What bugs me the most about this is how fixable it is.
Apple is now at a point where even the oldest supported devices on their latest operating systems have either decent dedicated ML hardware or are Intel-based machines with enough compute power that they should really be able to do a much better job with inferencing locally (and the server component is irrelevant to local device capabilities anyway).
It should’ve been possible to bin Siri’s backend and rebuild something better from scratch by now. That Apple is swimming in money just makes things like this more jarring.
It’s honestly my greatest gripe with Apple as a software company. They put a lot of effort into polishing what they release but seemingly very little into maintaining it afterwards. macOS is the worst offender, where basic and fundamental things stay broken for years on end (settings not being applied after selection in the UI, permissions being set in the UI but not actually taking effect despite being set, continuity stopping working silently in FaceTime and iMessage. It’s gotten so bad I just accept that some former tentpole features no longer exist because they’ve become so broken and no-one cares to fix them).
Back to Siri though, it’s gotten substantially worse than when I used to use it on an iPhone 4S. That’s a decade plus of consistent regression while Google Assistant continues to progress. It’s so bad, and can’t be replaced, that I wouldn’t be surprised if it’s a leading cause of people exiting the eco-system altogether.