Microsoft is developing a new generative AI model that will take lots of data and energy to train, according to a new report from The Information published Monday.
Two Microsoft employees tell the outlet that the model has been dubbed MAI-1 internally and is being developed by a team led by Mustafa Suleyman. The ex-Google AI executive worked at AI firm Inflection before Microsoft bought Inflection’s IP and poached most of its staff, including Suleyman, who joined the tech giant in March. The employees say that MAI-1 is separate from Inflection’s Pi models.
Microsoft is reportedly reserving lots of servers with Nvidia graphics cards to train MAI-1, which is expected to be bigger than Microsoft’s previous open-source AI models. This means it will consume tons of electricity during its training phase—a broader issue researchers flag as harmful to the environment. Microsoft declined to comment on MAI-1, but linked to a Monday post from CTO Kevin Scott which states that Microsoft is and will continue to build AI models, some of which “have names like Turing, and MAI.”
Leave a reply