AI... advantage? What advantage? Apple Intelligence doesn't exist yet. Its likely to be 2026 before we see the most significant features released. Customers by-and-large do not care about any of these features, whether built by Apple or Google. How much did Apple pay for this article?
1) Vertical integration, which enables end-to-end hardware (unified CPU/GPU/NPU memory, custom flash controller, etc.) and software optimization 2) focus on on-device and private processing (vs. personal and business data leakage through cloud AI), and 3) ability to roll out incremental updates with high uptake.
Microsoft and Google also have vertical integration on their own hardware.
But I think the "secret AI advantage" may be Apple's marketing advantage where it can tout its upcoming vertically integrated AI and software features to compensate for minimal hardware upgrades for the iPad mini (which is still a nice tablet even with "only" 8GB of RAM and a smartphone SoC.)
iPad Mini’s (and iPhone’s) 8gb memory will be very limiting. It’s sufficient for 3b models quantized to 4bits, but puts significantly more powerful 8b models just out of reach.
This isn’t a huge issue now, but with another 12 months of local AI progress it’ll leave a competitive opening for device-makers who ship more memory. (See Meta’s TPO paper for significant improvements to Llama, pending release.)
I skipped this iPhone cycle despite being on Apple’s iPhone upgrade program — a memory bump to the Pro line would’ve easily justified an upgrade.
I still dont understand why they name the chip it uses as A17 Pro when it has one GPU less that A17 Pro used on iPhone. They could have just call it A17.
Not only a Pro SoC in the Mini, it is also using the rare N3B rather than N3E. I guess Apple has committed to a minimum N3B usage from TSMC for them to continue. AFAIK Apple is the only one using it.
The nice thing is that the repaid transition to whole lineup being AI ready is that 8GB is finally standard on Smartphone, and hopefully at least 12GB or 16GB on Mac.
I assume the chips in this thing are a bunch of lower-binned chips that were otherwise destined for iPhones. It's not uncommon for them to ship binned variants under the same name. They ship 8- and 10-core GPU variants of the base M3, and iirc the entry-level M1 MacBook Air had 7 GPU cores while the next config up had 8.
> approx two years being others in AI response quality
I bet this doesn't wind up mattering much in the grand scheme of things. Not for Apple's vision anyway. They want things to work reliably, not some gimmick. A lot of what I see in Google's ads is features no one really wants, like unfettered image generation (cool, but it doesn't really help many people in practice).
> A lot of what I see in Google's ads is features no one really wants
That's true of the vast majority of AI features from every company; its not a problem unique to Google. No one cares about any of this; its stockholders making products to sell to other stockholders.
Yeah, I'd agree with that. I just think Apple is clearly pitching one thing to investors - "Apple Intelligence is here now and is basically AGI!!!" and another thing to customers - "Apple Intelligence is not available for awhile until we figure out the quirks, will mostly handle "basic" NLP style tasks such as summarization of notifications". Playing both sides of it.
Sure; but I feel that its a bit unfair to poo poo Google's strategy when, at least, they have a strategy; and there are AI features Google has released that I've actually seen normal people mention once or twice. Their AI call handling thing and the "ask Gemini to summarize a youtube video" feature come to mind.
Bringing up Google's imagegen features as something no one wants is reasonable: But Apple is releasing a copy-cat app under the Apple Intelligence banner. Is Genmoji a gimmick? That's an Apple feature, not Google.
So, again: I think the assertion that "Apple is late but does it right" is totally invalid when applied to AI, because there is no way to do it right, because no one wants it. You can at least be first and capture some vaporware market hype, but Apple is going to have a come-to-jesus moment next year when they have little new to announce, because they're still releasing features they pre-announced in 2024.
>So, again: I think the assertion that "Apple is late but does it right" is totally invalid when applied to AI, because there is no way to do it right, because no one wants it.
No one wants "Artificial Intelligence," but it's unclear if they want the features that it will drive -- i.e., "Apple Intelligence." This is something that can't be known for sure until the features are released and the companies that make them get analytics on them. For now, the very conservative set of features that comprise "Apple Intelligence" are reasonable in that they easily slot into the workflow of users by enhancing or making easier things that people already do. The fact that they are not impressive like what Google or Meta is doing is only important to people who either live their life on the cutting edge of technology or are tech investors. For the people buying phones every 3-5 years (as opposed to every 1 or 2), the story is yet to be written.
I think Apple has a hard line to walk here. They have project progress to investors so the stock will continue to rise, and they have to be the thoughtful curator of features that the customers demand -- that second impulse is what gave rise to the reputation that drove the stock originally.
Apple Intelligence really should have been kept in the oven until WWDC '25, because it's not clear that any of them really require external developers to make happen. But Apple had to keep the impression that they are 'doing something with AI,' and now they're stuck doing what they typically haven't done: promising things that clearly aren't ready. This puts them in the same league as, i.e., Google. Google labelled Gmail as "Beta" for half a decade (April 1, 2004 to July 7, 2009), and now has a reputation for shipping quarter-baked, basically-Alpha projects and then killing the ones that don't catch on.
I think Apple's "problem" is that their vision doesn't usually end up being very revolutionary. As a musician I really see this in the iPad - Apple assumed that having a headphone jack and USB-MIDI would make the iPad a breakout success between musicians. Their vision didn't exactly line up with reality though, people wanted custom drivers and desktop plugins that Apple wasn't willing to encroach on with their vision. As such, serious musicians don't really shackle themselves to the iPad unless they want to challenge themselves.
Same goes for the iPad with AI. What are people going to use it for? Not for coding, which seems to be the most opaque application we've seen yet. You won't do research or play with cool repos you find, since Metal is a second-class citizen everywhere you go. Do... you stream responses from OpenAI to your most serious questions? Generate clipart of a 7-fingered salesman for your sales pitch? Use it to SSH into an Nvidia box for your serious AI-based workloads?
At this point I'm getting really exhausted of Apple's vision. I'd much rather they implement a serious interface like ONNX or Vulkan instead of giving people toys and expecting them to treat it like a tool. Nobody wants to work on Apple platforms when they're this blatant about their EEE intentions.
Or, its the same as Apple Vision Pro, that device everyone has forgotten about. Their MO has always been: We don't build products for a customer profile, we build the products we want to build and expect customers to come to us. On the surface level, one problem Apple has had for a few years and I expect will continue to have is, that "we" in there used to mean "intelligent and respected product minds" but today means "supply chain experts and investors".
But on a deeper level, that strategy used to work because there was so much green field in the world of computing. Its easy to sell a vision to someone with no expectations. The iPhone was released into a world where it didn't even have copy/paste, and no one cared, because very few people had anything remotely like it. Vision Pro was released into a world where we already have sixteen Apple devices, I use my computers every day for work and play, I expect things of them, and releasing with a narrative of "it replaces other devices X Y and Z" actually maybe could never work, even if it could replace those devices, because I kinda like my X Y Z devices. Not only that, but it was released into a world where VR already exists, and has been available at your local Walmart for a couple hundred bucks: Apple never got to control the narrative for VR, so they couldn't communicate their vision.
> Same goes for the iPad with AI. What are people going to use it for?
I don't think it's necessarily Apple's fault that there haven't turned out to be a whole lot of genuinely productive or groundbreaking use cases for generative AI.
With regard to your other issues, it sounds like you just don't like Apple's vision. But that's not really relevant to how they make use of generative AI and probably belongs in a broader discussion about Apple.
My take on it, in any case, is that if Apple decides you're not an important market segment then you're not an important market for them. That's all there is to it. They may occasionally waver and indicate that e.g. triple-A gaming is coming, etc. etc. But really it's better to count on them not caring much about niche or even popular markets if it interferes with their broader goals.
Calling Apple's vision "revolutionary" is not the wording I would use. More like "highly opinionated, well-polished software with hardware that's just a bit ahead of the curve".
I don't trust Apple's software acumen to be opinionated in a way that puts it ahead of the curve. I got rid of my Mac for this reason - Homebrew and third-party distributors fight against Apple to get their apps to behave the way they want them to between minor updates.
On top of that, Apple's first-party offerings haven't proven at all that they can stand on their own. Metal is entirely ignored as a 3D API for serious applications and it's most impressive demos are Japanese gachapon games for iPhone. Their hardware isn't competitive with Nvidia despite their advanced access to TSMC nodes and their software isn't even remotely gunning to replace CUDA or adopt a similar featureset a-la Apple's own (retired) OpenCL. They have given up on anything that doesn't entail a) locking in their users or b) making recurring revenue from services. They spurned Khronos, abandoned cross-platform API support and stopped competing on stats... for what? Why?
I think it's fine if you support this on a personal or ideological level because it doesn't bother you. But you also have to admit that the average user won't ever spend more than 5 minutes with AI features because they buy iPads to watch YouTube and porn, not design sports cars and rocket engines. Apple is building features for a market that doesn't exist while entirely neglecting rational uses of their time and development talent. It's absolutely pitiful to see a business as large as Apple fall that far - they deserve everything antitrust regulation has to give.
> They want things to work reliably, not some gimmick.
Siri is still hot garbage far too often. On CarPlay, “send text message to [partner]” “I can’t find [partner] in your contacts”, often followed ironically by “new text message from [partner]”
And there's lots of gossipy talk about who is "beloved" and who isn't. Like who has time to care about these things if you're not directly reporting to them?
AI... advantage? What advantage? Apple Intelligence doesn't exist yet. Its likely to be 2026 before we see the most significant features released. Customers by-and-large do not care about any of these features, whether built by Apple or Google. How much did Apple pay for this article?
1) Vertical integration, which enables end-to-end hardware (unified CPU/GPU/NPU memory, custom flash controller, etc.) and software optimization 2) focus on on-device and private processing (vs. personal and business data leakage through cloud AI), and 3) ability to roll out incremental updates with high uptake.
Microsoft and Google also have vertical integration on their own hardware.
But I think the "secret AI advantage" may be Apple's marketing advantage where it can tout its upcoming vertically integrated AI and software features to compensate for minimal hardware upgrades for the iPad mini (which is still a nice tablet even with "only" 8GB of RAM and a smartphone SoC.)
iPad Mini’s (and iPhone’s) 8gb memory will be very limiting. It’s sufficient for 3b models quantized to 4bits, but puts significantly more powerful 8b models just out of reach.
This isn’t a huge issue now, but with another 12 months of local AI progress it’ll leave a competitive opening for device-makers who ship more memory. (See Meta’s TPO paper for significant improvements to Llama, pending release.)
I skipped this iPhone cycle despite being on Apple’s iPhone upgrade program — a memory bump to the Pro line would’ve easily justified an upgrade.
https://archive.ph/UPOg6
I still dont understand why they name the chip it uses as A17 Pro when it has one GPU less that A17 Pro used on iPhone. They could have just call it A17.
Not only a Pro SoC in the Mini, it is also using the rare N3B rather than N3E. I guess Apple has committed to a minimum N3B usage from TSMC for them to continue. AFAIK Apple is the only one using it.
The nice thing is that the repaid transition to whole lineup being AI ready is that 8GB is finally standard on Smartphone, and hopefully at least 12GB or 16GB on Mac.
I assume the chips in this thing are a bunch of lower-binned chips that were otherwise destined for iPhones. It's not uncommon for them to ship binned variants under the same name. They ship 8- and 10-core GPU variants of the base M3, and iirc the entry-level M1 MacBook Air had 7 GPU cores while the next config up had 8.
The Pro designates more than just the core count. If you look at the A18 vs A18 Pro, they’re not simply binned versions of each other.
The Pro includes features like better display controllers, thunderbolt support and more video codec silicon.
So it’s more fair to say this is a binned A17 Pro than what would fit if they removed the Pro nomenclature.
Summary is they’ve rolled out AI chips to their devices quickly but are approx two years being others in AI response quality.
> approx two years being others in AI response quality
I bet this doesn't wind up mattering much in the grand scheme of things. Not for Apple's vision anyway. They want things to work reliably, not some gimmick. A lot of what I see in Google's ads is features no one really wants, like unfettered image generation (cool, but it doesn't really help many people in practice).
> A lot of what I see in Google's ads is features no one really wants
That's true of the vast majority of AI features from every company; its not a problem unique to Google. No one cares about any of this; its stockholders making products to sell to other stockholders.
Yeah, I'd agree with that. I just think Apple is clearly pitching one thing to investors - "Apple Intelligence is here now and is basically AGI!!!" and another thing to customers - "Apple Intelligence is not available for awhile until we figure out the quirks, will mostly handle "basic" NLP style tasks such as summarization of notifications". Playing both sides of it.
Sure; but I feel that its a bit unfair to poo poo Google's strategy when, at least, they have a strategy; and there are AI features Google has released that I've actually seen normal people mention once or twice. Their AI call handling thing and the "ask Gemini to summarize a youtube video" feature come to mind.
Bringing up Google's imagegen features as something no one wants is reasonable: But Apple is releasing a copy-cat app under the Apple Intelligence banner. Is Genmoji a gimmick? That's an Apple feature, not Google.
So, again: I think the assertion that "Apple is late but does it right" is totally invalid when applied to AI, because there is no way to do it right, because no one wants it. You can at least be first and capture some vaporware market hype, but Apple is going to have a come-to-jesus moment next year when they have little new to announce, because they're still releasing features they pre-announced in 2024.
>So, again: I think the assertion that "Apple is late but does it right" is totally invalid when applied to AI, because there is no way to do it right, because no one wants it.
No one wants "Artificial Intelligence," but it's unclear if they want the features that it will drive -- i.e., "Apple Intelligence." This is something that can't be known for sure until the features are released and the companies that make them get analytics on them. For now, the very conservative set of features that comprise "Apple Intelligence" are reasonable in that they easily slot into the workflow of users by enhancing or making easier things that people already do. The fact that they are not impressive like what Google or Meta is doing is only important to people who either live their life on the cutting edge of technology or are tech investors. For the people buying phones every 3-5 years (as opposed to every 1 or 2), the story is yet to be written.
I think Apple has a hard line to walk here. They have project progress to investors so the stock will continue to rise, and they have to be the thoughtful curator of features that the customers demand -- that second impulse is what gave rise to the reputation that drove the stock originally.
Apple Intelligence really should have been kept in the oven until WWDC '25, because it's not clear that any of them really require external developers to make happen. But Apple had to keep the impression that they are 'doing something with AI,' and now they're stuck doing what they typically haven't done: promising things that clearly aren't ready. This puts them in the same league as, i.e., Google. Google labelled Gmail as "Beta" for half a decade (April 1, 2004 to July 7, 2009), and now has a reputation for shipping quarter-baked, basically-Alpha projects and then killing the ones that don't catch on.
I think Apple's "problem" is that their vision doesn't usually end up being very revolutionary. As a musician I really see this in the iPad - Apple assumed that having a headphone jack and USB-MIDI would make the iPad a breakout success between musicians. Their vision didn't exactly line up with reality though, people wanted custom drivers and desktop plugins that Apple wasn't willing to encroach on with their vision. As such, serious musicians don't really shackle themselves to the iPad unless they want to challenge themselves.
Same goes for the iPad with AI. What are people going to use it for? Not for coding, which seems to be the most opaque application we've seen yet. You won't do research or play with cool repos you find, since Metal is a second-class citizen everywhere you go. Do... you stream responses from OpenAI to your most serious questions? Generate clipart of a 7-fingered salesman for your sales pitch? Use it to SSH into an Nvidia box for your serious AI-based workloads?
At this point I'm getting really exhausted of Apple's vision. I'd much rather they implement a serious interface like ONNX or Vulkan instead of giving people toys and expecting them to treat it like a tool. Nobody wants to work on Apple platforms when they're this blatant about their EEE intentions.
Or, its the same as Apple Vision Pro, that device everyone has forgotten about. Their MO has always been: We don't build products for a customer profile, we build the products we want to build and expect customers to come to us. On the surface level, one problem Apple has had for a few years and I expect will continue to have is, that "we" in there used to mean "intelligent and respected product minds" but today means "supply chain experts and investors".
But on a deeper level, that strategy used to work because there was so much green field in the world of computing. Its easy to sell a vision to someone with no expectations. The iPhone was released into a world where it didn't even have copy/paste, and no one cared, because very few people had anything remotely like it. Vision Pro was released into a world where we already have sixteen Apple devices, I use my computers every day for work and play, I expect things of them, and releasing with a narrative of "it replaces other devices X Y and Z" actually maybe could never work, even if it could replace those devices, because I kinda like my X Y Z devices. Not only that, but it was released into a world where VR already exists, and has been available at your local Walmart for a couple hundred bucks: Apple never got to control the narrative for VR, so they couldn't communicate their vision.
> Same goes for the iPad with AI. What are people going to use it for?
I don't think it's necessarily Apple's fault that there haven't turned out to be a whole lot of genuinely productive or groundbreaking use cases for generative AI.
With regard to your other issues, it sounds like you just don't like Apple's vision. But that's not really relevant to how they make use of generative AI and probably belongs in a broader discussion about Apple.
My take on it, in any case, is that if Apple decides you're not an important market segment then you're not an important market for them. That's all there is to it. They may occasionally waver and indicate that e.g. triple-A gaming is coming, etc. etc. But really it's better to count on them not caring much about niche or even popular markets if it interferes with their broader goals.
Calling Apple's vision "revolutionary" is not the wording I would use. More like "highly opinionated, well-polished software with hardware that's just a bit ahead of the curve".
I don't trust Apple's software acumen to be opinionated in a way that puts it ahead of the curve. I got rid of my Mac for this reason - Homebrew and third-party distributors fight against Apple to get their apps to behave the way they want them to between minor updates.
On top of that, Apple's first-party offerings haven't proven at all that they can stand on their own. Metal is entirely ignored as a 3D API for serious applications and it's most impressive demos are Japanese gachapon games for iPhone. Their hardware isn't competitive with Nvidia despite their advanced access to TSMC nodes and their software isn't even remotely gunning to replace CUDA or adopt a similar featureset a-la Apple's own (retired) OpenCL. They have given up on anything that doesn't entail a) locking in their users or b) making recurring revenue from services. They spurned Khronos, abandoned cross-platform API support and stopped competing on stats... for what? Why?
I think it's fine if you support this on a personal or ideological level because it doesn't bother you. But you also have to admit that the average user won't ever spend more than 5 minutes with AI features because they buy iPads to watch YouTube and porn, not design sports cars and rocket engines. Apple is building features for a market that doesn't exist while entirely neglecting rational uses of their time and development talent. It's absolutely pitiful to see a business as large as Apple fall that far - they deserve everything antitrust regulation has to give.
> They want things to work reliably, not some gimmick.
Siri is still hot garbage far too often. On CarPlay, “send text message to [partner]” “I can’t find [partner] in your contacts”, often followed ironically by “new text message from [partner]”
This really isn't just Siri though. My car's built in nav tried to direct me to Mexico (I'm in CT), rather than a restaurant one town over.
Turns out it had the restaurant name backwards too. Barn Thai, rather than Thai Barn.
Both a data entry problem, and an interpretation problem. Why it thought I really wanted to drive 1300 miles over instead of 10 is beyond me.
And there's lots of gossipy talk about who is "beloved" and who isn't. Like who has time to care about these things if you're not directly reporting to them?