That's my point - I expect it to get incrementally better between major releases[1]. That doesn't seem to happen - at least not on the response handling. Voice recognition seems to improve slightly, but that's the nature of having lots of training data.

[1] On new functionality that requires resources installed on the phone (the new sports scorecard things, for example), I understand that only happening on new releases. When I say "take me home" and it only starts navigation sometimes, it clearly has the ability to start nav based on something, so I expect that to happen more reliably.