Microsoft is Preparing for New AI Model to Compete Against OpenAI

Microsoft is currently in the process of crafting a groundbreaking large-scale AI language model known as MAI-1. This development positions Microsoft in potential competition with leading models emerging from industry giants such as Google, Anthropic, and OpenAI, as per insights from The Information.

This is the largest AI model Microsoft has created internally since it invested more than $10 billion in OpenAI in exchange for the ability to utilize the startup's models. Microsoft Copilot and ChatGPT are powered by OpenAI's GPT-4.

Mustafa Suleyman, formerly a leader in Google's AI division and more recently CEO of the AI startup Inflection, is spearheading the development of MAI-1. Following Microsoft's acquisition of the majority of Inflection's personnel and intellectual assets for a whopping $650 million in March, Suleyman's leadership now steers the creation of this new AI venture.

While drawing upon the expertise brought over by former Inflection employees, MAI-1 is said to be an entirely novel large language model (LLM), as affirmed by insiders at Microsoft familiar with the project. Boasting approximately 500 billion parameters, MAI-1 dwarfs Microsoft's previous open-source models like Phi-3, necessitating significantly more computational resources and training data.

This places MAI-1 in the same echelon as OpenAI's speculated GPT-4, rumored to exceed 1 trillion parameters in a mixture-of-experts framework, surpassing smaller models such as those developed by Meta and Mistral, which hover around 70 billion parameters.

What's unique about the Microsoft ongoing chatbot model MAI-1?

The emergence of MAI-1 underscores Microsoft's multifaceted approach to AI, balancing the development of compact, locally operated language models for mobile platforms with the creation of expansive, state-of-the-art models powered by cloud infrastructure. Apple is purportedly pursuing a similar strategy. Additionally, it reflects Microsoft's independent exploration of AI development outside of its collaboration with OpenAI, whose technology presently underpins Microsoft's most ambitious generative AI endeavors, including an integrated chatbot feature within Windows.

Interestingly, the specific function of MAI-1 remains undetermined, even within Microsoft circles, with its optimal application contingent upon its performance, according to insiders cited by The Information. To train this model, Microsoft has allocated a substantial cluster of servers equipped with Nvidia GPUs and aggregated training data from diverse sources, including text generated by OpenAI's GPT-4 and publicly available internet data.

Anticipating potential milestones in the near future, reports suggest that Microsoft may unveil a preview of MAI-1 as early as its Build developer conference later this month, according to sources cited by The Information.