Microsoft Training New AI Model Big Enough to Compete With Major LLMs

Microsoft is developing a new AI large language model (LLM) codenamed MAI-1, which has the potential to compete with the AI LLMs developed by Google and OpenAI.

It’s the first time that Microsoft has developed such a large-scale LLM since investing billions of dollars in OpenAI for the rights to deploy the latter’s technology across its suite of productivity software.

The development of MAI-1 is being overseen by Google DeepMind and Inflection co-founder Mustafa Suleyman, a report by The Information revealed. In March 2024, Microsoft acquired a number of Inflection’s staff in a $650 million deal.

According to the report, which was released on May 6, MAI-1 will have approximately 500 billion parameters, placing it somewhere between the reported one trillion parameters of OpenAI’s GPT-4 and 70 billion parameters of Meta’s Llama 3 AI model.

It is also expected to be “far larger” and more expensive than any of Microsoft’s previous, smaller, open-source AIs (Phi-3 and WizardLM-2), as it will require more computing power and training data.

While MAI-1 may leverage techniques and training data from Inflection, it remains distinct from any models or technologies produced by OpenAI or Inflection. According to Microsoft employees who are acquainted with the project, MAI-1 is a completely novel LLM developed internally by Microsoft.

Microsoft has not yet announced the exact purpose of MAI-1, and its exact use will depend on its performance. In the meantime, the company has been allocating a large cluster of servers with Nvidia GPUs and using large amounts of data from various sources to improve the model.

Depending on its progress, reports say the company may preview MAI-1 at the Build developer conference later this month, but this isn’t confirmed .