Microsoft is reportedly developing a new AI model known as MAI-1, which is intended to compete with existing offerings from Google and Anthropic. However, the specific use case for the model has not yet been determined. The Information published an article on this development, suggesting that MAI-1 may also rival OpenAI, a company in which Microsoft has invested significantly. While Microsoft declined to comment on these reports, its CTO Kevin Scott mentioned in a recent blog post their plans to continue building large supercomputers for OpenAI in the future.
According to Scott, OpenAI uses these supercomputers to train cutting-edge models that are then made available in products and services for wider use. The collaboration between Microsoft and OpenAI is seen as mutually beneficial and poised to have a long-lasting impact. The oversight of MAI-1 is entrusted to Mustafa Suleyman, a co-founder of Google DeepMind and former CEO of AI start-up Inflection, who joined Microsoft along with his team in March.
Media sources have revealed that MAI-1 is distinct from the models previously released by Inflection, for which Microsoft acquired the intellectual property for $65 million. This new model is expected to be significantly larger than Microsoft’s previous open-source models, requiring more data and computing power, and therefore, a higher cost. MAI-1 is anticipated to have approximately 500 billion parameters, less than OpenAI’s GPT-4 which has over 1 trillion, but larger than smaller models from other companies. Microsoft’s Phi-3 mini model, unveiled recently, has 3.8 billion parameters.
