Microsoft has introduced Mu, a new small language model integrated into the Windows 11 operating system. The new artificial intelligence model has been incorporated into the operating system as part of a structure aimed at improving the user experience locally by providing natural language support, particularly in the Settings app.
An artificial intelligence model has been integrated into Windows 11
Mu works directly with the operating system and stands out with its natural language search feature. This new model, tested in the Dev Channel of the Windows Insider program, powers the “smart agent” feature in the Settings app.
This agent enables users to interact with the system using multi-word natural language queries instead of short keywords. For example, when a user types “Bluetooth,” the relevant settings screen is listed in the traditional way. However, when a query like “How do I connect a new Bluetooth headset?” is entered, Mu understands the request and directly opens the Bluetooth device addition page.
The model was developed to be compatible with hardware featuring an NPU (Neural Processing Unit), part of Microsoft’s Copilot+ PC initiative. Previously, a model called Phi-Silica was used in these systems. However, with the new integration, the language model used in the Settings app is now Mu. Details about the underlying architecture and performance metrics of the model were also shared.
Microsoft announced that Mu is based on an encoder-decoder architecture and contains a total of 330 million parameters. The number of parameters was reduced by distributing some weights. The training process was carried out on Microsoft’s Azure Machine Learning platform. Nvidia A100 GPUs were used in the training.
Windows 11 is preparing to offer an unexpected feature. Users will now be able to create GIFs much faster.
In performance comparisons, Mu achieves approximately 47% lower first token latency compared to similar-sized decoder-based models. It also achieves 4.7 times faster decoding speed. According to Microsoft’s data, Mu offers similar output quality despite being only one-tenth the size of the Phi-3.5-mini model.
