Low confidence: "Our engineers work tirelessly to ensure that frontier research doesn't step on the toes of users. It's a deliberate touch to keep people in the drivers seat, emphasizing the importance of human coexistence with technology. Apple Intelligence can never replace our users or the human touch they bring to everyday life, and that's what makes Apple products so special to all our lives."
You realize this isn’t an Apple marketing piece, it’s an extremely well known tech reviewer who used to work for the NYT and was one of the two(?) who was allowed to test the iPhone before its release.
Actually think that’s a good thing for Apple. Compare how much money companies have absolutely shoved into the AI race, and how much circular investment is happening. Apple is one of the few massive players which hasn’t massively extended themselves and taken on a ton of financial risk. How do you make money on frontier models? By selling it to software companies, I guess.
Apple just needs a capable personal AI that respects privacy. They could eventually get that by self-hosting open model inference on their private cloud compute. They don’t need to light billions upon billions of dollars on fire to build data centers and do training to make a frontier model that’s nearly already a commodity. If they’re not selling tokens (and why would I want them to), that approach seems financially stupid.
> If they’re not selling tokens (and why would I want them to)
That’s one of the two things that has stood out most to me. Siri AI answers my questions. And that’s it. She stops.
Google, OpenAI, and Anthropic all constantly try and get you to answer follow-up questions. Because if you do that, you’ll use up more tokens and get more “addicted“ and end up buying a plan.
Apple doesn’t have one. I guess iCloud, but let’s face it, they already pushed that a ton anyway. They’re not trying to get you to use it for the sake of use. It’s not trying to be your friend.
And just like David Pogue says, it doesn’t lie. I just asked it an incredibly stupid question. Can you get pregnant from a picture?
“While I'm not a doctor, no, you cannot get pregnant from looking at a picture of someone. Pregnancy requires physical contact and the transfer of sperm to an egg. For accurate information regarding reproduction, consult a healthcare professional.”
Look at that! It immediately says it’s not a doctor. It gives a good answer, and then it says you need to seek out real advice to be sure things are accurate. Of the summer, I also got responses that ended with something along the lines of “I am an AI, please check information for accuracy.”
I’ve gotten incorrect answers, but they come with disclaimers and it doesn’t act like some omnipotent god that is desperate to be your friend.
David Pogue is right, it’s very impressive. He and basically everyone else make the same comment, and I fully agree: it’s hard to remember to even use it.
I used Siri a ton before, but always for basically the same four commands. Create reminder, call X, send a text to Y, set a timer for Z minutes.
And after years of that, it’s hard to remember to do more. But when I do it can often shock me.
It’s not fast. Which is quite annoying for those quick actions I used to do. But I’m OK with that on my phone. Not happy with it on my watch.
It’s not ChatGPT 5.6. And I’m sure they’ll get it faster over time. But in my mind they finally delivered.
I was just rambling about it to my wife - it's finally good and it finally can understand me correctly.
Apple always had access to app's internal API for AppleScript and accessibility, so now that they exposed it to an LLM (with upcoming claude and gemini support), they are a serious power. And iPhones are in the hands of a lot of folks around the globe. Exciting.
I seriously doubt they’re going to give anyone else access. They’re fighting the EU over it. But I guess we’ll see.
More than once has has told me something, and I had to ask how it knew. And it would tell me exactly how and what do you know it was there somewhere on my phone.
11 comments
[ 2.4 ms ] story [ 23.2 ms ] threadI like how this is cast as a selling point instead of tacit admission that Apple is not on the frontier.
But hey perhaps i still don't understand your point properly
Low confidence: "Our engineers work tirelessly to ensure that frontier research doesn't step on the toes of users. It's a deliberate touch to keep people in the drivers seat, emphasizing the importance of human coexistence with technology. Apple Intelligence can never replace our users or the human touch they bring to everyday life, and that's what makes Apple products so special to all our lives."
High confidence: https://variety.com/2024/digital/news/why-apple-ipad-ad-cont...
Apple just needs a capable personal AI that respects privacy. They could eventually get that by self-hosting open model inference on their private cloud compute. They don’t need to light billions upon billions of dollars on fire to build data centers and do training to make a frontier model that’s nearly already a commodity. If they’re not selling tokens (and why would I want them to), that approach seems financially stupid.
That’s one of the two things that has stood out most to me. Siri AI answers my questions. And that’s it. She stops.
Google, OpenAI, and Anthropic all constantly try and get you to answer follow-up questions. Because if you do that, you’ll use up more tokens and get more “addicted“ and end up buying a plan.
Apple doesn’t have one. I guess iCloud, but let’s face it, they already pushed that a ton anyway. They’re not trying to get you to use it for the sake of use. It’s not trying to be your friend.
And just like David Pogue says, it doesn’t lie. I just asked it an incredibly stupid question. Can you get pregnant from a picture?
“While I'm not a doctor, no, you cannot get pregnant from looking at a picture of someone. Pregnancy requires physical contact and the transfer of sperm to an egg. For accurate information regarding reproduction, consult a healthcare professional.”
Look at that! It immediately says it’s not a doctor. It gives a good answer, and then it says you need to seek out real advice to be sure things are accurate. Of the summer, I also got responses that ended with something along the lines of “I am an AI, please check information for accuracy.”
I’ve gotten incorrect answers, but they come with disclaimers and it doesn’t act like some omnipotent god that is desperate to be your friend.
David Pogue is right, it’s very impressive. He and basically everyone else make the same comment, and I fully agree: it’s hard to remember to even use it.
I used Siri a ton before, but always for basically the same four commands. Create reminder, call X, send a text to Y, set a timer for Z minutes.
And after years of that, it’s hard to remember to do more. But when I do it can often shock me.
It’s not fast. Which is quite annoying for those quick actions I used to do. But I’m OK with that on my phone. Not happy with it on my watch.
It’s not ChatGPT 5.6. And I’m sure they’ll get it faster over time. But in my mind they finally delivered.
Apple always had access to app's internal API for AppleScript and accessibility, so now that they exposed it to an LLM (with upcoming claude and gemini support), they are a serious power. And iPhones are in the hands of a lot of folks around the globe. Exciting.
More than once has has told me something, and I had to ask how it knew. And it would tell me exactly how and what do you know it was there somewhere on my phone.