Next Big Windows Update Will Bring Hardware-Accelerated AI (zdnet.com)

Mary Jo Foley, writing for ZDNet: Every tech vendor these days is quick to slap the AI label on products and services. Up until today, I thought Microsoft had done an admirable job in refraining from doing this with Windows. But the shark has been jumped as of March 7, the company's latest Windows Developer Day. Cue the eye rolls. Microsoft is telling developers that the next release of Windows 10, which we are still calling by its codename, "Redstone 4," will enable developers to "use AI to deliver more powerful and engaging experiences." Microsoft execs say there's now an AI platform in Windows 10 that enables developers to use "pre-trained machine learning in their apps on Windows 10 devices."

Comments

more powerful and engaging experiences

[sinij]

"more powerful and engaging experiences" is a euphemism for spying on users.

Hardware acceleration?

[nmb3000]

Microsoft execs say there's now an AI platform in Windows 10 that enables developers to use "pre-trained machine learning in their apps on Windows 10 devices."

That's not hardware acceleration, because you need, ya know, specialized hardware for that which you can't send via a software update. In fact, the word "hardware" isn't even in the linked article, so where did this silly headline even come from?

Sounds reasonable

[EndlessNameless]

Per the article, they will let developers train their AI in Azure and then import directly into applications. Training a neural network is exactly the kind of limited-duration, CPU-heavy activity that the cloud is designed for. Borrow a thousand CPUs to knock it out in short order and get on with your work.

And imagine if you wanted to train an algorithm with different inputs to see which method yields the best results in your application. You can burn through the training process in parallel in the cloud quickly, and then start building packages for testing immediately. You can iterate faster to fine tune things once you've picked the best baseline training. Without paying for an expensive AI "render farm" up front. The idea is promising, although the devil is always in the details.

And, obviously, any decent hardware-level support for AI would be great. The article only refers to the Azure integration though, so it appears the Slashdot headline is misleading.