Depends what you mean by "smart assistant". Spew a bunch of trivia on a random subject? Sure. Understand your speech commands better? Maybe. Actually manage your stuff (meetings, reminders)? Nope.
ChatGPT is good at making up pieces of text that come very very close to what you wanted (more so in the spirit rather than the letter of it), but it's not very good at anything else.
ChatGPT has potential to create meeting invitations. I asked it:
Today is 2023-08-07. I’m in Tokyo. Draft a calendar invitation for coffee meeting at Starbucks in Shibuya during lunch time on Wednesday. Answer in ics format.
It answered:
BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//ChatGPT//EN
CALSCALE:GREGORIAN
BEGIN:VEVENT
SUMMARY:Coffee Meeting at Starbucks Shibuya
LOCATION:Starbucks Shibuya, Tokyo
DTSTART:20230809T120000
DTEND:20230809T130000
DESCRIPTION:Let's meet up for a coffee and chat at Starbucks in Shibuya during lunchtime.
END:VEVENT
END:VCALENDAR
Requires a bit of prompt engineering to make it work, but this can probably be improved with some fine tuning. All you need is something that detects and parses ics invitation in the output.
Sure it can output .ics but the question is, would you trust it to manage your calendar? To get every single meeting right?
Software at its core is a precision tool. When it goes wrong it was either made wrong or you used it wrong. But ML introduces another layer for which the software maker doesn't want to take responsability, and you the user have no idea if you're using it right. When it fails there's "nobody to blame" yet you're the one left holding the bag.
With fundamentally flawed software like this (ambiguous by design) it's fun to play (especially if it seems to make sense 80-90% of the time), but it really sucks to have to depend on it for actually important stuff.
Depends what you mean by "smart assistant". Spew a bunch of trivia on a random subject? Sure. Understand your speech commands better? Maybe. Actually manage your stuff (meetings, reminders)? Nope.