Blog

American multinational firm Microsoft is training a new AI language model to compete with the offerings of OpenAI and Google.

Microsoft’s latest in-house model, called MAI-1, is under the supervision of Mustafa Suleyman, a recent hire and Co-founder of Google DeepMind, who previously served as the CEO of AI startup Inflection. According to a report by The Information, the development was confirmed by two Microsoft employees familiar with the effort.

In March, Microsoft appointed Suleyman to lead its consumer AI unit and recruited multiple employees from Inflection. The report clarified that the new model isn’t directly inherited from Inflection, although it might have leveraged training data from the startup.

The report added that the AI model is expected to be “far larger” in size than the smaller, open-source models previously trained by Microsoft, suggesting higher costs associated with its development and maintenance.

MAI-1 will reportedly feature around 500 billion parameters, while OpenAI’s GPT-4 is reported to boast one trillion parameters and Phi-3 mini has 3.8 billion parameters.

@media (max-width: 769px) {
.thumbnailWrapper{
width:6.62rem !important;
}
.alsoReadTitleImage{
min-width: 81px !important;
min-height: 81px !important;
}

.alsoReadMainTitleText{
font-size: 14px !important;
line-height: 20px !important;
}

.alsoReadHeadText{
font-size: 24px !important;
line-height: 20px !important;
}
}

Also Read

AI will create enormous wealth and have bigger impact than industrial revolution, says Ema’s Surojit Chatterjee

The report said the model’s exact purpose hasn’t been determined yet and will rely on its performance. It added that the company plans to give a sneak peek of the new model at its Microsoft Build developer conference later this month.

In April, the tech behemoth rolled out Phi-3-mini, a smaller AI model to appeal to a broader range of clients with more budget-friendly options. With investments in OpenAI totalling billions of dollars, Microsoft has integrated its technology into its suite of productivity software. 

The report added that the US-based company has been allocating a huge cluster of servers, outfitted with Nvidia’s graphic processing units, along with extensive datasets to enhance the model.