| Microsoft also introduced MAI-1-preview – a platform language model that was trained entirely in-house. (Source: Mashable India) |
On August 29, Microsoft announced the first two artificial intelligence (AI) models developed and trained internally by the company, marking a significant milestone in its strategy to reduce reliance on external technology.
The first product, called MAI-Voice-1, focuses on creating natural-sounding speech. According to Microsoft, this model boasts superior capabilities, being able to produce one minute of audio in less than one second using only a single graphics processing unit (GPU).
MAI-Voice-1 is now integrated into several Copilot services, such as Copilot Daily – which provides daily audio news bulletins, and podcast-style talk shows to explain specific topics. Users can also experiment with the model through the Copilot Labs platform, with options to adjust intonation and style.
Alongside this, Microsoft also introduced MAI-1-preview – a platform language model trained entirely in-house. The training process utilized up to 15,000 Nvidia H100 chips, allowing the model to process text instructions and provide helpful answers to everyday needs.
According to Microsoft, the MAI-1 preview is just a "trial version," providing a glimpse into what Microsoft will implement in the Copilot ecosystem in the future. The model has begun testing on the LMArena platform, which evaluates the performance of AI systems, and will gradually be integrated into several Copilot services in the coming weeks.
This move comes as Microsoft's Copilot services still rely heavily on technology from OpenAI.
Microsoft believes that building specialized, multi-functional AI toolkits to serve a wide range of needs and purposes will bring immense value to users and pave the way for a new phase of competition in the global AI race.
Source: https://baoquocte.vn/microsoft-trinh-lang-2-mo-hinh-tri-tue-nhan-tao-ai-tu-phat-trien-326083.html






Comment (0)