The development of its own AI models could enable Microsoft to compete with the companies that it currently partners with for AI projects, such as OpenAI.
Microsoft has released two in-house artificial intelligence (AI) models. The tech giant has been focused on building AI foundation models since last year and has developed both a strong team and the relevant infrastructure, it claims.
Last year, it hired co-founder of DeepMind and Inflection AI, Mustafa Suleyman, to create Microsoft AI, a new division focused on advancing the company’s consumer AI products – such as Copilot – and its research in this field. And in January, it announced a new AI engineering group to be led by former Meta engineering chief, Jay Parikh.
The first of the new models is the MAI-Voice-1, which Microsoft has described as a system that is expressive and capable of generating natural speech. It can be found through Copilot Daily, Podcasts and as a new Copilot Labs experience. It can be used across single and multi-speaker scenarios.
The second model, MAI-1-preview, has just entered the public testing phase on LMArena. According to Microsoft, it is an end-to-end trained foundational model that will be rolled out in Copilot for certain text-based use cases over the coming months. It was trained on roughly 15,000 Nvidia H-100 GPUs.
In an interview with Semafor, Suleyman said that the company used techniques developed by the open-source community to stretch the capabilities of the models with minimal resources.
“Increasingly, the art and craft of training models is selecting the perfect data and not wasting any of your flops on unnecessary tokens that didn’t actually teach your model very much,” he said.
Microsoft has largely relied on other organisations when working on projects that demand AI-driven innovation, such as OpenAI, however, its latest venture could enable it to achieve a degree of independence and establish itself as a key player in the AI race.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
