Microsoft develops MAI-1 AI model to compete with GPT-4 and Gemini
Microsoft is working on a new language model of artificial intelligence called MAI-1, which could potentially compete with the most advanced models from Google, Anthropic, and OpenAI.
This is the first time Microsoft has developed its own AI model of this scale after investing more than $10 billion in OpenAI. GPT-4 works not only in ChatGPT, but also in Microsoft Copilot.
MAI-1 development is led by Mustafa Suleiman, a former Google AI executive who recently served as CEO of Inflection startup before it was acquired by Microsoft for $65 billion in March, Ars Technica reports citing The Information.
MAI-1 will have approximately 500 billion parameters, while OpenAI GPT-4 has one trillion parameters. It will be much larger than previous open-source models from Microsoft, such as Phi-3, which has 3.8 billion parameters. It also significantly exceeds smaller models such as Meta and Mistral, which have 70 billion parameters.
The development of MAI-1 implies a dual approach to AI at Microsoft, focusing on both small local language models for mobile devices and larger modern models running in the cloud.
Reportedly, the exact purpose of MAI-1 has not been determined (even within Microsoft), and its most ideal use will depend on its performance, according to one of The Information’s sources. To train the model, Microsoft has dedicated a large cluster of servers with Nvidia GPUs and collects training data from various sources, including text generated by OpenAI’s GPT-4 and publicly available data from the Internet.
Depending on the progress made in the coming weeks, Microsoft may unveil MAI-1 at its Build developer conference later this month, The Information reports, citing one of its sources.
