Interestingly, that kind of parallels the real world too: if you want a quick and high level answer, talk to someone in person; if you want something detailed and info-dense, get them to write it down.

Turning advanced voice still leaves "regular" voice interaction which are actually (for me:) much much better - it's just the regular response, verbalized :). Voice quality isn't worse, it just doesn't try to summarize in one casual sentence.

(I still hate that the voice is getting more and more "natural" - the umms and ahhs and weird pauses)

The AI attached to their voice chat is running a completely different model. Ask any question and you quickly realize it is completely, unapologetically lobotomized. If you want to talk to it about how you feel after your gf/bf broke up with you, it is fine. If you want to ask it something about tunneling machines and how tunneling through different types of rock impact engineering decisions, it is going to skim the first four sentences of some blog article and then defend whatever hill it has chosen until it dies, regardless of what the larger body of work on the topic says. OpenAI's voice chat being so bad and being totally divorced from their SOTA models is largely why I cancelled my subscription. I am tempted to wire up piper/whisper and the OAI api to get back what I actually want/need. But today you cannot have a conversation about engineering questions and get anything close to factually reliable answers out of it.