Microsoft is done relying on others for its models. According to reports, though the company had been training smaller models like Orca and Phi all this while (using GPT and incorporating Meta’s Llama on its platform), this time it is training a model large enough to compete with others.

Referred to as MAI-1 (possibly Microsoft AI-1), the model is being developed internally by the company, and is around 500 billion parameters in size.

Its development is being headed by Mustafa Suleyman, formerly a leader in AI at Google and most recently CEO of the AI startup Inflection, who now oversees Microsoft’s AI division. In March, Microsoft acquired a majority of Inflection’s staff and paid $650 million for its intellectual property rights.

However, MAI-1 is a Microsoft-developed model, distinct from those previously developed by Inflection. While it may leverage training data and technology from the startup, it is an independent project.

Though the exact purpose has not been disclosed just yet, it is possible that Microsoft might incorporate its products into all its Copilot products. This would mean that the company will move away from OpenAI’s GPT and Codex models.

This comes in the backdrop of OpenAI partnering with Stack Overflow to improve its products. The big-tech company is also planning to release its search engine powered by Bing possibly as a way to compete with Google, which Microsoft has a history of doing.

No competition?

Microsoft’s strategy has highlighted three trends for 2024 – small language models, multimodal AI, and AI in science.

This time around, it seems like the company has decided to incorporate multimodal AI with large language models. The small language models will continue to be incorporated into the company’s on-edge use cases, such as laptops, while larger ones might be available in other core products.

All of this comes only weeks before the Microsoft Build conference. So it is possible that the model will be introduced at the conference. To train the new model, Microsoft has allocated a significant cluster of servers equipped with NVIDIA GPUs.

To make the case clear, Microsoft’s CTO Kevin Scott went on LinkedIn to explain that this is not in any way a competition with OpenAI. “I’m not sure why this is news, but just to summarise the obvious: we build big supercomputers to train AI models. Our partner OpenAI uses these supercomputers to train frontier-defining models; and then we both make these models available in products and services so that lots of people can benefit from them. We rather like this arrangement,” he said. 

He further stated that the company has always been building bigger and better supercomputers for OpenAI to further AI research and wishes to continue this arrangement. “There’s no end in sight to the increasing impact that our work together will have,” he added. 

So it seems like Sam Altman’s $50 billion dream of building AGI is going to get funded by Microsoft. 

Scott also clarified that all the research that Microsoft does is about building models: “AI models turn out to be interesting things to work on, and our researchers do great work studying and building them.”

He also confirmed that there would be more models coming out soon, including MAI, Phi, and even Turing. 

But is it true?

Even though Scott clarified that there is no competition between OpenAI and Microsoft, it is still worth a wonder if a model as big as MAI-1 would actually be used for Microsoft products instead of OpenAI’s GPT.

Moreover, it would be ideal for Microsoft to have a backup plan just in case the deal with OpenAI falls through, as has been the case with several others in the field. Chief Satya Nadella seems to be playing a different AI game. Under him, Microsoft has invested in all kinds of AI companies, from OpenAI and Mistral to Databricks and Figure AI. 

Recently an email from 2019 resurfaced, in which Scott told the Microsoft team that they needed to invest in OpenAI as the competition with Google was rising and that their AI model was “scarily good”.

Suleyman recently posted on X saying, “AI is everything at Microsoft”. He also highlighted that the company is building massive products using AI and has a definite vision for Copilot.

Everything about this seems forced. There seems to be no other reason for Microsoft to build such large models and spend so much on compute if they’re not making it for commercial purposes. Moreover, on his hiring, Suleyman was touted as the “new” Sam Altman. 

So, is Microsoft possibly becoming the OG of AI, aka OpenAI?