Microsoft works on a new artificial intelligence model: what we know about MAI-1

If there is one company that is particularly active in investing in artificial intelligence, it is it Microsoft. According to what was stated in an agency report The Informationthe Redmond giant, despite having already invested 10 billion dollars in OpenAI (the company that develops ChatGPT, Sora and Voice Engine), would have decided to work on a new generative AI modelcalled NEVER-1. The latter would compete not only with Alphabet (the holding company that controls Google) but, paradoxically, also with OpenAI itself.

Again according to the report in question, the project would be followed by Mustafa Suleymanco-founder of Google DeepMind and former CEO of the startup Inflection AIwhich gave birth to the “empathetic chatbot” Pi.

In addition to Suleyman, Microsot would also have hired a good part of Inflection AI's employees and would have paid a fortune 650 million dollars to grab material from the latter to use for part of the MAI-1 training process (second The Information for the other part of training, data generated by the model would be used GPT-4 of OpenAI).

Once MAI-1's training is concluded, it should handle approx 500 billion parameterswhich is a significant figure considering the fact that, according to some (unofficial) estimates, the GPT-4 model would be able to manage 220 billion of them trained on 8 sets of separate weights for a total of 1760 billion parameters.

The reason why Microsoft would create such a powerful new AI model is not clear. In any case, we may find out soon given that, according to the aforementioned report, Microsoft could reveal MAI-1 during the Microsoft Build conferencewhich will be held from 21st to 23rd May. Interviewed by ReutersMicrosoft declined to comment on the story.