Fuck AI

You need to be more precise. Assistant was using traditional voice recognition technology, which is "AI" but has been around a lot longer than LLMs have and doesn't have any of the same problems. It's tried and true.

Voice controls have long been a beneficial technology both for navigating hands-free as well as an assistive technology.

AI is an entire field of computer science, and purpose-built models have many practical applications and are much, much better than shoehorning everything under the sun into generative language models.

The primary issue of LLMs is everyone trying to use it for things it was never intended to do. They are a decent model of human language, and can produce text that is a reasonable facsimile of text a human might produce. Not semantically, but in form. They are not a search engine or source of knowledge or reasoning, as so many try to use them.

But despite this mismatch, AI companies just try to brute force the problem away with scale by creating astoundingly massive models from all of the data on the internet in an attempt to encode "knowledge" present on the internet in a very incidental way. This makes for something that can sometimes incidentally produce that encoded information, which fools people into thinking it's "intelligent", when it is, of course, anything but. It's a very roundabout and unreliable way of doing things, which is why it's so incredibly dumb that people have lost their fucking minds over it.