There is no AI that Apple could add that would even tempt me to upgrade.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Spending a 'Tremendous Amount of Time and Effort' on AI, Details Coming 'Later This Year'
- Thread starter MacRumors
- Start date
- Sort by reaction score
There is no AI that Apple could add that would even tempt me to upgrade.
it's not really the AI features themselves that will drive the upgrade. It's that they'll make it so unoptimized for older iPhones that the old phones feel slow and reduced battery life that will force you to update. And they'll make it so you can't disable those new features to keep your device running great.
Siri has been TERRIBLE for a decade. I have little faith they’ll get this right.
Microsoft investing in ChatGPT was the right move. I use it multiple times per day. Apple’s focus on Vision caused them to completely miss the boat here.
Now they are way behind.
I’m a huge Apple stan too.
Microsoft investing in ChatGPT was the right move. I use it multiple times per day. Apple’s focus on Vision caused them to completely miss the boat here.
Now they are way behind.
I’m a huge Apple stan too.
A mix between the rabbit and that halo pin? I like the idea of trusting Apple as the central hub to connect all my apps and services. Would make for a really great assistant (what Steve kind of promised) … but on my phone or watch
Well, maybe with all this AI push, Tim can finally get Siri to be closer to Google Assistant. I’ve been running 9 original HomePods, and 3 Google Nest Home hubs, and by far Google Assistant is more useful. 🤷♂️
Well for those clamouring for AI what is it you want the phone to do besides improving from “I found this on the Internet.”
Are you expecting Tony Stark’s Jarvis?
Are you expecting Tony Stark’s Jarvis?
Not sure this is a good comparison. Alexa is absolute garbage. I have 3 Echo Shows and 4 HomePods and I use them with a narrow focus on what they can do well, like controlling my home, setting reminders and timers and playing music. That generation of voice assistants were built with very strict commands and are absolutely horrible when compared with LLM's.
Alexa is in a league of its own in how bad it is because it sets expectations with all of its "Skills", none of which I can ever remember how to invoke when I need them. Siri at least is more flexible in how I can ask for things, but I know its limitations and set my expectations accordingly.
If Siri moves into generative AI as it seems it will, holy crap, having that kind of natural conversational assistant tied into the ecosystem that most of us have assembled between HomePods, iPhones, iPads, Macs, Watches, AirPods, this is going to tremendously useful in a way that chatGPT can't because it's not built into hardware ecosystems.
Doesn't need to be on-device to respect users' privacy.
Siri, for instance, most of the times already uses servers to elaborate a reply, yet privacy is paramount.
Wether you'd like to have generative AI on device so that it's gonna be super fast, that's another story 😉
I'm cautiously optimistic. I'm confident that AI is great at pretending. Pretending to know what you're talking about. Example: Amazon has new AI-powered review summaries. The AI reads through thousands of review on a product, and produces a summary of those reviews in a short paragraph. Seems great at first, until you read them closely and realize the AI has no understanding of it at all...it's just collating frequency into a summary, and repeating it. The summaries will often cite concerns or praise that make absolutely no sense in context, because they misunderstand what people are talking about.
AI Siri may be able to pretend to have a conversation with you, but will it be able to translate what you're asking into useful commands? Will it continue to fail to respond for inexplicable reasons? Will it continue to perform actions you didn't ask for?
I don't think generative AI is important for Siri. I think comprehension of commands is important. I think speed of response and execution of commands is important.
AI Siri may be able to pretend to have a conversation with you, but will it be able to translate what you're asking into useful commands? Will it continue to fail to respond for inexplicable reasons? Will it continue to perform actions you didn't ask for?
I don't think generative AI is important for Siri. I think comprehension of commands is important. I think speed of response and execution of commands is important.
To state the obvious: Training a LLM that is close to capability of GPT-4 is something Apple surely can deliver. They would need access to talent, training sets, and Nvidia GPUs. None of these are out of the reach for Apple. Furthermore, they can build high scale services for inference and integrate model to their products.
Where Apple is not great is fast paced iteration in public. That is against their culture: They want to ship a product when it is done. Unfortunately AI training does not work like that - for it one needs a large number of users giving feedback. With this, I would expect Apple to still be able to ship GPT-4 -level LLM, but lag 1-2 years behind the frontrunners on the market (namely OpenAI and secondly Google). While not exciting, this might be acceptable.
In a couple of places Apple has a huge advantage:
(1) Seamless vertical integration and UX. Siri is already always present. It just needs to become less dumb. I would expect Apple to become better than anyone else in this due to having control over 2B devices people use. It does not matter if GPT-5 is a bit smarter if Apple has baked GPT-4 -level LLM access to device price and it is already omnipresent.
(2) Local inference. Apple has included inference HW into their silicon for the past 7+ years. I would expect them to be best positioned to ship local "good enough" LLM in a smartphone. This will work offline, retain privacy, and be instant to respond. We should expect Mistral 7B level capabilities and thus would still need to allow routing harder queries to cloud. Interesting to see how UX challenge for this is solved.
(3) Reputation for privacy. Many people trust Apple over anyone else for actually trying to protect their privacy. This is due to building the reputation for many years and having a business model that is aligned with privacy.
With all this, I would not bet against Tim Cook when he says that they have something seriously cool cooking.
Where Apple is not great is fast paced iteration in public. That is against their culture: They want to ship a product when it is done. Unfortunately AI training does not work like that - for it one needs a large number of users giving feedback. With this, I would expect Apple to still be able to ship GPT-4 -level LLM, but lag 1-2 years behind the frontrunners on the market (namely OpenAI and secondly Google). While not exciting, this might be acceptable.
In a couple of places Apple has a huge advantage:
(1) Seamless vertical integration and UX. Siri is already always present. It just needs to become less dumb. I would expect Apple to become better than anyone else in this due to having control over 2B devices people use. It does not matter if GPT-5 is a bit smarter if Apple has baked GPT-4 -level LLM access to device price and it is already omnipresent.
(2) Local inference. Apple has included inference HW into their silicon for the past 7+ years. I would expect them to be best positioned to ship local "good enough" LLM in a smartphone. This will work offline, retain privacy, and be instant to respond. We should expect Mistral 7B level capabilities and thus would still need to allow routing harder queries to cloud. Interesting to see how UX challenge for this is solved.
(3) Reputation for privacy. Many people trust Apple over anyone else for actually trying to protect their privacy. This is due to building the reputation for many years and having a business model that is aligned with privacy.
With all this, I would not bet against Tim Cook when he says that they have something seriously cool cooking.
I'm expecting something more like HAL from 2001: Space Odyssey
the biggest update for Siri would be it no longer telling me “here’s what I found on the web for …”
That is surprising. I uses the faces feature in photos quite a lot and find it very effective. It does a pretty good job of finding faces, particularly when I go through the faces it thinks might be and rule some out. I have several dozen names faces in my photo library. It often find someone’s face in a photo where I would have missed it. It does have trouble distinguishing two of my nieces but then I have a hard time telling which is which in many photos, too.
I‘m not sure that Apple intends to just turn Siri into a ChatGPT-style answerbot. That is not really their style and doesn’t enhance their devices or the software in them. If they do try to do something like that with Siri, they are more likely to license some large bodies of text and use that as a legal, and controlled data source. That is what Adobe is doing for the source images for their AI image generator. Apple might license news from known sources like NYT, AP news, and others and use data sources like Wikipedia.
I think it is more likely that Apple would use AI to give Siri a better understanding of the context and intent of our requests and a better understanding of the capabilities of the apps and functions on the device so that when you ask Siri to do something you get a more useful result, even if Siri needs to ask clarifying questions. That would enhance Apple’s offering and provide real assistance to users.
You know, you can offer cloud services and be privacy preserving too… it doesn‘t all have to be on device only to be good and work with their privacy alignments.
That being said, on device LLM are the next logical step in pushing the technology forward imp, so excited to see if they can manage to pull it off.
That’s kinda what I meant by putting more thought into it. I don’t need Siri to draft emails for me, but I’d love to tell my phone whatever in natural language, even slang or my own terms, have it understand me, remember from context that this is how I talk about XYZ feature, and then go do stuff in the background to achieve what I’m asking for or tweaking things to anticipate my needs.
Like, if I’m replying to a text, offer suggestions in my voice with things I usually say. Type like me. If I’m playing media, learn what volume I like to have things at and normalize audio to that. If I’m looking for an ATM, know from what bank apps I have on my phone to prioritize those ATMs.
Im sure some of this doesn’t need AI but I think it illustrates the point of the level of thoughtfulness I want. If I still have to tell my phone to do a bunch of stuff, I might as well be doing it myself.
I think it speaks to the broader arc of Cook's Apple being bean counters rather than innovation drivers. An obvious example of this is discontinuing AirPort and Time Capsule. There was no money in it and the bean counters wanted to sell iCloud storage. They have HomePods because it ties to Apple Music service they can sell. There's no service to sell with HomeKit (other than some half baked security camera storage). Apple could have collaborated with device makers, such as Philips, Leviton, but the bean counters said no.
