I went back to iOS recently, and with AirPods or CarPlay, I found Siri to be a really pleasant surprise. If I need something done, like a text message sent or a quick calculation done, I can just ask without taking my hands or eyes off what I’m doing. It’s also gotten a lot better at taking the hot word + command + arguments without breaks. “hey siri, send a message to dan I’m here” gets you “to dan: I’m here, send it?” For CarPlay, “hey siri, navigate to xyz restaurant” works great to start the navigation if you change your mind about where you’re going once you’re already on the road.
Combine that with “announce notifications” and it really feels like the future, “dan says, are you on your way” and it listens for a few seconds, where you can just say “reply yep I’m 15 minutes away”.
They’re already bringing AI into it a bit, if you get a picture it will try to describe what’s happening in the picture “dan sent a photo of a piece of paper with text on it” which is interesting but it’s not getting enough detail in the descriptions to be of much use just yet. A good multimodal model could definitely improve that, though it must be noted I prefer a worse on-device model to a server based model that can collect a bunch of data.
I think for really off the wall questions/brainstorming, going out to an LLM would yield better results than the current behavior of pointing you towards Google in the browser. “hey siri, what are the typical speeds people achieve on a SPI bus” would be nice if it could give some answer even if it’s not always 100% perfect, it’s kinda like bouncing questions off a slightly unreliable coworker, but you know, better than nothing when you can do it hands/eyes free.
I also have a hope that AI (although maybe not specifically LLMs) would help these assistants handle more complex tasks, like “add eggs to my grocery list” where your grocery list is just in a note and not in some special place. It doesn’t feel out of the question given current technology, I just think nobody’s done it yet.
I have the opposite experience with Siri. Maybe it’s my Australian accent. I have HomePods around, and I use them for lots of stuff. But the voice assistant is barely useful.
“Hey siri, turn on the amp.” “Ok, your media will play louder.” “Siri Set alarm for 4:18” “I’ve set an alarm for 8 o’clock” “hey siri cancel alarm” - radio silence. (Listening to podcast, and my alarm goes off) “hey siri stop” (the alarm stops, so does the podcast). “Hey siri play” “ok, here is some media” (random music plays) … uuurrrrrgghhhh.
Combine that with “announce notifications” and it really feels like the future, “dan says, are you on your way” and it listens for a few seconds, where you can just say “reply yep I’m 15 minutes away”.
They’re already bringing AI into it a bit, if you get a picture it will try to describe what’s happening in the picture “dan sent a photo of a piece of paper with text on it” which is interesting but it’s not getting enough detail in the descriptions to be of much use just yet. A good multimodal model could definitely improve that, though it must be noted I prefer a worse on-device model to a server based model that can collect a bunch of data.
I think for really off the wall questions/brainstorming, going out to an LLM would yield better results than the current behavior of pointing you towards Google in the browser. “hey siri, what are the typical speeds people achieve on a SPI bus” would be nice if it could give some answer even if it’s not always 100% perfect, it’s kinda like bouncing questions off a slightly unreliable coworker, but you know, better than nothing when you can do it hands/eyes free.
I also have a hope that AI (although maybe not specifically LLMs) would help these assistants handle more complex tasks, like “add eggs to my grocery list” where your grocery list is just in a note and not in some special place. It doesn’t feel out of the question given current technology, I just think nobody’s done it yet.