1) Making it say silly things about hiding dead bodies in ravines.
2) Trying to make it text things, only to give up and type the message by hand. Usually done while driving, and usually spanning the course of around a minute.
It's not even remotely close to something usable yet, at least not from what I've seen. Just look at how people use it, they don't even understand [or maybe they just don't like?] the interface for it!
Most people I see address it by name, like you would the computer on star trek. They hold down the button, and then say:
"Siri, where can I get a blah blah blah nearby?"
Which usually doesn't return a useful result. In this time I'm usually searching on google maps, and telling them the instructions before they've finished their attempt at an interaction.
You don't invoke the service by addressing it. The service was invoked by pressing the button.
Seriously I have yet to see a person who uses siri for something useful.
It is much, much quicker to say "remind me about X tomorrow at 6pm" than unlock screen, find the app, open it up, create event etc.
- change playlists when when I'm working out or biking
- set the alarm when I go to bed
- reminders (a lot)
- the weather
- looking up something on wikipedia and imdb
- and sending short sms messages
I'd say my use has grown over the last few weeks as I'm biking a lot more. There was an initial trust issue, and I still sometimes have this weird feeling of being on the spot when I'm dictating to Siri, but otherwise best "app" on the phone for me.
But, here we are, with people praising Siri. It's how technology works-- every input modality is useful for some tasks, and at some level of maturity. Dismissing technological potential based purely on past technology is short-sighted. [For examples of how Pen+touch is a good idea see http://billbuxton.com/Pen+TouchUIST.pdf. Add type, and it could be powerful indeed]