According to sources, Microsoft is working behind the scenes on MAI-1: an LLM with more than 500 billion parameters. It would be the largest AI model that Microsoft develops without interference from OpenAI.
The Information believes it knows that Microsoft has big AI plans. Behind the scenes, work is underway on an LLM called MAI-1. If The Information’s information is correct, the LLM would be officially announced at the Build conference on May 21. With 500 billion parameters, the MAI-1 would be the largest model from Microsoft’s stable.
Microsoft poached almost the entire staff of the start-up Inflection AI and also took over the intellectual property for $650 million. The model was developed completely from the ground up indoors. Project leader Mustafa Suleyman is the co-founder of Google DeepMind.
Who has the largest?
With a large LLM, Microsoft is taking a different path. Previous models that it developed in-house were lightweight models with a ‘limited’ number of parameters: Phi-3 Medium has fourteen billion parameters. The big advantage of small AI models is that they require less energy to run. This reduces energy costs and makes mobile applications possible.
read also
OpenAI CEO: ‘GPT-4 is the dumbest LLM you will ever use’
Due to the collaboration with OpenAI, there was no need for Microsoft to develop large models itself. The Copilot applications all run on GPT-4. It is not yet clear how Microsoft would like to integrate the new model into its own ecosystem. We may find out more during Build.
Microsoft has already pumped billions of dollars into collaborating with OpenAI and developing its own models. Despite these significant investments, the strategy does not seem to be doing Microsoft any harm. Microsoft surpasses Apple as the most valuable company with a market value of more than three trillion dollars.
