Microsoft"s long-time partnership with ChatGPT OpenAI has resulted in the company getting a huge leap from its competition like Google, Amazon, and others in the generative AI industry. However, the company is reportedly going at it alone in making a new large language model that could end up rivaling ChatGPT.
According to The Information, via unnamed sources, the in-house LLM from Microsoft is called MAI-1 (presumably "M" stands for "Microsoft). It"s being developed with 500 billion parameters, compared to OpenAI"s GPT-4 which has over 1 trillion parameters.
The report says that the development of MAI-1 is being headed up by Microsoft"s new AI chief Mustafa Suleyman. He was hired by the company in March after co-founding Google"s DeepMind AI division, and then later forming his own AI startup Inflection.
Before being hired by Microsoft, Suleyman was working on another LLM for Inflection. The Information reports that this new MAI-1 LLM that he is supervising does not share any code with his previous work with Inflection, but it may use some training data from that startup.
Microsoft likely wants to be less dependent on OpenAI for help in creating AI services like its various Copilot versions, which all currently use some version of ChatGPT.
Today"s report does not state what services will be used by the upcoming MAI-1 LLM. The Information says that it"s possible Microsoft could reveal more about this project at its annual Build developers conference, which gets underway in Seattle on May 21.
Microsoft has already released a number of smaller and lightweight open-source LLMs. Last week, it released the new Phi-3 family of small LLMs. They work with as low as 3.8 billion parameters. They are also designed to run on local devices such as smartphones, but they are still made to handle advanced natural language processing for a wide variety of tasks.