Microsoft scientists wrote in October inside of a site post asserting the company’s substantial Megatron-Turing NLG model, built in