Microsoft has started the development of a LLM to be a new (Large Language Model), which will be in a position to challenge the Gemini a Google + and a GPT-4 of OpenAI. According to the sources of The Information, the model is called the MAI-1, and it will be around 500 billion parameters.
The work of Mustafa Suleyman
Microsoft has incorporated the Copilot on the different types of services and Windows 11. Assistant to the digital, however, is based on the model of GPT-4 of OpenAI. There is an agreement with the billions of dollars between the two companies, which provides for the use of the models, GPT, and DALL THE Microsoft, as the OpenAI use the infrastructure to Azure training.
The company's Redmond has been developed by some of the models, the smaller IT (and some open source). According to The Information, MAY-1 to about 500 billion parameters, and thus belongs to the same category and as a Gemini (Google), and GPT-4 (OpenAI). Therefore, it seems that Microsoft wants to compete with the company, led by Sam Altman. In fact, GPT-4 still holds an advantage of a well (of which there are over 1 trillion parameters).
The training will be use in servers with GPU NVIDIA, and data from a variety of sources, including public records, in the internet and in the text of the generated by the GPT-4. The project is headed by Mustafa Suleyman, the former co-founder of the Inflection that HE and the CEO of Microsoft, HE. Another view of the previous and the model can be displayed throughout the conference, and Build 2024 (21-23 may).
A confirmation of the indirect came to the Head of Technology, Kevin Scott, via a post on LinkedIn. The executive notes that this isn't anything new after the Microsoft build superkompjuterë for the training of the models, OpenAI, and has been developed by the effectiveness of IT for a few years, a few of which are open source.
Discussion about this post