Hallucination is a specific technique. Of course an AI like chatgpt can hallucinate as a side effect, but with fine-tuning and a specialized tool / context, the hallucination falls to zero. So, if you train an AI specifically to set alarms and / or use a specific toolset, you can make it damn reliable.
You are correct in theory, but the point is that Apple hasn't been doing it. The point is not about generative AI, but APPLE making generative AI when they've made Siri a moron. So it's useless to state that it can be "damn reliable" when Apple hasn't been making it reliable.
Siri often doesn't even understand very regular commands for tasks it's specifically designed and marketed to do. This is common knowledge. Now we keep assuming that Apple will fix things, but they haven't fixed Siri. In fact, by many accounts Siri has gotten worse these several years.
If you tell it to do something, sometimes it will do exactly that, but sometimes it will do something entirely unrelated. It cannot even get speech to text right. The issue does not exist, or does not exist to nearly the same extent, in Google Assistant and Amazon Alexa, or even Dragon NaturallyFreakingSpeaking.
Often it will hear you correctly and still do something unrelated. Last night I told it to play a song in three different ways. Each time Siri heard me correctly as it displayed the text of what I said. It did three different things except play the song. I tried naming the song, and the song and the album, and the song and which playlist it's in. Finally I had to pick up my iPad and tap into the app to do it myself.
Admittedly, Siri has been pretty good at setting alarms, at least for me.
Now, without any rational basis for saying this, and with a decade's worth of evidence to the contrary, I still assume that Apple will make a capable generative AI. But certainly not soon.
