May 2024

Tech breakthroughs often flash then fade. Once upon a time, MS-DOS changed how we talked to computers. Soon after, Windows 1.0 put a mouse in the house. Recently, we’ve reached for the cloud. Now, we live in the time of AI.

While each digital era differs, they’re all grounded by the same word – an expression that’s core to countless hours of coding and the IT wonders they produce: Build.

In tech, “the build” is the bedrock of every solution. If we’re getting technical – and hey, that’s literally what we do here – “build” means converting source code into standalone software that can be run on a computer.

Today, Microsoft developers continue those builds full bore. This month, we'll tell you about one of their transformative creations, Copilot+ PCs, the fastest, most intelligent Windows PCs ever made.

We’ll also share a snapshot from Microsoft Build, our annual event for developers, where we showcased tools like Microsoft Fabric, which helps developers leverage data in motion to craft intelligent apps. And we’ll detail a new dispatch from the bustling intersection of AI and work.

For this edition, we were hoping if they built it, you would come. So glad you did.

Imagine a PC that allows you to perform tasks and tricks you can’t do on any other personal computer.

This month, Microsoft introduced a new class of Windows devices that do exactly that: Copilot+ PCs.

Available June 18 via preorder, Copilot+ PCs will let you generate and refine AI images in near real time directly on your PC. They’ll also translate audio into English from more than 40 languages.

And they’ll finally relieve one digital headache we all share: That moment when you desperately need to track down the cool thing you recently saw on your PC. Often, you spend way too long searching for that nugget of digital gold in your file folders or on the websites you visited or – gasp – in your stack of emails.

Copilot+ PCs can quickly find that wayward tidbit with Recall, a feature that lets you virtually access snapshots of everything you’ve seen on your PC, as if you have a photographic memory. To protect your privacy, the snapshots are yours and stay locally on your PC. You can delete individual snapshots. And you can filter apps and websites from ever being saved. 

To break all that new ground, Windows developers reimagined the PC, starting with powerful new silicon that can perform 40+ trillion operations per second. With AI at the center of it all, Copilot+ PCs mark the most significant change to the Windows platform in decades.

Meanwhile, all across the AI frontier, the lingo is shifting as quickly as the tech. Such as this fresh phrase: “frontier models.” It refers to large-scale systems so advanced, they sometimes surprise us with what they can achieve.

At Microsoft Build, we announced new frontier models that allow developers to explore multimodal capabilities supporting text, images, video and other types of data in their AI applications.

For example, Phi-3-vision – the newest member of Microsoft’s family of small language models – provides the ability to input images and text and receive text responses. It’s optimized for personal devices and available in Azure.

And there’s GPT-4o, OpenAI’s new flagship model, which reasons across audio, vision and text in real time. It’s now available in Azure AI Studio and as an API.

Two other reveals from Build will help streamline your work in the moment:

• Microsoft Fabric, a cloud-based suite of data analytics tools, now helps developers and customers build intelligent apps by ingesting and querying data in real time for instant insights. A new Fabric module called Real-Time Intelligence can, say, trigger an alert to production managers when equipment is overheating. That capability is now in preview.
• Team Copilot will help facilitate your work meetings, manage your meeting agendas and take notes along the way. Coming in preview later this year, Team Copilot also will expand the AI muscle of Microsoft for Copilot 365 by acting as a collaborator in chats – surfacing relevant info, tracking action items and addressing unresolved issues.

In all, some 60 new products and solutions were unveiled at Build.

BYO … AI?

Holy unread emails!

Nearly 70% of workers say they’re struggling to keep up with the rising pace of their jobs, according to Microsoft’s latest Work Trend Index. In fact, the typical worker reads four emails for each one they send.

To survive this desktop deluge, three in four workers are now using AI to complete tasks. In 2024 alone, Copilot has been used 4.4 million times to synopsize email threads.

When it comes to AI, however, there’s a disconnect between employees and leadership. Most leaders are worried their organizations lack a plan to implement AI while 78% of workers take their own AI tools to work. That trend – dubbed Bring Your Own AI (BYOAI) – endangers company data.

“Organizations need to ride the wave of employees’ AI energy and channel that energy into business impact,” the Work Trend Index suggests.

Meanwhile, there’s good news for employees craving a career change. Most leaders say they’re worried about attracting enough talent to fill key roles. That’s sparking a huge opportunity for workers willing to skill up on AI. “In this moment,” the report says, “fortune favors the bold.”

Watts: The big idea

Picture Paris in the evening. Now imagine that across the City of Light, every light (along with every machine) is powered for two years only by renewable electricity. That would require 23.6 million megawatt hours of renewable electricity – energy from sources that regenerate and can be sustained indefinitely, like solar and wind power.

That’s not entirely theoretical. Microsoft used that same amount of renewable electricity during its 2023 fiscal year as the company continued its work to become carbon negative, water positive and zero waste all by 2030.

According to its latest Environmental Sustainability Report, Microsoft is working to reduce its direct and indirect carbon emissions to near zero by boosting energy efficiency, decarbonizing its operations and by reaching 100% renewable energy by 2025 – all big goals the company does not take, well, lightly.

Honey, I shrunk the AI

Good tech often comes in small packages – and many times, it arises from the smallest moments.

Just ask Ronen Eldan, a machine learning expert with Microsoft Research. One night, while reading to his 4-year-old daughter, he wondered: How much can an AI model learn, using just the words a child can understand?

His question led fellow Microsoft researchers to build a discrete dataset spanning 3,000 basic words. The team then asked a large language model (LLM) to create a children’s story by plucking one noun, one verb and one adjective from their list. They fed that same prompt to the LLM millions of times, generating millions of children’s stories.

Finally, they leveraged all those tiny tales to train small language models (SLMs), a lightweight version of LLMs, designed to perform simpler tasks while requiring less computational power – and fewer resources – than LLMs.

Their work spawned the Phi-3 family of open models, the most cost-effective SLMs available. Not exactly child’s play. But inspired by a child, nonetheless.

We hope this issue inspires you to build your own story of tech discovery!

Between issues, follow the Microsoft News and Stories LinkedIn page for the latest company news, or visit us at Microsoft Source to learn about people doing extraordinary things with technology.