MAI-1: Microsoft’s new artificial intelligence to compete with Google, OpenAI and Anthropic

Microsoft is training a new Artificial Intelligence (AI) model, known internally as MAI-1, which handles 500,000 million parameters and is “large enough” to compete with next-generation models launched by other technology companies, such as Google, Anthropic and OpenAI.

The team in charge of its development is supervised by the former Google AI leader and CEO of the startup Inflection AI, Mustafa Suleyman, as recently reported by The Information.

This medium has also indicated that the new AI model that Microsoft is working on, known internally as MAI-1, will be much larger in size than any of the smaller open source models that Microsoft has previously developed.

This means that this model, which will have approximately 500,000 million parameters, will require more computing power and training data, which will make it more expensive, according to sources consulted by this medium.

Additionally, MAI-1 may be based on training data and other technologies from Inflection, which focuses on the development of personalized AI. Among his works stands out Pi, a ‘chatbot’ capable of providing knowledge based on the interests and needs of users.

With this new AI, which Microsoft could refer to at the next Microsoft Build developer conference (scheduled for May 21-23), the technology company hopes to compete with the latest generation models from Google, Anthropic and OpenAI.

It is worth remembering that in recent months Microsoft has presented different language models, such as Phi-1, which works with 1.3 billion parameters and was updated a few months ago with Phi-1.5, which achieved performance comparable to that of models with a size up to five times larger.

The latest iteration is Phi-2, capable of obtaining reasoning and language understanding results with “state-of-the-art” performance among models with less than 13 billion parameters, as advanced a few weeks ago.