Microsoft’s new AI model may challenge GPT-4 and Google Gemini

Microsoft is working on a new large-scale AI language model called MAI-1, which could compete with the latest models from Google, Anthropic, and OpenAI, according to a report from Microsoft. the information. This is the first time Microsoft has developed an internal AI model of this size since investing more than $10 billion in OpenAI for the rights to reuse the startup’s AI models. OpenAI’s GPT-4 not only powers ChatGPT, but also Microsoft Copilot.

MAI-1’s development is being led by Mustafa Soliman, the former Google AI leader who most recently served as CEO of AI startup Inflection before Microsoft acquired the majority of the startup’s employees and intellectual property for $650 million in March. Although MAI-1 may be based on technologies brought by former Inflection employees, it is said to be an entirely new large language model (LLM), as confirmed by two Microsoft employees familiar with the project.

With nearly 500 billion parameters, MAI-1 will be much larger than Microsoft’s previous open source models (such as Phi-3, which we covered last month), requiring more computing power and training data. This is said to put MAI-1 in a similar league as OpenAI’s GPT-4, which is… Rumor For over a trillion parameters (in a mix of experts) and much higher than smaller models like the Meta and Mistral models with 70 billion parameters.

The development of MAI-1 signals a dual approach to AI within Microsoft, focusing on small language models that run locally for mobile devices and newer, larger models that run by the cloud. Apple is said to be exploring a similar approach. It also highlights the company’s desire to explore AI development independently of OpenAI, whose technology currently powers Microsoft’s most ambitious AI features, including a chatbot built into Windows.

The exact purpose of MAI-1 has reportedly not been determined (even within Microsoft), and its optimal use will depend on its performance, according to one information source. To train the model, Microsoft dedicated a large cluster of servers with Nvidia GPUs and collected training data from various sources, including text generated by OpenAI’s GPT-4 and public Internet data.

Depending on progress in the coming weeks, information suggests that Microsoft may preview the MAI-1 as early as its Build Developer conference later this month, as reported by one of the sources cited by the post.