I haven't found an LLM that gives correct responses often enough to be consistently useful with typed requests, let alone spoken ones.

Really ?! I don't think I've ever had a case where Claude has given a response that wasn't helpful relative to whatever I was asking, certainly not for cases where I'm just trying to use it, vs probing for shortcomings.

Noways I only interact with Siri via voice, and all these companies have excellent voice recognition - at least as good as your phone keyboard typing accuracy.

That’s wild. What kinds of things do you ask about?

I fine all of the modern LLMs to be very very good, with some errors but no worse than would turn up in a google search.

That's a pretty low bar, at least over the last several years. I feel like any time I ask Google something uncommon, it presumes a meant something else, and is hard set on answering that question, no matter how many quotes or minus symbols I add to the search.

It is a mistake to use modern Google like a Boolean search engine of old. You almost have to use natural language to get decent results.