It is reported that Microsoft is developing its own large language model (LLM) called MAI-1, competing with AI products from OpenAI, Google and Meta.
Microsoft ‘s latest large language model, MAI-1, has 500 billion parameters, according to The Information . This value is higher than all the LLMs that Microsoft has ever created, far exceeding the 70 billion USD of Meta Llama 3, but less than a trillion USD of OpenAI GPT-4.
In addition, after many years of using OpenAI products for more than 10 billion USD, Microsoft announced the first LLM developed by itself. Previously, Microsoft announced the Phi-3 mini small-scale model with 3.8 billion parameters for Ollama, Azure and Hugging Face services.
The new LLM requires a lot of computing power and training data. Microsoft is said to have provided many servers equipped with Nvidia chips for the project.
Mustafa Suleyman, co-founder of Google DeepMind and former CEO of startup AI Inflection , is the head of MAI-1. In March, Microsoft spent $650 million in an unusual deal to poach personnel and models created by AI Inflection.
However, the source said that MAI-1 is an independent product of Microsoft instead of being given to Suleyman’s former company. Microsoft has not issued any statement yet. At the Build event held at the end of May, the company can introduce MAI-1 to developers.
Despite not having a well-known language model, Microsoft has supported ChatGPT and OpenAI with investment and a relationship with CEO Sam Altman. Microsoft’s Copilot chatbot , working on GPT-4 Turbo, is one of the best chatbots available today.