Personally, I think they should focus on edge on-device AI, because cloud servers will just add to costs, and they'll probably have to charge us more because of that. I also believe they should make it voice-based, which means making voice recognition a lot better, because ChatGPT blows Apple out of the water with that. They should also make Siri sound better, because again, ChatGPT blows them out of the water with that. I also believe while you're talking to Siri, they should be able to generate images and videos and have actual conversations with you, and I also think instead of the AI having all these amazing capabilities, we should have to download capabilities from the App Store. That way, the AI is not over-capable, only knows what we want it to know, and it saves Apple's cost of moderating it, and it provides a better experience because it only has those capabilities that you download and you don’t have to worry about it making a mistake
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.
What do you think Apple should do with AI, and how should it implement it?
- Thread starter TheApplenator
- Start date
- Sort by reaction score
While I agree on all of what you write I wonder how much it cost in terms of CPU power and batterydrain.
I know there was reports of big steps being made on "on-device" AI, but I wonder if that is enough.
Another question is if this will be offered or can be offered on older devices? I do like the idea of "downloading abilities", that way maybe they could offer some parts of AI to older devices.
I know there was reports of big steps being made on "on-device" AI, but I wonder if that is enough.
Another question is if this will be offered or can be offered on older devices? I do like the idea of "downloading abilities", that way maybe they could offer some parts of AI to older devices.
And when it comes to older devices, the only difference is in the chip capabilities. Unless the iPhone 16 or 17 has something entirely new. Well, the new chips and the ray tracing, but I'm not sure if that would affect generative AI or not. It could, probably. But I'm assuming older devices will at least have some capabilities. And the newer ones might have more, but if they do have more, it would be because the hardware limitations.
Of course, there would be concerns about battery drain, but if that was happening, then it's the user's fault for using the AI so much. Like any other thing, the more you use your phone, and the more power that consumes, the more your battery is going to drain. That's just how it works. There's really nothing that can be done about that, unless, and I'm sure they will, the only thing Apple can really do is optimize the system where it can scale easily from using a lot of power to using very little power when you aren't using it.
Absolutely, its more that if Apple provides something that "built-in" and it would drain the batteries you get a ********* from forums like this with all sorts of weird conspiracy theories. But if they release it as an add-on, app style or have a basic package (that shows in battery if its being used) and more that can be downloaded I'm sure they could get away with it just as with other apps.
I mean, I'm talking about the artificial intelligence being built into iOS still. I just mean, allow developers to build apps and submit them to the App Store that would add capabilities to the AI, if that makes sense. Whatever Apple didn't add to it naturally or add with an update. That way, it would allow Apple to expand on it without doing the work themselves. And it would also give the AI more capabilities and it would also allow Apple to limit what the AI was capable of by making you download additional capabilities. Which means less misinformation and less of a chance that the AI could lie or anything because that app would have to be approved by Apple before it could be on the App Store to add additional capabilities to the AI. And you could limit what the AI does on its own because you can download additional capabilities from the App Store.
Ah, now I get what you mean. Not a bad idea, provide the AI plattform basics in the OS, AI APIs to developers and keep some sort of control of what can be downloaded. Also give the user the option if it should be enabled or not just like Siri works today.
Not sure how that would apply to the new alternative store idea that's happening in the EU, but hopefully they will keep some sort of control of it or could potentially choose not to offer certain features to the alternative store.
The way Apple set it up, the alternative stores basically operate like the App Store does. The same rules apply or Apple won't approve the apps from the alternative stores and allow them to be downloaded. It's basically like submitting them to the App Store, but a different App Store.
Absolutely not unless Apple trains it on their own data. There are already way too many issues about algo-generation based on tons of stolen artwork as it is. Not just images, but text and people's likenesses and voices.
Not violating copyright is a great start, so train it on licensed materials. Not hallucinating fake information is also great. Train it on data that is not poisoned with bias, discrimination and misinformation. Not breach a user's privacy. Be completely transparent with how their AI is trained and deployed. This one worries me the most, since Apple has a bad track record with transparency.
Ideally, no generative text and image stuff. One area I can see it being useful is to make Siri less useless by parsing natural speech commands more competently and enabling more sophisticated, chained commands.
E.g. "Hey Siri, I'm going for a run. Put the new [podcast name] on and remind me to get some cash if I go near an ATM" — it should start an outdoor run, queue up the specific podcast to play when my headphones are in, search for ATMs near me based on my running ability, and set a proximity based reminder.
Maybe it could even lend itself to better activity tracking on the watch with more advanced analysis of watch data.
But basically anything that's not the pointless garbage that every other "AI" tool is doing these days would probably be a good start.
E.g. "Hey Siri, I'm going for a run. Put the new [podcast name] on and remind me to get some cash if I go near an ATM" — it should start an outdoor run, queue up the specific podcast to play when my headphones are in, search for ATMs near me based on my running ability, and set a proximity based reminder.
Maybe it could even lend itself to better activity tracking on the watch with more advanced analysis of watch data.
But basically anything that's not the pointless garbage that every other "AI" tool is doing these days would probably be a good start.
Agree 100%
No need for fancy stuff until Siri can do the simple things right.
I asked this question:
"Siri, when is my flight to Greece?"
Answer:
And yet, it's been in my Apple Calendar ever since I made the booking in November.
As it is right now, Siri is unable to connect the dots.
I actually agree with you here. Siri should be able to connect all the information on your device while doing it on-device to give you information that you need and to do things on-device.
Maybe they could develop something where... Siri can automatically train itself based on your own information, but do it all on-device without any of it reaching Apple servers or anything.
First, Improve current Siri.. By leaps and bounds.
Sure they can replace the current model of Siri with some cutting edge on-device "AI" Siri replacement. However that would put such a processing load on devices just for basic tasks. While also limiting it to newer devices. Their MLX does sound impressive, it would also add a lot of wear on an iPhones NAND.
There's no excuse why they can't update Siri to be contextually aware of everything under your iCloud. iCloud content both saved/cached locally and stored remotely on an iCloud server. Without relying on LLM/GenAI. Siri can do content recognization in a random photo and describe it to me, but can't give me details on a calendar event? Unless I use very specific phrasing. It can't figure it out on its own. That alone should not require A server farm of A200 cards.
I'm a firm believer that Siri lost the assistant battle only because it didn't have the sheer amount of user data to train on. Unlike Google and its "Your data is our Data". It wasn't until maybe a few iOS versions ago that a user can opt-in to providing Dictation and Siri queries snippets back to Apple. Even today that option is buried in the Settings app.
Apple should make a ChatGPT competitor and improve on-device MLX and other LLM generative AI. While also using it to improve current Siri. Offering the newer significantly improved Siri as a base and using MLX and other LLM techniques for more complex user interactions. And why the hell not offload the rare too complex on-device scenarios to a server farm where a larger model can parse and respond.
While they're add it... maybe develop something that would remove PII on-device. Scramble voice pitch etc and upload these now anonymous interactions for training on a server somewhere.
Sure they can replace the current model of Siri with some cutting edge on-device "AI" Siri replacement. However that would put such a processing load on devices just for basic tasks. While also limiting it to newer devices. Their MLX does sound impressive, it would also add a lot of wear on an iPhones NAND.
There's no excuse why they can't update Siri to be contextually aware of everything under your iCloud. iCloud content both saved/cached locally and stored remotely on an iCloud server. Without relying on LLM/GenAI. Siri can do content recognization in a random photo and describe it to me, but can't give me details on a calendar event? Unless I use very specific phrasing. It can't figure it out on its own. That alone should not require A server farm of A200 cards.
I'm a firm believer that Siri lost the assistant battle only because it didn't have the sheer amount of user data to train on. Unlike Google and its "Your data is our Data". It wasn't until maybe a few iOS versions ago that a user can opt-in to providing Dictation and Siri queries snippets back to Apple. Even today that option is buried in the Settings app.
Apple should make a ChatGPT competitor and improve on-device MLX and other LLM generative AI. While also using it to improve current Siri. Offering the newer significantly improved Siri as a base and using MLX and other LLM techniques for more complex user interactions. And why the hell not offload the rare too complex on-device scenarios to a server farm where a larger model can parse and respond.
While they're add it... maybe develop something that would remove PII on-device. Scramble voice pitch etc and upload these now anonymous interactions for training on a server somewhere.