Microsoft is building a massive LLM called MAI-1 (without OpenAI)

It's no secret that Microsoft has leaned heavily on OpenAI's tech to power its AI ambitions. But, things are changing! They're building their own in-house powerhouse AI model, dubbed MAI-1, to go toe-to-toe with giants like Google and OpenAI itself.

What’s going on here?

Microsoft is building a massive LLM named MAI-1.

What does this mean?

Microsoft has built baby LLMs like Phi (1-3B parameters), but MAI-1 is seriously beefed up. With roughly 500 billion parameters, it'll outmuscle Microsoft's previous efforts. OpenAI has supplied Microsoft with such big models until now (and it still is).

This signals a two-pronged strategy: they're still into smaller, cheaper models for everyday apps, BUT they're also swinging for the fences with state-of-the-art AI. Microsoft isn't content relying solely on OpenAI anymore.

Mustafa Suleyman, ex-Deepmind bigwig and founder of Inflection, leads the charge. Microsoft basically scooped up his team and tech two months ago. It is setting aside Nvidia GPUs just for MAI-1 and even using data from OpenAI's GPT-4 model to train (a big no-no for other companies as per OpenAI’s terms).

Why should I care?

Recently leaked emails show that Microsoft invested in OpenAI because CTO Kevin Scott and CEO Satya Nadella were legit scared of how good Google’s AI lead was. They were seriously behind. By all means, this investment has played out well.

Could it be that OpenAI was Microsoft’s way to win the catch-up? And now, since they have, Microsoft’s not relying on them 100%. It’s investing in OpenAI’s rivals (Mistral), going after the same enterprise customers, and building models that go toe-to-toe with GPT-4.

Well, we will keep tabs on MAI-1 for y’all, we might see a sneak peek at Microsoft's Build conference this month!