On May 7th, reports surfaced that Microsoft is training an in-house artificial intelligence language model named MAI-1, which rivals the scale of large models from Google and OpenAI.
According to two Microsoft employees, the MAI-1 model is led by Mustafa Suleyman, the former CEO of Inflection and co-founder of Google DeepMind, recently recruited by Microsoft.
The exact purpose of the MAI-1 model is yet to be determined by Microsoft, depending on its performance. It is speculated that Microsoft may preview this new model at the Build developer conference later this month.
Microsoft has declined to comment on this development.
The scale of MAI-1 is said to be much larger than Microsoft's previously trained small open-source models, indicating that it will also be more costly.
Last month, Microsoft introduced a small artificial intelligence model called Phi-3-mini, aiming to attract a broader customer base with cost-effective products.
Microsoft has invested billions of dollars in OpenAI and integrated OpenAI's artificial intelligence technology into its productivity software, establishing a leading position in the field of generative artificial intelligence in the early competition.
To enhance this model, Microsoft has reportedly set up large server clusters equipped with NVIDIA GPUs and accumulated a vast amount of data.
It is reported that MAI-1 will have approximately 500 billion parameters. In comparison, OpenAI's GPT-4 is said to have 1 trillion parameters, while Phi-3 mini has 3.8 billion parameters.
In March of this year, Microsoft appointed Suleyman as the head of the newly established consumer AI division and also hired several employees from Inflection.
The reports suggest that the new model is not directly from Inflection but may have utilized training data from this startup company.
