According to the Information report, Microsoft is getting ready to launch a brand-new in-house AI language model that can take on even Alphabet’s Google or OpenAI. This new model is named MAI-1, and this project is being supervised by the newly hired extremely efficient Mr. Mustafa Suleyman. He was the former CEO of Inflection, an AI startup company, and a co-founder of Google DeepMind.
However, the main purpose of the model is still unclear, and it is likely to be decided later after its performance is evaluated. According to sources, Microsoft could be previewing its latest venture anytime, especially now that the Build developer conference is approaching, which is scheduled to be held later this month.
Besides, the tech giant has refused to comment on this topic when asked by Reuters. All we can say now is that MAI-1 will be huge, “far larger” than the smaller open-source models that were trained by Microsoft before. This means that it’s likely to be priced higher than previously trained models.
It was only last month that Microsoft came up with Phi-3-mini, a much smaller AI model, as the company aimed to target a wider customer reach with its attractive and affordable options.
Microsoft is making bold strides in technological advancement by integrating generative AI into its suite of productivity software. The company is investing billions of dollars in OpenAI and deploying ChatGPT’s technologies into its software, leaving no stone unturned to build something amazing.
According to sources, the tech giant has been stocking a large cluster of servers powered by Nvidia’s graphic processing units and a huge collection of data to refine the model. As OpenAI’s GPT-4 and Phi-3 mini are said to constitute one trillion parameters and 3.8 billion parameters, respectively, MAI-1, the latest invention, is expected to possess around 500 billion parameters, according to the report.
Microsoft hired the profound AI professional Mustafa Suleyman in March as the head of its newly formed AI unit, along with several other employees from Inflection. Although the new project may be developed by leveraging training data from Inflection, as the report clarifies, it is not an addition or carryover from the start-up.
