Yeah, the “new” Siri described in this article is not the Siri I’ve been interacting with. Basic speech to text is still flawed; I have to enunciate slowly and clearly with no background noise or it will make absurd mistakes that would be obvious if the context of the prompt and the phone were considered. Adding a reminder to a list is now slower and more error prone than with the old Siri, like you said. I have to sit and delicately hold the phone or watch to avoid disturbing it while it thinks, lest it decide that I’ve moved on and give up on its task. I tried to push it a little by asking for the title of the last podcast episode I listened to; it started talking about a podcast I’d never heard of nor even pulled up on the phone before garbling the second half of its sentence and then going silent.

I suspect it's regressed because they actually don't have good e2e evals comparing the two experiences to test the regression and UX. Like the simplest evals you want is just a shit ton of STTs across human speech accounting for dialects, accents etc and then figuring out how the distribution of errors has moved or increased on the axis you care about. That they likely have. But then you have to integrate that into the product to capture the entire funnel of errors, which I suspect is the part they don't have. The stacks are not "vertically integrated" to test quickly.

Dictation being so terrible is truly wild to me and I don't understand it. On my M5 MBA literally any of the vibe coded local dictation apps that come up (and there are like, hundreds of them!) are better than the built in macOS dictation. They are much faster and the AI text cleanup is so much better than whatever the Apple foundation models are doing.