I've been wanting something like this ever since GPT-2 was a thing. Language models seem like a pretty natural fit for voice assistants. It's just a question of actually building one, and of being comfortable enough with all the new failure modes that enables to actually release the product despite that.
Hopefully the existence of ChatGPT has popularized the idea enough that we'll start to see companies actually implementing this in short order. Or at the very least, open source solutions taking advantage of LLM APIs outcompeting proprietary assistants from companies too terrified of the idea of Alexa or Siri being caught on camera saying something racist to implement a system where there's even a remote possibility of that happening.
Hopefully the existence of ChatGPT has popularized the idea enough that we'll start to see companies actually implementing this in short order. Or at the very least, open source solutions taking advantage of LLM APIs outcompeting proprietary assistants from companies too terrified of the idea of Alexa or Siri being caught on camera saying something racist to implement a system where there's even a remote possibility of that happening.