Windows ML

Yeah, I mean, local, it's, um, it's just incredible, and this week I think is a special week, too.

Microsoft and NVIDIA announced the, you know, a real, a, a partnership around the new Windows ML stack and the new RTX Spark-grade hardware, which is you're basically getting a petaflop in a, in a desktop computer.

And the-

Explain what the RTX is for the audience in case they don't know.

To make, uh, running AI agents safe on Windows, Microsoft created that execution environment, MCX, uh, which is now generally available.

Uh, app, they have i- in fits and starts improved and then ignored, um, Windows ML.

They've done that again with this.

They're g- They're, "Okay, this is how you should do it." Um, the, the little asterisks I gave about there being no local AI that would benefit anyone, it will, uh, actually apply to Copilot+ PCs too.

Thinking of someone that can be rude.

Zoe, for example, wouldn't have to learn how to install, uh, Ollama or, uh, Windows ML or LM Studio.

It will be built into the operating system itself and, and know when to use the right task for using that local model, which is, you know, that's what it should be really.

Really excited about that.

Also, Microsoft talked about this new thing, which is not really a new thing, but they call it hybrid intelligence.

Now, this is going to be based off, I believe, like Windows ML.

Which, here's my bit of concern related to this.

One, as a person who's actually developed an application, used Windows ML, and then the sands underneath shifted in front of me because Microsoft moved things around with these co-pilot runtimes, APIs, and everything else, trying to use a non-device NPU with Connection Explorer.

Whew.

Um...

Like when was the last time anybody thought about Jeff? I'll tell you what though, we've already integrated it for certain things within clairvoyance.

But that's my point is that this stuff is moving so rapidly fast and forward.

Um, DeepSeek V4 Flash, um, which runs in 60 gigabytes of RAM, I think is already available.

Um, there's the Windows ML, which is kind of like the runtime that scales across all the different processor types you have in these SOCs, blah, blah, blah.

You know, uh, llama.cpp is coming to Windows ML, uh, which is gonna bring more, you know, open source-

You couldn't run it in Windows before? No.

Well, I think it's gonna be...