Skip to content

Microsoft is training its own AI model: MAI-1

  • by
  • 2 min read

After investing nearly $10 billion in OpenAI and getting access to the company’s popular GPT models, Microsoft is now gearing up to create its own AI model from scratch. The new model, internally called MAI-1, will reportedly have 500 billion parameters and is currently the Windows maker’s largest foray into AI models so far.

The 500 billion parameters aim to make the model stand its ground against the competition. While OpenAI, arguably the biggest company in all things AI, runs its latest model, GPT4, at around a trillion parameters, smaller AI models, such as those from Meta, are trained on around 70 billion parameters. Google’s Gemini (formerly Bard) is also rumoured to have been trained close to a trillion parameters, although the company hasn’t disclosed the exact number.

The parameter size is an important metric when measuring the performance of an AI model, as it can be a good indication of how accurate the resulting output will be. The higher the number of parameters, the greater the model’s accuracy. However, it also means that the model will require more computing power and training data to run, making it more expensive.

Microsoft has been hard at work acquiring said computation power and training data. The Information reports that the company is purchasing many server clusters from Nvidia and different AI training datasets to fuel the development of the new AI model.

The new model will be developed under Mustafa Suleyman, former CEO of AI startup Inflection. Suleyman isn’t the only Inflection employee being poached by Microsoft, either. The company has reportedly hired a majority of Inflection’s staff following a $650 million deal in March to acquire rights to the startup’s intellectual property.

It’s also important to note that MAI-1 will be completely built from scratch. Microsoft isn’t using anything from OpenAI or any of the models from Inflection as a base. The company isn’t even using Phi-3-mini, another in-house Microsoft AI model developed on a smaller scale to offer cost-effective AI options to a broadened clientele.

In the News: OpenAI partners with Stack Overflow to utilise coding content

Yadullah Abidi

Yadullah Abidi

Yadullah is a Computer Science graduate who writes/edits/shoots/codes all things cybersecurity, gaming, and tech hardware. When he's not, he streams himself racing virtual cars. He's been writing and reporting on tech and cybersecurity with websites like Candid.Technology and MakeUseOf since 2018. You can contact him here: yadullahabidi@pm.me.

>