Microsoft is reportedly in the process of training a new, in-house AI language model as it aims to compete with counterparts from Google and OpenAI, according to a report by The Information on Monday.
The project, internally dubbed MAI-1, is being supervised by Mustafa Suleyman, a co-founder of Google DeepMind and former CEO of AI startup, Inflection, as per information from two Microsoft employees familiar with the initiative.
While the exact purpose of MAI-1 is yet to be finalised, its potential applications will depend on its performance. Microsoft might provide a preview of the model during its upcoming Build developer conference later in May.
Reportedly, MAI-1 will be “far larger” than Microsoft’s previous smaller, open-source AI models, implying higher costs. Last month, Microsoft introduced a smaller AI model named Phi-3-mini, aiming to reach a wider client base with cost-effective options.
The tech giant has made major investments in OpenAI and integrated its technology, including ChatGPT, into its suite of productivity software, positioning itself at the forefront of the generative AI landscape.
In its efforts to enhance the model, Microsoft has dedicated significant server clusters equipped with Nvidia’s graphic processing units and vast amounts of data, according to the report.
MAI-1 is expected to comprise approximately 500 billion parameters, compared to OpenAI’s GPT-4 with one trillion parameters and Microsoft’s Phi-3 mini with 3.8 billion parameters.
Suleyman, who joined Microsoft in March to lead its newly formed consumer AI unit, has been instrumental in hiring several former Inflection employees. However, the report clarifies that the new model is not a direct continuation of Inflection’s work, although it may leverage training data from the startup.
