Learn how this innovation is unlocking new possibilities in artificial intelligence.

Clara’s desk was cluttered, but her monitor’s colors were always off. She spent her mornings searching for the brightness slider in Windows Settings. “It’s always just there,” she’d say, frustrated, “but where is there today?” This search took away from her creative time, leaving her drained before she even started.

Her old PC couldn’t handle her needs for quick system tweaks.

Then, she got a new Copilot+ PC, promising a new era of on-device AI.

The world of artificial intelligence is changing fast. It’s moving from experiments to real changes in our lives. Microsoft is leading this change with AI that’s safe and secure. Microsoft Mu is a big step in this direction, bringing AI to your PC.

Mu is not just an update. It’s a big change in how we use technology. It works fast and locally, making AI experiences better and more private. You don’t need the internet to use it.

The Imperative of On-Device AI: Why Mu Matters Now

AI has promised a lot, but it’s often been frustrating. Clara, a digital artist, spent hours in Windows Settings trying to fix her monitor’s colors. Her old PC couldn’t handle simple tasks, wasting her time.

Microsoft Mu solves these problems. It works well on your PC, even when you’re offline. It makes everyday tasks easier and keeps your data safe.

Decoding Mu: Understanding This Micro-Sized Powerhouse

So, what is Mu? It’s a small, but powerful language model. Despite its size, it’s very capable. Mu is designed for small PCs and works fast, even offline.

The Engineering Marvel: Mu’s Architectural Innovations for Peak Performance

Mu’s design is truly brilliant. It’s small but powerful, thanks to its engineering.

Encoder-Decoder Architecture: The Efficiency Game-Changer

Mu uses a special architecture that’s different from old models. It has an encoder that changes your input into a fixed size, and then a decoder makes output tokens from that. This makes Mu very efficient, using less computation and memory.

On devices like Qualcomm Hexagon NPU, Mu is much faster. It’s about 47% quicker for the first token and 4.7 times faster for decoding. These improvements are key for AI that needs to work fast.

Mu was made to work well on NPUs and edge devices. It was carefully designed to fit the NPU’s needs. For example, it uses the right sizes for operations to run smoothly.

Mu also saves memory by sharing weights in certain parts. It only uses operators that the NPU can speed up. This makes the most of the NPU’s power.

To get better accuracy and speed, Mu has three transformer upgrades:

Dual LayerNorm (pre- and post-LN) stabilizes training without adding much overhead.

Rotary Positional Embeddings (RoPE) help the model understand long sequences better. It can handle sequences longer than it was trained on.

Grouped-Query Attention (GQA) reduces memory and latency by sharing keys and values. This makes the model faster and uses less power.

Mu was trained on many GPUs and learned from other models. It started with pre-training on educational tokens. Then, it learned from Microsoft’s Phi models.

Low-rank adaption (LoRA) methods helped Mu get even better. This made it very efficient in learning from task-specific data.

Microsoft used advanced techniques to make Mu run well on devices. They changed the model’s weights and activations to use less memory and compute. This made Mu fast and efficient.

Mu is now part of the Windows Settings AI agent. It helps change system settings easily. This makes it simpler to find and change settings.

Mu’s AI agent is integrated into the Settings app. It responds quickly to user queries. For example, it can change settings in under 500 milliseconds.

Remember Clara’s struggle with her monitor settings? She got a new Copilot+ PC and typed into the Settings search box: “Make my secondary monitor colors warmer” [conversation history]. Mu, the AI agent, quickly suggested adjusting the color temperature. With one click, her monitor’s colors were perfect [conversation history].

She then asked for more complex settings: “Set my notifications to only show from my art software when I’m in fullscreen mode, and mute everything else.” Mu, designed for complex tasks, understood her query well. It offered a precise solution [conversation history, 37, 38]. Mu worked fast, processing over 100 tokens per second [conversation history, 44, 59].

To achieve this precision, the Settings Agent’s training was carefully refined. Microsoft increased training samples by 1300 times and expanded settings coverage. They used synthetic labeling and diverse phrasing to improve the model. This way, the agent can handle short or partial-word inputs and multi-word queries for precise responses.

The Mu-powered AI agent is now available to Windows Insiders in the Dev Channel. It works with Copilot+ PCs running Windows 11 Build 26120.3964 (KB5058496) or higher.

The Broader Impact: Beyond Settings to a Transformed Future

Microsoft Mu is more than a single feature. It’s a step towards a future where AI is part of our daily lives.

The Rise of Intelligent Agents:

Mu is part of a new generation of AI agents. These agents can handle complex tasks and work on your behalf. They are seen as the “apps of the AI era,” changing how we work and live. Microsoft ensures human oversight is key in this AI-powered world.

AI Companions for Everyday Life:

Outside work, Microsoft Copilot, powered by Mu, will be your AI companion. It will simplify daily tasks, save time, and protect your privacy. Imagine Copilot Daily summarizing news or Copilot Vision understanding web pages to help you.

Efficiency and Sustainability at Scale:

As AI grows, so does its need for resources. Microsoft is working to make AI more efficient and sustainable. They’re using custom silicon and liquid cooling for large AI systems. Microsoft aims to be carbon negative, water positive, and zero waste by 2030.

Accelerating Scientific Breakthroughs:

AI is also changing scientific research and innovation. It’s driving advances in supercomputing and biomolecular science. AI models like Mu are solving big problems, leading to new discoveries in drug development and sustainable materials.

Conclusion: A New Era of Personal Computing

Microsoft Mu is a big step in on-device AI, making AI experiences powerful and efficient on your PC. Its architecture and training show Microsoft’s commitment to AI advancements. Mu is changing how we interact with technology, making it more intelligent and seamless. This is a step towards a future where technology is deeply integrated into our lives, safely and securely.

Take the Leap: Experience On-Device AI with Mu

Are you ready to experience the next generation of personal computing?

For Windows Insiders:

If you have a Copilot+ PC, you can try out the Mu-powered AI agent in Windows Settings right now. Just install Windows 11 Build 26120.3964 (KB5058496) or higher. Provide your feedback to Microsoft as they continue to refine this groundbreaking experience! Explore Copilot+ PCs: Learn more about the hardware designed to unleash the full potential of these powerful on-device AI capabilities.