Microsoft Challenges Google, OpenAI with New AI Model
Microsoft’s ambitious efforts to develop a large-scale AI language model, named MAI-1, signal a significant shift in the company’s AI strategy. This new model, set to compete with leading models from Google, Anthropic, and OpenAI, represents Microsoft’s first in-house AI model of such magnitude since its substantial investment in OpenAI. OpenAI’s GPT-4 currently powers many of Microsoft’s generative AI features, including Microsoft Copilot. However, the development of MAI-1 indicates Microsoft’s desire to establish its own cutting-edge AI capabilities.
Leading the development of MAI-1 is Mustafa Suleyman, the former head of Google AI and CEO of the AI startup Inflection. Microsoft acquired most of Inflection’s staff and intellectual property for $650 million in March, bringing valuable expertise to the project. According to two Microsoft employees familiar with the initiative, MAI-1 is a completely new large language model (LLM) that, while potentially building on techniques from former Inflection staff, stands as a unique endeavour.
With approximately 500 billion parameters, MAI-1 is poised to be significantly larger than Microsoft’s previous open-source models, such as Phi-3. This places MAI-1 in a competitive league with OpenAI’s GPT-4, which is rumoured to have over 1 trillion parameters. The sheer scale of MAI-1 requires substantial computing power and training data, highlighting the advanced infrastructure Microsoft is dedicating to this project. By allocating a large cluster of servers with Nvidia GPUs and compiling training data from various sources, including text generated by GPT-4 and public internet data, Microsoft is ensuring that MAI-1 has the resources needed for its development.
The creation of MAI-1 reflects Microsoft’s dual approach to AI, focusing on both small, locally-run language models for mobile devices and larger, state-of-the-art models powered by the cloud. This approach is similar to Apple’s reported strategy, indicating a broader industry trend. Moreover, the development of MAI-1 underscores Microsoft’s willingness to explore AI development independently from OpenAI, despite their existing partnership. OpenAI’s technology currently powers some of Microsoft’s most ambitious generative AI features, including a chatbot integrated into Windows.
The exact purpose of MAI-1 has not yet been determined, even within Microsoft. Its most ideal use will depend on its performance, according to one of The Information’s sources. This flexibility suggests that MAI-1 could potentially serve a variety of applications, adapting to different needs as it evolves. Depending on the progress made in the coming weeks, Microsoft may preview MAI-1 as early as its Build developer conference later this month, providing a glimpse into the future of its AI capabilities.
The development of MAI-1 comes amid a broader context of Microsoft’s strategic investments in AI. On Monday, OpenAI announced a
