It’s one of those transformative little features. But it’s especially great in PDFs where it’s not a “real” PDF but just a collection of page sized images so you couldn’t normally select the text before without OCR software of some sort.

Having it built into the OS is so handy.

I tried it with one of these image-based PDFs and didn't get it to work. Is this Apple-Silicon-only? Or is there any other step you need to do?

It is possible that it is AS only. It may use the machine learning components of AS.

Edit: As of July of 2021, Live Text was available in beta on Intel Macs so by the time Monterey shipped it worked on both Intel and AS.

Whether it works on a given document may depend on how that app displays the image. I can see where it would be complicated in a PDF where you may have combinations of text, images, and OCR’d text behind images.

I’m not going to pretend it works 100% of the time. I don’t know if that’s the issue.

It works with anything that works with Monterey. So if you are running macOS Monterey you fit the requirements.

Particularly useful for Twitter announcements on Twitter, where in order to say anything of value they have to tweet an image of some text.