The first Copilot+ PCs made their debuts about 13 months ago (June18, 2024). I’ve been waiting to see some low-hanging AI fruit from that stock ever since, mostly with only ho-hum results. But lately, it looks like those PCs are gaining some useful capabilities accessible to ordinary mortals (like your humble correspondent). Indeed, I just read an Ashwin story for Ghacks with great interest. Entitled Microsoft rolls out a bunch of AI features for Windows 11, it shows Copilot AI-driven Setting helpful, given the right prompts. You can see some visual evidence in the lead-in graphic. Note: MS has a cool demo video about this on YouTube.

What Makes Copilot AI-driven Settings Helpful?

The impetus here is to ask Settings for help and information to address specific problems or issues. After monkeying with cursor size on one of my Copilot+ test PCs, I realized  I’d made it too big. So I prompted “cursor too big.” You can see what popped up before I hit the Enter key. Pretty helpful, and going in the right direction.

After I hit Enter, things got more focused and even more helpful. Here’s what showed up (including my overlarge cursor positioned at far right).

These are just the controls I needed to see, with the “Size” item the very thing I needed to fix my problem. That got me started on trying problems or issues in settings, to see where AI would take me. In most cases, it took me right where it would do some, and often the most, good. That’s not good, that’s great!

What Else Ya Got?

The afore-linked Ghacks story  provides further discussion of AI-driven facilities in Click-to-Do, Snipping Tool, Copilot Vision, and more. Could be that spending some time digging in might be rewarded. After all the hoopla and hype around AI of late, I’m glad to see something that’s helpful and potentially useful that’s also easy to understand and implement. Good-oh.

I’ll know that AI is really on my side in a constructive way when I can say to it “Examine these files and give me a specific analysis” (where that analysis depends on the contents) and get something useful back right away. AI can do that right now, of course, but it still leaves users responsible for the set-up and framing to put everything in context. When it can handle that part, too, then we’ll really have something.

Vital Thanks and Shout-Outs

First, I’d like to thank Ashwin from Ghacks for sharing his article. It’s what encouraged my to take the AI-driven Settings facility for a spin. Muchos gracias, mi amigo!

Second, I’d like to thank Jeff Witt and Amanda Heater in the Lenovo Reviews org for providing me with ongoing parade of test PCs. It’s been going on for years and years now and has been a great working relationship. Right now I’ve got TWO (2!) Copilot+ PCs for testing: a 2024 ThinkPad T14s Gen 6 (Snapdragon X1E-78-100 CPU) and a 2025 Yoga AIO 32ILL10 (Intel Ultra 7 258V) with a gorgeous 31″ display. They’re giving me the opportunities I need to learn and dig more deeply into Copilot+ features and functions. Thanks, thanks, thanks.

Very interesting. I was checking WU this morning on the peachy-keen Lenovo AIO 9i the company sent last month. Seems there’s an update available for a local NPU focused language model called Phi Silica (see lead-in graphic). Seems it’s focused on handling SLMs (Small Language Models) on behalf of Windows 11. It drives the OS’s new “AI facilities,” such as Recall, Click-to-Do, and so forth. As MS updates Phi Silica for Copilot+ PCs, I decided to dig in and learn more… So I asked Copilot, and it told me a LOT.

After MS Updates Phi Silica for Copilot+ PCs, Then?

Turns out there’s a special Update History section for this kind of thing, as the lead-in graphic shows. it’s called “AI Component Updates” and it indicates that Phi Silica has already been updated twice on this machine.

Phi Silica is a small-language model (aka SLM) purpose built for Copilot+ PCs. It runs as well as it can on such a PC’s Neural Processing Unit (aka NPU). It lets such models execute locally without having to use a cloud-based back end to do the heavy lifting.

Here’s how Copilot itself describes Phi Silica: It’s “a 3.3 billion parameter model, derived from Phi-3-mini, optimized for speed, accuracy and low power usage.” It “runs directly on the Snapdragon X Series NPU” (and obviously also their AMD and Intel counterparts, because the AIO 9i is an Intel PC), ” enabling fast private and offline AI tasks.” Phi Silica is what’s behind Click-to-do, on-device rewrite and summarization in Word and Outlook, and Windows Recall.

What Can Users Do with Phi Silica?

Quite a lot, as it turns out. When I asked Copilot if I could use my source files for the hundreds of books and thousands of articles and stuff I’ve written as the base for my own SLM, it said “Sure.” Then it pointed to the GitHub-based, open-ource Phi Cookbook as a good place for me to get started. I’m not sure I’m ready to go there, but it’s nice to know that door is open on Copilot+ PCs to someone with the time, wit and energy to make such a thing happen.