Want to boost your productivity? Microsoft Edge now has built-in AI features, like text summarization! Get started now. #MicrosoftEdge #AIinBrowser #EdgeAI
Explanation in video
AI Right in Your Web Browser? Microsoft Edge is Making it Happen!
Hey everyone, John here! You know how Artificial Intelligence, or AI, seems to be popping up everywhere these days? Well, get ready for something pretty neat: AI tools are starting to be built right into things we use every day, like our web browsers! Today, we’re going to talk about how Microsoft is bringing some clever AI magic into its Edge browser. And don’t worry, if any of this sounds like rocket science, my trusty assistant Lila is here to ask the questions you might be thinking!
First Things First: What’s the Big Deal with AI Anyway?
You’ve probably heard of AI tools like ChatGPT. They can write, summarize, and even chat with you. These are often powered by what we call Large Language Models. They’re super powerful, but they also need a LOT of computer power to run.
Lila: “John, hold on a sec! What exactly are these ‘Large Language Models,’ or LLMs, you mentioned? And when you say they need a lot of computer power, what does ‘compute, power, and cooling of a modern inferencing data center’ actually mean?”
John: “Great questions, Lila! Think of a Large Language Model as a GIANT brain that’s been trained by reading billions of pages of text from the internet, books, and more. It learns patterns and how words fit together, so it can understand and generate human-like text. As for ‘inferencing data centers’ – ‘inferencing’ is just a fancy word for when the AI is actually doing its job, like answering your question or writing a summary. Because these LLMs are so big and complex, they usually run in huge buildings packed with powerful computers. These are ‘data centers,’ and they use a massive amount of electricity (‘power’) and generate a lot of heat, so they need serious ‘cooling’ systems to stop them from overheating. It’s like having a thousand gaming PCs running at once in one room!”
Introducing the Little Brothers: Small Language Models (SLMs)
So, those big LLMs are amazing, but they’re kind of like using a giant crane to lift a feather for some tasks. That’s where Small Language Models, or SLMs, come in. These are like the smarter, more efficient little siblings.
Lila: “Okay, so SLMs are smaller. But what makes them ‘small’? And you mentioned they’re ‘trained with fewer parameters’ and on ‘tokens’. That sounds a bit technical, John!”
John: “You’re right, Lila, let’s break that down! SLMs are ‘small’ because they are trained on less data than the giant LLMs, and they have fewer ‘parameters’. Think of parameters as tiny little dials or settings inside the AI’s brain. An LLM might have trillions of these dials, while an SLM might have ‘only’ a few billion. Fewer dials mean it’s smaller and needs less power. And ‘tokens’? Imagine you’re teaching a kid to read. You break words down into smaller pieces, like ‘cat’ is ‘c-a-t’. Tokens are similar for AI; they are common pieces of words or sentences that the AI learns to understand and use. So, an SLM like Microsoft’s ‘Phi-4-mini-instruct’ (that’s its name!) is trained on billions of these tokens, which is a lot, but still less than the biggest AIs.”
The cool thing about these SLMs is that they can run on much less powerful hardware, like your personal computer, or even what some call ‘edge hardware’.
Lila: “Edge hardware? Is that like the edge of my desk?”
John: “Haha, not quite, Lila! ‘Edge computing’ or ‘edge hardware’ means processing data closer to where it’s being generated, instead of sending it all the way to a distant data center. So, in this case, it means running the AI right on your PC or a small local server, rather than on some massive computer far away. This makes things faster and can be better for privacy.”
Why Put AI in Your Web Browser?
Microsoft has been working hard on these SLMs. The idea is to have AI tools available right on your PC. But sometimes, getting everyone to download and install new AI models can be tricky. Plus, you want these AI tools to work the same way for everyone, no matter what computer they have.
Lila: “You mentioned ‘PC-based inferencing architecture’ and using ‘ONNX runtimes with GPUs and NPUs’. That sounds like a mouthful!”
John: “It does, doesn’t it? Let’s simplify. ‘PC-based inferencing architecture’ just means the setup or structure that allows the AI to do its thinking (‘inferencing’) on your personal computer. ‘ONNX runtimes’ are like a universal translator that helps AI models work on different types of computer chips. And ‘GPUs’ and ‘NPUs’? A GPU (Graphics Processing Unit) is the part of your computer that usually handles graphics for games and videos, but it turns out they’re also really good at AI math! An NPU (Neural Processing Unit) is a newer type of chip specially designed to make AI tasks run even faster and more efficiently. Some new PCs, called ‘Copilot+ PCs’, are starting to include these NPUs.”
So, what’s a great place where these AI tools could live and be easily accessible? Your web browser! We spend so much time in our browsers – writing emails, filling out forms, reading articles. Microsoft thinks it’s the perfect spot, and they’re starting by building AI into their Edge browser.
Lila: “And how do they make the AI talk to the browser? You said something about ‘APIs’?”
John: “Exactly! An API (Application Programming Interface) is like a special menu that one piece of software provides so other software can ‘order’ up its services. In this case, Edge will provide AI APIs. This means web developers can write code that tells the AI in Edge to do things like summarize text, without needing to become AI experts themselves.”
Meet Phi-4-mini: Edge’s New Built-in AI Helper
Microsoft is testing a new feature in the experimental versions of Edge (called Dev and Canary builds) that includes one of these SLMs, specifically the Phi-4-mini model, right inside the browser. This is super convenient because users won’t have to mess around with complicated setups.
Lila: “The article mentioned developers won’t need to set up ‘WebNN or WebGPU, or even WebAssembly’. Are those other ways to run AI in a browser?”
John: “Yes, Lila. Those are web technologies that can help run more complex applications, including AI, directly in web pages. WebAssembly is like a way to run super-fast code in your browser. WebGPU and WebNN (Web Neural Network API) are newer technologies aimed at letting web pages use your computer’s GPU or NPU for heavy-duty tasks like AI. But what Microsoft is doing with Edge aims to make it even simpler for developers by providing ready-to-use AI functions.”
The Perks of Having AI Right in Your Browser
This approach has some really nice advantages:
- Better Privacy: Because the AI runs locally on your computer, your personal data (like the text you want summarized) doesn’t have to be sent over the internet to a big company’s server. It stays with you. This is a big win for privacy!
- Saves Money: Using those big cloud-based AIs can get expensive. Running it locally in your browser means you (or the website developer) don’t have to pay for that cloud computing time for these specific tasks.
- Works Offline (Mostly!): Once the AI model is downloaded by the browser, it can often perform its tasks even if you’re not connected to the internet. Super handy!
- Easy Updates: The browser itself takes care of downloading the AI model and updating it when new versions are available. You don’t have to worry about it.
Lila: “So, the browser ‘hosts the model’? And developers use ‘JavaScript APIs’ to make it work? Can you explain that a bit more?”
John: “Sure! ‘Hosts the model’ just means the browser stores the AI program (the model) and runs it. And JavaScript is the main programming language used to make websites interactive. So, website developers can use JavaScript commands – these are the ‘APIs’ we talked about – to tell the AI model inside Edge what to do, like ‘summarize this paragraph’ or ‘help me rewrite this sentence’.”
What Can This In-Browser AI Do (So Far)?
Right now, in these early test versions of Edge, the built-in AI (using Phi-4-mini) can help with a few text-based tasks:
- Summarizing text: Great for quickly getting the gist of a long article.
- Writing and rewriting text: It can help you draft emails, rephrase sentences, or generate ideas.
- Basic prompt evaluation: You can give it a specific instruction (a “prompt”) and it will try to follow it.
Microsoft also plans to add support for translation services in the future. Imagine your browser being able to instantly translate web pages for you, right there on your device!
Lila: “The original article talked about a ‘Prompt API’ and ‘JSON format constraint schema’ when developers test this. That sounds complicated!”
John: “It can sound a bit techy, but the idea is simple. A ‘prompt’ is just the instruction you give the AI. The ‘Prompt API’ is the tool developers use to send that instruction. Now, ‘JSON’ is just a way to organize data in a simple, text-based format that computers can easily understand. A ‘constraint schema’ for JSON is like giving the AI a very strict template for its answer. For example, if you ask the AI to analyze if a customer review is positive or negative, you can tell it: ‘Your answer MUST be in this JSON format: just tell me the sentiment (positive/negative) and how confident you are.’ This helps make sure the AI’s output is predictable and useful for the application.”
It’s important for developers to write good, clear system prompts and use these constraints. This helps keep the SLM focused and prevents it from giving weird or unhelpful answers. Think of it like giving very precise instructions to a helper to ensure they do the job correctly.
A Quick Look at How Developers Might Use It
For any developers reading, getting started involves enabling some special settings (called feature flags) in the test versions of Edge. There’s even a sample webpage where they can try it out. The browser downloads the Phi model the first time it’s needed.
Lila: “The article mentioned an ‘asynchronous process’ and ‘N-shot prompting’. What are those?”
John: “Good catch! An ‘asynchronous process’ means a task that can happen in the background without freezing up everything else. So, downloading the AI model or getting a response from it can happen while you can still use the browser. ‘N-shot prompting’ is a clever trick: you give the AI a few examples (the ‘N’ shots) of a prompt and the kind of answer you expect. This helps ‘teach’ the AI in that specific moment how you want it to behave for similar tasks, making its responses more accurate and tailored.”
Developers will be able to set a general instruction (a system prompt), provide these examples, and then give the AI user-specific tasks. They can even choose if they want the answer as plain text or formatted text (like Markdown).
What’s Powering This Magic: GPU or NPU?
Right now, it seems this in-browser AI in Edge uses your computer’s GPU (the graphics card) to do its work. This makes sense because many computers have capable GPUs, especially those used by developers who would be testing this first.
Lila: “So, it’s not using those special ‘NPU’ chips you mentioned earlier, even on the ‘Copilot+ PCs’?”
John: “Not yet, it seems. But that’s likely to change. As more PCs come with NPUs (those dedicated AI chips), Microsoft will probably update Edge to use them when available. They have something called Windows ML…”
Lila: “Windows ML? What’s that?”
John: “Windows ML (Machine Learning) is a platform from Microsoft that helps developers run AI models efficiently on Windows devices. It’s smart enough to figure out the best hardware to use on your PC – whether that’s the main processor (CPU), the GPU, or an NPU if you have one. So, in the future, Edge could use Windows ML to automatically pick the best chip for the job, making the AI run even better without developers having to worry about the specific hardware a person has.”
The Big Picture: A Smarter, Safer Web Experience
Having these AI tools built into a trusted platform like your browser, which knows how to securely use your computer’s hardware, is a big step. It means developers can write AI-powered web features once, and they should work for anyone using that browser, regardless of their specific computer setup. This could lead to all sorts of helpful, privacy-respecting AI features becoming a normal part of our web experience.
Our Quick Thoughts
John: “Personally, I think this is really exciting. Making AI more accessible, more private, and integrated into tools we use daily like a web browser is fantastic. It’s like giving everyone a little smart helper for common online tasks, without them needing to be tech wizards. It’s still early days for this Edge feature, but the direction is very promising!”
Lila: “Wow, so the AI can actually work on my computer without always sending my information to the internet for these tasks? That sounds much better for keeping my stuff private! And it’s great when you explain these techy words, John. It makes it all so much clearer!”
What do you all think about having AI tools built directly into your browser? Let us know in the comments!
This article is based on the following original source, summarized from the author’s perspective:
Taking advantage of Microsoft Edge’s built-in AI