Microsoft takes on Google, OpenAI with first in-house AI models
What's the story
Microsoft has launched its first-ever in-house artificial intelligence (AI) models, namely the MAI-Voice-1 and MAI-1-preview. The tech giant claims that the new MAI-Voice-1 speech model can generate a minute-long audio in less than a second using just one GPU. Meanwhile, the MAI-1-preview "offers a glimpse of future offerings inside Copilot."
Feature integration
MAI-Voice-1 is already powering Microsoft's Copilot daily
The MAI-Voice-1 model is already powering a number of Microsoft features, including Copilot Daily. This feature has an AI host read out the day's top news stories. The model is also being used to create podcast-style conversations to explain different topics. You can test out the capabilities of MAI-Voice-1 on Copilot Labs, where you can customize what you want it to say as well as its voice and speaking style.
Advanced training
MAI-1-preview was trained on 15,000 NVIDIA H100 GPUs
As per Microsoft, the MAI-1-preview model was trained on some 15,000 NVIDIA H100 GPUs. It is designed for users looking for an AI model that can follow instructions and "provide helpful responses to everyday queries." Microsoft's AI chief Mustafa Suleyman had previously said that the company's internal AI models aren't focused on enterprise use cases. "My logic is that we have to create something that works extremely well for the consumer and really optimize for our use case."
Future integration
Microsoft's plans for MAI-1-preview integration and public testing
Microsoft plans to integrate the MAI-1-preview model into certain text use cases in its Copilot AI assistant, which currently uses OpenAI's large language models. The company has also started public testing of its MAI-1-preview model on the AI benchmarking platform LMArena. This move shows Microsoft's intention to push boundaries and take on leaders like OpenAI and Google with its in-house developed AI models.