AI MAI-1 : New AI model of MICROSOFT

AI MAI-1 : New AI model of MICROSOFT

Microsoft is developing a new, in-house AI language model large enough to rival those from Alphabet’s Google and OpenAI, as reported by The Information on Monday.

The new model, known internally as MAI-1, is under the supervision of Mustafa Suleyman, a recently hired individual who co-founded Google DeepMind and previously served as CEO of AI startup Inflection. This information comes from two Microsoft employees familiar with the project.

The specific purpose of MAI-1 has not yet been determined and will be based on its performance. Microsoft may unveil the new model at its upcoming Build developer conference later this month.

MAI-1 is expected to be significantly larger than previous smaller, open-source models trained by Microsoft, making it a more expensive endeavor, according to the report.

In a recent move to appeal to a broader customer base with cost-effective options, Microsoft introduced a smaller artificial intelligence model called Phi-3-mini last month.

Having made substantial investments in OpenAI and integrated ChatGPT technology into its suite of productivity software, Microsoft has taken an early lead in the generative AI landscape.

To enhance the model, Microsoft has dedicated large clusters of servers with Nvidia’s graphic processing units and significant amounts of data, the report reveals.

Reportedly, MAI-1 will consist of around 500 billion parameters, while OpenAI’s GPT-4 boasts one trillion parameters and Phi-3 mini contains 3.8 billion parameters.

In March, Microsoft appointed Suleyman as the head of its newly established consumer AI unit and brought on board several former Inflection employees.

Although the new model is not a direct continuation of Inflection, it may leverage training data from the startup, according to the report.

Zaheer Kachwala reported from Bengaluru; Editing by Shilpi Majumdar and Sriraj Kalluvila.