Microsoft is reportedly developing a new AI model known as MAI-1, with the aim of competing with offerings from Google and Anthropic. The specific purpose of this model has not yet been determined, but it could potentially compete with OpenAI, in which Microsoft has made a significant investment.
According to reports, MAI-1 is overseen by Mustafa Suleyman, a co-founder of Google DeepMind who was hired by Microsoft in March along with his team from AI start-up Inflection. Despite Microsoft paying $65 million for the rights to Inflection’s intellectual property, MAI-1 is distinct from the models Inflection released.
Microsoft representative declined to comment on the reports about MAI-1 but referred to a blog post by CTO Kevin Scott. In the blog post, Scott mentioned plans to continue building large supercomputers for OpenAI in the future. OpenAI uses these supercomputers to train cutting-edge models that are then made available in products and services for widespread use.
MAI-1 will be much larger than the smaller, open-source models that Microsoft has previously trained. This means it will require more data and computing power, making it more expensive to develop. Roughly 500 billion parameters are estimated for MAI-1, in comparison to over 1 trillion parameters for OpenAI’s GPT-4. Smaller models released by other companies like Meta Platforms and Mistral have around 70 billion parameters, while Microsoft’s Phi-3 mini model released last month offers 3.8 billion parameters.
