According to The Information, Microsoft's latest large language model (LLM), named MAI-1, contains 500 billion parameters. This is larger than any LLM that Microsoft It has been developed with a parameter count exceeding 70 billion—higher than Meta Llama 3’s 70 billion, but lower than OpenAI GPT-4’s one trillion.
This is also the first LLM developed in-house by Microsoft to be unveiled after a long period of relying on products from OpenAI, where the company has invested more than $10 billion over the years. Previously, Microsoft also announced the Phi-3 mini, a smaller-scale model with 3.8 billion parameters, which is used in services such as Azure, Hugging Face, and Ollama.
Given its powerful capabilities, the new LLM requires significant computational power as well as training data. Microsoft is reported to have allocated a large number of servers equipped with Nvidia chips to support this project.
MAI-1 is supervised by Mustafa Suleyman, co-founder of Google DeepMind and former CEO of the startup AI Inflection. Back in March, Microsoft made an unusual move when it spent $650 million to acquire most of the staff and models developed by AI Inflection.
However, according to sources, MAI-1 is an independent product developed by Microsoft, not a spin-off from Suleyman’s former company. Microsoft has not yet commented. The company may introduce MAI-1 to developers at the Build event, scheduled for late May.
Microsoft is currently one of the leading Big Tech companies in the field of AI. Although it does not own a well-known large language model, the company has played a major role in promoting ChatGPT and OpenAI through its investment and relationship with CEO Sam Altman. Microsoft’s Copilot chatbot, which runs on GPT-4 Turbo, is also considered one of the best chatbots available today.
Luu Quy
EZ Fast Tech – Partnering with you to build a successful website using AI technology!
Contact us today for free consultation and support:
- Website: https://ezfasttech.com
- Hotline: 0772869820
- E-mail: [email protected]
