One thing I do enjoy a lot is text to speech and I think that this is an example where statistical models have helped to make TTS more natural sounding and therefore more useful.

Don’t get me wrong: The AI voices do get intonation as well as decisions of what to read out loud and how (“Roman 11 Jinping” instead of Xi Jinping) wrong all the time, but it is still a boon in my book, because I wouldn’t be able to consume some of the longer form content without it.

But with TTS I can listen to interesting articles while walking the dog, cooking or doing rote tasks at work. That’s pretty great.

When it comes to longer form (semi-)academic articles, I have noticed it can work as a “first pass”. I frequently have to re-read passages before I would claim I have consumed that content. Nonetheless: it’s a great supplemental way to consume content while doing something else.