Microsoft develops a new large language model to compete with those of Google and OpenAI

In early 2023, OpenAI announced an agreement with Microsoft to receive a “multi-year, multi-million dollar investment.” It was not the first time that Microsoft carried out one in the company of Sam Altmanfounded in 2015, but the one with the highest value until then: 10 billion dollars. Since then, the language model GPT-4 has been integrated with a wide variety of Microsoft products under the name Copilot, but that could change in the future. According to The Information, Microsoft is working on a big new language model, developed in-housewhich will compete with those of OpenAI and Googleamong others.

His name is MAI-1 and its development is directed by Mustafa Suleymanthe former Google AI leader who later became CEO of the start-up Inflectionwho created the chatbot Pi. Microsoft hired most of the staff of this company and bought its intellectual property for $650 million last March. However, MAI-1 will not have to do with Pi, but is about a completely new development.

MAI-1 is a large language model, LLM, which will have 500 billion parameters. This is half of those used by GPT-4, the LLM behind ChatGPT Plus and Copilot, of which a total of 1 billion parametersbut it is not something that OpenAI has revealed.

This model will also be “much larger” than previous, smaller, open source models the company has worked on. Microsoft last month launched an artificial intelligence model called Phi-3-miniwith 3.8 billion parameters, For example. The new model will therefore be much more expensive to train and operate. Microsoft has been assigning a large cluster of servers with Nvidia GPUs for MAI-1 and collecting training data from various sourcesincluding text generated by GPT-4 and public Internet data.

According to published information, Microsoft has yet to determine the exact purpose it will give to MAI-1 and it will do so depending on its performance. The company is likely to provide the first details about MAI-1 at the developer conference Build which will take place at the end of this month.

The question raised by this development is how it will affect the relationship between OpenAI and Microsoft and whether it will continue to use GPT-4 in its products. About this theme, Kevin ScottCTO of Microsoft, explained on LinkedIn, after the publication of the news, that both companies will continue to work closely in the future.