Microsoft could challenge GPT-4 and Google with new AI model: MAI-1

Microsoft is claimed to be coming into one other participant into the AI mannequin market, with its latest iteration titled ‘Mai-1,’ and experiences counsel it could possibly be higher than the remaining.

Though the expertise large has already invested over $10 billion into OpenAI, the corporate is claimed to be coaching a brand new in-house AI mannequin that may compete.

The information comes from The Information who say the event of the massive language mannequin is being led by Mustafa Suleyman, the previous Google AI chief who served because the CEO of the AI startup Inflection.

Inflection is a California-based firm that creates conversational AI you can speak to. Its group created the private AI known as Pi which was launched in Could 2023. Microsoft was considered one of its buyers in a $1.3 billion funding spherical.

Each Suleyman and the opposite co-founder of Inflection, Karén Simonyan, have been poached by Microsoft in March 2024. Different members of the group joined the duo at Microsoft too.

MAI-1 is anticipated to be created solely by Microsoft and is separate from Inflection, however it may construct on coaching information and different expertise as they paid for the start-up’s rights to its mental property for the sum of $650 million in March.

What we find out about Microsoft’s MAI-1

Other than this report, there hasn’t been any affirmation or information about MAI-1 formally printed.

The rumor suggests it may have roughly 500 billion parameters, that means MAI-1 might be considerably bigger than Microsoft’s previous open-source models like Phi-3.

OpenAI’s GPT-4 is rumored to have over 1 trillion parameters and the French AI instrument Mistral is regarded as at 70 billion parameter fashions. Microsoft and Mistral signed a multi-year partnership in February.

Whereas start-ups are discovering their ft and making waves in AI like Anthropic and xAI, Microsoft is a large firm that ought to have the massive quantities of information wanted to coach the MAI-1 mannequin.

Microsoft has been allocating a big cluster of servers with Nvidia GPUs and compiling coaching information from varied sources for MAI-1, in keeping with the supply. This contains textual content generated from OpenAI’s GPT-4, alongside public web information.

The Data additional experiences how a preview of MAI-1 may come as early because the Microsoft Construct convention later this month (Could 23 – 25) if progress is made inside improvement.

Featured Picture: By way of Ideogram