> We believe in training our models using diverse and high-quality data. This includes
data that we’ve licensed from publishers, curated from publicly available or open-
sourced datasets, and publicly available information crawled by our web-crawler,
Applebot.
> We do not use our users’ private personal data or user interactions when training
our foundation models. Additionally, we take steps to apply filters to remove certain
categories of personally identifiable information and to exclude profanity and unsafe
material.
> Further, we continue to follow best practices for ethical web crawling, including
following widely-adopted robots.txt protocols to allow web publishers to opt out
of their content being used to train Apple’s generative foundation models. Web
publishers have fine-grained controls over which pages Applebot can see and how
they are used while still appearing in search results within Siri and Spotlight.
When Apple inevitably partners with OpenAI or Anthropic, which by their definition isnt doing "ethical crawling", I wonder how I should be reading that.
I mean they also buy from companies with less ethical supply chain practices than their own. I don’t know that I need to feel anything about that beyond recognizing there’s a big difference between exercising good practices and refusing to deal with anyone who does less.
You shouldn't believe Big Tech on their PR statements.
They are decades behind in AI. I have been following AI research for a long time. You can find best papers published by Microsoft, Google, Facebook in past 15 years but not Apple. I don't know why but they didn't care about AI at all.
Apple used to be at the edge of AI. They shipped Siri before "AI assistant" went mainstream, they were one of the first to ship an actual NPU in consumer hardware and put neural networks into features people use. They were spearheading computational photography. They didn't publish research, they're fucking Apple, but they did do the work.
And then they just... gave up?
I don't know what happened to them. When AI breakthrough happened, I expected them to put up a fight. They never did.
One problem with Apple's approach here is that they were scraping the web for training data long before they published the details of their activities and told people how to exclude them using robots.txt
> Using our web crawling strategy, we sourced pairs of images with corresponding alt-texts.
An issue for anti-AI people, as seen on Bluesky, is that they're often "insisting you write alt text for all images" people as well. But this is probably the main use for alt text at this point, so they're essentially doing annotation work for free.
> An issue for anti-AI people, as seen on Bluesky, is that they're often "insisting you write alt text for all images" people as well. But this is probably the main use for alt text at this point, so they're essentially doing annotation work for free.
How did you come to the conclusion that those two groups overlap so significantly?
Gotta polish that fig-leaf to hide Apple's real stance towards user privacy: arstechnica.com/tech-policy/2023/12/apple-admits-to-secretly-giving-governments-push-notification-data/
> Apple has since confirmed in a statement provided to Ars that the US federal government "prohibited" the company "from sharing any information,"
All I can say is, I asked Siri today (verbatim): What is 75 degrees fahrenheit in celsius, and what is 85 degrees in fahrenheit — and it offered a web search about fahrenheit. The "and" completely disabled its most basic ability to do metric conversions.
So, it's nice to see Apple is doing research and talking about it, but we're out here waiting, still waiting, for anything useful to make of it all on our thousand-dollar devices that literally connect us to the world and contain our entire life data. It's what I would've expected from one of the most valuable companies in the world.
A tool that can handle more than one question at a time is useful. Modern LLMs handle that with ease. So it's completely reasonable to be critical of that limitation.
"holding it wrong" was exactly the right phrase given how that phrase was used with the iPhone antenna bridging problem. This is an Apple product failing.
First, most of the English speaking world is not native.
"As of 2022, there were about 400 million native speakers of English. Including people who speak English as a second language, estimates of the total number of Anglophones vary from 1.5 billion to 2 billion."
Second, all popular models I tested did well with that query, including Gemini on Android (aka "ok Google"), except Apple's.
That exact phrase "What is 75 degrees fahrenheit in celsius, and what is 85 degrees in fahrenheit" given to ChatGPT produces the correct result (it infers that the second degrees must be Celsius) and ChatGPT gives me a nicely laid out formula for the math of the conversion.
Your usage of Siri today (probably on an old version of iOS) frankly has nothing to do with the article we are discussing. Sorry to say this but it is going to take time. Comparing the performance of a chatgpt running in a big data center with a model running locally on a phone device... give it a few years.
People have been giving Siri a few years for a decade now. Siri used to run in a data center (and still does for older hardware and things like HomePods) and it has never supported compound queries.
Siri needs to be taken out back and shot. The problem with “upgrading” it is the pull to maintain backwards compatibility for every little thing Siri did, which leads them to try and incorporate existing Siri functionality (and existing Siri engineers) to work alongside any LLM. Which leads to disaster, and none of it works and just made it all slower. They’ve been trying to do an LLM assisted Siri for years now and it’s the most public facing disaster the company has had in a while. Time to start over.
> Your usage of Siri today (probably on an old version of iOS) frankly has nothing to do with the article we are discussing.
Yes, but isn't that infuriating? The technology exits! It even exists, as evidenced by this article, in the same company that provides Siri!
At least I feel that way every time I interact with it – or for that matter my Google Home speaker, ironically made and operated by the company that invented transformer networks.
Despite all the “Apple is evil” or “Apple is behind” (because they don’t do evil). Well, what they made with the Foundation Model is great. The fact that they build a system within the Swift language that allows you to specify structured data models (structs) to be used like any other model in a modern programming language, and you actually get back generated data in that format is great. Unlike a lot of other AIs where you might get back a well formatted JSON after a carefully crafted request, but still you never can’t be sure and need to implement a bunch of safeguards. Obviously it’s still the beginning and other tools might do something similar. But as an iOS developer that makes the usage of AI so much simpler. Especially with the bridge to external AIs that still allows you to map back to the type safe structured Swift models. I try not to be a hater, every progress, even slow or underwhelming at first might lead to improvements everywhere else.
Sorry if I didn’t use the correct terms. Didn’t catch up on all the terminology coming from my native language. ;) But yes, I agree, the fact that parts, different parameters, of the model can be completed asynchronous by streaming the output of the model, is quite unique. Apple/swift was late with async/await, but putting it all together, it probably plays well with the ‘never’ (I know ) asynchronous and reactive coding.
How do you think their implementation works under the hood? I'm almost certain it's also just a variant of "structured outputs", which many inference providers or LLM libraries have long supported.
Huh? Grammar-based sampling has been commonplace for years. It's a basic feature with guaranteed adherence. There is no "carefully crafting" anything, including safeguards.
Every time I see a paper from Apple I just feel like, OK so why isn’t my iPhone actually doing any of this yet?
Why give this to developers if you haven’t been able to get Siri to use it yet? Does it not work or something? I guess we’ll find out when devs start trying to make stuff
Yes, but why do I have to open a third-party app to do these things when Apple, the company that primarily popularized the entire genre of mobile voice assistants, could very feasibly bake all of that into theirs?
I mean, the thing even lets me ask ChatGPT things if I explicitly ask it to! But why do I need to ask in the first place?
> why isn’t my iPhone actually doing any of this yet?
Probably Apple is trying to distill the models so they can run on your phone locally. Remember, most, if not all, of Siri is running on your device. There's no round trip whatsoever for voice processing.
Also, for larger models, there will be throwaway VMs per request, so building that infra takes time.
They just launched "Private Cloud Compute" with much fanfare to enable server-side LLM processing, so between that and the fact that Siri has been server-based for most of its existence (local processing is fairly new), I don't think that's their main constraint at this point.
That said, "Private Cloud Compute" does run on proprietary Apple hardware, so availability might be a concern (assuming they don't want to start charging for it).
I know apple is methodical and don’t show their hand but I cannot help but feel they are releasing all this research because they haven’t integrated any into the phone or provided a compelling AI functionality for their users. This is their only way to say “hey we are good with AI too”.
AFAICT this is the first commercial model trying to be marketed as responsibly-sourced. Love it, but it also seems like the noise around this issue has died down. Is this for legal cover? Or more apple-privacy marketing
"Sorry we are hilariously far behind everyone else in the industry after having made a huge amount of fanfare about 'Apple Intelligence' for years. It's just that we have shot ourselves in the knee to satisfy Bluesky posters and the NY Time's lawyers"
My son (he's 11 years old now and fairly skilled with all the main AI tools, eg chatgpt, gemini, etc) and I retry her every month or so, and this past time we just laughed. Can't handle basic questions - hears the question wrong, starts, stops, takes us to some random ass webpage, etc, etc.
"She's so jacked up!" he said.
Apple needs to get this under control and figured out, stat!
Looks nice. I just wish they’d improve the models behind dictation on both iPhone and Mac to have better accuracy and on the fly multiple language transcription.
I'd really like to be able to use this 3B model on my little 4GB GPU card!
It looks very capable for a reasonable weight.
Maybe one day on HhuggingFace
This isn’t the Apple I remember. Product integration falls apart at every seam, but don’t worry—we’ve got plenty of impressive technical documentation to compensate. I’m sure Jobs would be thrilled to see his ‘it just works’ philosophy replaced with ‘it barely works, but here’s a 50-page PDF explaining why.
The philosophy is the same, and since it was never implemented in the mythical era of Jobs, so is the practice. So he'd be as thrilled as he was back then?
As someone that was around in the days of Apple before bankruptcy, the same, there is no Jobs more around, and is getting back to Gil Amelio kind of Apple.
Tim Cook might be better at squeezing the juice, but he is not a product person.
This time around they need another solution, otherwise regardless of how much money they have, they will stay as the iOS/iPad company, given the relevance of macOS on desktop market worldwide.
It does. You can use it directly on iOS 26 beta - without writing a line of code I can toy with the on-device model through Shortcuts on my 16 Pro. It’s not meant to be a general purpose chatbot… but it can work as a general purpose chatbot in airplane mode which is a novel experience.
It would be interesting to see the tok/s comparison between the ANE and GPU for inference. I bet these small models are a lot friendlier than the 7B/12B models that technically fit on a phone but won't accelerate well without a GPU.
As someone mentioned, this model is available in the beta version of iOS 26; it's also part of macOS 26, iPadOS 26 and visionOS 26. Anyone with a free developer account can install the developer betas; the public beta is expected next week.
There's a WWDC video "Meet the Foundation Models Framework" [1].
> The new Foundation Models framework gives access to developers to start creating
their own reliable, production-quality generative AI features with the approximately
3B parameter on-device language model. The ∼3B language foundation model at the
core of Apple Intelligence excels at a diverse range of text tasks like summarization,
entity extraction, text understanding, refinement, short dialog, generating creative
content, and more. While we have specialized our on-device model for these tasks,
it is not designed to be a chatbot for general world knowledge. We encourage app
developers to use this framework to design helpful features tailored to their apps
There are even already some local AFM to Open AI API bridge project on GitHub - that lets you point basically any Open AI compatible client at the local models. Super nice for basic summarisation and completions.
The more I think about Apple, the more I realize that Apple is so far behind. While other companies are pushing the envelope (OpenAI, Anthropic, Google ..) Apple's ambitions seem much much smaller.
And this is after they made very big claims with Apple Intelligence last year, when they had everyone fooled.
This is like watching a train-wreck in slow motion.
Apple's ambitions are actually bigger than openai or anthropopic. Only Google's ambition (surprise surprise) is similar. Apple fundamentally wants the llm to be a tool. It doesn't want the llm to be the product.
They're not a model company. The risks of deploying something half-baked to their users is unacceptable. They're taking it slow and trying to do it in a way that doesn't damage/erode their brand.
Wait it out, let the best model(s) rise to the surface (and the hallucination problems to get sufficiently mitigated), and then either partner with a proprietary provider or deploy one of the open source models. Makes more sense than burning billions of dollars training a new foundation model
This is a reasonable approach, but unfortunately misses what made Apple soooo successful. Apple is the master of controlling the brand. Apple DOES NOT like to highlight their suppliers. Nobody knows who makes iPhones displays, or sensors, or RAMs.
They love to "invent" brands that they control, so that they can commodotize the underlying supplier. Hey user, it is a retina display and dont worry whether it is LG or Samsung is making it.
Apple tried this with AI, calling it "Apple Intelligence". Unfortunately that faltered. Now Apple will have to come out and say "iPhone with ChatGPT" or "Siri with Claude". AND APPLE HATES THAT. HATES IT WITH PASSION.
People will start to associate smartness with ChatGPT or Claude, and Apple loses control and OpenAI/Anthropic's leverage goes up.
Apple has painted themselves into a corner. And as I said elsewhere, it is a train-wreck happening in slowmotion.
They already deployed half-baked models (eg needing to disable news summaries because they were so bad), and haven't delivered on other aspects of apple intelligence. This is hard to call being cautious, this is them not being able to keep up.
Exactly. Another mobile.me moment that adversely impacts customers is worse than making something useful that works. Anyone that “needs” AI can use an app.
I wouldn’t go as far as GP, but yes, absolutely, they must compete with large models on the internet. Customers are now used to being able to ask a computer a question and get something better than “I just ran a web search for what you said, here are the uncurated, unsummarized results”.
Yes, this is in fact what people want. Apple is the biggest company in the world (don’t quibble this y’all, you know what I mean) and should be able to deliver this experience. And sure, if they could do it on device that would be aces, but that’s not an item on the menu, and customers seem fine with web-based things like ChatGPT for now. To act like Apple is doing anything other than fumbling right now is cope.
I see it as the opposite. Apple is absolutely positioned to own "chat". I am not worried they'll soon sort things out — and eventually we'll have an LLM integrated into the iPhone; call it Siri or otherwise.
With my history encrypted in the cloud, and the trust that Apple has built around privacy ... I think they're going to come out alright.
But they have de facto admitted failure of most of the strategy if the rumours are true that they are switching much harder to OpenAI/Anthropic for upcoming LLM products.
This is the first time in 10+ years I've seen Apple so far on the back foot. They usually launch category defining products that are so far ahead of the competition, even by the time they work through the 'drawbacks' in the first versions of them they are still far ahead. OS X, the iPhone and the iPad were all like that. They are still way ahead of the competition on Apple Silicon as well.
I am not very confident on their on device strategy at least in the short to medium term. Nearly all their devices do not have enough RAM and even if they did SLMs are very far behind what users "know" as AI - even the free ChatGPT plan is leap years ahead of the best 3B param on device model. Maybe there will be huge efficiency gains.
Private cloud is used AFIAK for virtually 0 use cases so far. Perhaps it will be more interesting longer term but not very useful at the moment given the lack of a suitable (ie: non Chinese), large (>500b param) model. They would also struggle to scale it if they roll it out to billions of iOS devices especially if they put features that use a lot of tokens.
Then they've got OpenAI/Gemini/Anthropic via API. But this completely goes against all their private cloud messaging and gives those providers enormous potential control over Apple, which is not a position Apple usually finds itself in. It will also be extremely expensive to pay someone per token for OS level features for billions of iOS/Mac devices and unless they can recoup this via some sort of subscription will hit services margins badly.
To me its clear the future of "OS" is going to involve a lot of agentic tool calling. These require good models, with large context windows and a lot of tokens - this will definitely not work on device. Indeed this is exactly what the Siri vapourware demo was.
I'm sure they can potentially get to a great UX (though these missteps are making me question this). But having such a core feature outsourced does not leave them in a good position.
Apple is only "behind" if you think they're in the same race. They haven't shown any interest in developing frontier models or taking on the enormous costs of doing so.
Did you even watch Apple Intelligence ads? They were very much in the race, just that they got ahead of themselves a bit.
They were touting the same features that other companies are now delivering. Point the phone at something, and it'll tell you what you're looking at. Or summarize news articles etc. Instead we got .. emojithingy
I'm having trouble understanding, do you think people are going to stop buying iPhones because Siri isn't as good as ChatGPT? Do you think Apple users are going to flood over to the Pixel phone to use Gemini?
I feel like this is the most exciting news today about AI on hn. I really hope apple shows that small models can be just as capable as the bigger ones. Maybe they have the people on perplexity working on these small models.
Lol and yet, Google has AI image descriptions in their screen reader, TalkBack, before Apple. Apple is supposed to be the accessibility king. But with AI, they just can't, even if they obviously have access to ChatGPT which has vision capabilities. Granted, I don't know what model Google uses because tech news don't do Android Accessibility Suite APK teardowns, but it works pretty well, and fast too.
It's hard to know what it isn't for certain but there are many other reasons papers list contributors in a flat structure (be it random or alphabetical order). Particularly with large numbers of collaborators.
Not very hard to look people up on LinkedIn and figure out who the core researchers are. I think this is just a very surface-level overview paper that encompasses a bunch of different research projects conducted by different teams, and it would be difficult to order the contributors in any meaningful way.
Considering a large portion of the contributors have names originating in a script and language that has no relationship whatsoever to English’s already arbitrary letter ordering, this list configuration is as good as any.
This is the first time that millions of people will actually download and run a model on their own devices.
The question is… will Apple be constantly tweaking these models, or only during OS upgrades?
I for one really like local software. Call me old-fashioned, but I enjoy when a company doesn’t switch up software anytime on the server, or phone the results home all the time in order to extract more profits from their collective users.
> “Adapters produced by the toolkit are fully compatible with the Foundation Models framework. However, each adapter is compatible with a single specific model version, meaning that a new adapter must be trained for each new version of the base model.”
Any changes should require retraining any LoRA adapters that has been built & distributed by third party developers, so they wouldn’t update the models outside OS updates at the drop of a hat I don’t think.
LoRA adapters can be distributed via Background Assets, but the base model itself should be version-locked to the OS build (e.g. iOS 26.0 → 26.1) and updates only when Apple ships a new OS image.
In the meantime, when I ask Siri to set a timer for 15 minutes, about 10–15% of the time it just says, “Here’s what I found about setting a timer for 15 minutes,” instead of actually setting the timer"
> We believe in training our models using diverse and high-quality data. This includes data that we’ve licensed from publishers, curated from publicly available or open- sourced datasets, and publicly available information crawled by our web-crawler, Applebot.
> We do not use our users’ private personal data or user interactions when training our foundation models. Additionally, we take steps to apply filters to remove certain categories of personally identifiable information and to exclude profanity and unsafe material.
> Further, we continue to follow best practices for ethical web crawling, including following widely-adopted robots.txt protocols to allow web publishers to opt out of their content being used to train Apple’s generative foundation models. Web publishers have fine-grained controls over which pages Applebot can see and how they are used while still appearing in search results within Siri and Spotlight.
Respect.
When Apple inevitably partners with OpenAI or Anthropic, which by their definition isnt doing "ethical crawling", I wonder how I should be reading that.
They already partnered with OpenAI, right?
6 replies →
In theory Apple could provide their training data to be used by OpenAI/Anthropic.
1 reply →
"Good artists copy; great artists steal"
- Famous Dead Person
I mean they also buy from companies with less ethical supply chain practices than their own. I don’t know that I need to feel anything about that beyond recognizing there’s a big difference between exercising good practices and refusing to deal with anyone who does less.
[dead]
Same way as the other parts of their supply chain I suppose.
You shouldn't believe Big Tech on their PR statements.
They are decades behind in AI. I have been following AI research for a long time. You can find best papers published by Microsoft, Google, Facebook in past 15 years but not Apple. I don't know why but they didn't care about AI at all.
I would say this is PR to justify their AI state.
Apple used to be at the edge of AI. They shipped Siri before "AI assistant" went mainstream, they were one of the first to ship an actual NPU in consumer hardware and put neural networks into features people use. They were spearheading computational photography. They didn't publish research, they're fucking Apple, but they did do the work.
And then they just... gave up?
I don't know what happened to them. When AI breakthrough happened, I expected them to put up a fight. They never did.
3 replies →
One problem with Apple's approach here is that they were scraping the web for training data long before they published the details of their activities and told people how to exclude them using robots.txt
They documented it in 2015: https://www.macrumors.com/2015/05/06/applebot-web-crawler-si...
1 reply →
Uncharitable.
Robots.txt is already the understood mechanism for getting robots to avoid scraping a website.
13 replies →
> Using our web crawling strategy, we sourced pairs of images with corresponding alt-texts.
An issue for anti-AI people, as seen on Bluesky, is that they're often "insisting you write alt text for all images" people as well. But this is probably the main use for alt text at this point, so they're essentially doing annotation work for free.
I think it is entirely morally consistent to provide alt text for accessibility even if you personally dislike it being used to train AI models.
5 replies →
> An issue for anti-AI people, as seen on Bluesky, is that they're often "insisting you write alt text for all images" people as well. But this is probably the main use for alt text at this point, so they're essentially doing annotation work for free.
How did you come to the conclusion that those two groups overlap so significantly?
3 replies →
Alt text gives you 2k characters. All I gotta say is there's quite a bit of poisoned data
[flagged]
Respect, but its going to be terrible compared to every other company. You can only hamstring yourself so much.
Respect actions, not words and PR.
Gotta polish that fig-leaf to hide Apple's real stance towards user privacy: arstechnica.com/tech-policy/2023/12/apple-admits-to-secretly-giving-governments-push-notification-data/
> Apple has since confirmed in a statement provided to Ars that the US federal government "prohibited" the company "from sharing any information,"
I mean if you throw out all contrary examples, I suppose you are left with the simple lack of nuance you want to believe
2 replies →
All I can say is, I asked Siri today (verbatim): What is 75 degrees fahrenheit in celsius, and what is 85 degrees in fahrenheit — and it offered a web search about fahrenheit. The "and" completely disabled its most basic ability to do metric conversions.
So, it's nice to see Apple is doing research and talking about it, but we're out here waiting, still waiting, for anything useful to make of it all on our thousand-dollar devices that literally connect us to the world and contain our entire life data. It's what I would've expected from one of the most valuable companies in the world.
You asked 2 questions in a system made for 1 question at a time. Split these up and Siri answers them fine. You’re holding it wrong.
A tool that can handle more than one question at a time is useful. Modern LLMs handle that with ease. So it's completely reasonable to be critical of that limitation.
10 replies →
Never mind that Infocom games running on my Apple ][+ could handle that sort of command in 1983.
(Well, with multiple direct objects, anyway.)
"holding it wrong" was exactly the right phrase given how that phrase was used with the iPhone antenna bridging problem. This is an Apple product failing.
"You haven't contorted your comically simple query enough to make the brittle tool work. Throw the chicken bones better next time."
5 replies →
> What is 75 degrees fahrenheit in celsius, and what is 85 degrees in fahrenheit
Err, what? As a native English speaker human that's a pretty confusing question to me, too!
First, most of the English speaking world is not native.
"As of 2022, there were about 400 million native speakers of English. Including people who speak English as a second language, estimates of the total number of Anglophones vary from 1.5 billion to 2 billion."
Second, all popular models I tested did well with that query, including Gemini on Android (aka "ok Google"), except Apple's.
https://en.m.wikipedia.org/wiki/English-speaking_world
2 replies →
I just tried this on my phone and just got two pop ups with the conversions appear in quick succession.
>> What is 75 degrees fahrenheit in celsius, and what is 85 degrees in fahrenheit
Probably wouldn't have made a difference but the second half of that statement isn't exactly clear. 85 degrees what?
I also think when you're chaining these two separate calculations together you get a problem when it comes to displaying the results.
That exact phrase "What is 75 degrees fahrenheit in celsius, and what is 85 degrees in fahrenheit" given to ChatGPT produces the correct result (it infers that the second degrees must be Celsius) and ChatGPT gives me a nicely laid out formula for the math of the conversion.
So yeah, Apple is way behind on this stuff.
the fact is that gemini responds with this: 75 degrees Fahrenheit is 23.89 degrees Celsius, and 85 degrees Celsius is 185.00 degrees Fahrenheit.
Meanwhile users have been conditioned to expect a system that understand the multiple queries and answers them appropriately.
1 reply →
Your usage of Siri today (probably on an old version of iOS) frankly has nothing to do with the article we are discussing. Sorry to say this but it is going to take time. Comparing the performance of a chatgpt running in a big data center with a model running locally on a phone device... give it a few years.
People have been giving Siri a few years for a decade now. Siri used to run in a data center (and still does for older hardware and things like HomePods) and it has never supported compound queries.
Siri needs to be taken out back and shot. The problem with “upgrading” it is the pull to maintain backwards compatibility for every little thing Siri did, which leads them to try and incorporate existing Siri functionality (and existing Siri engineers) to work alongside any LLM. Which leads to disaster, and none of it works and just made it all slower. They’ve been trying to do an LLM assisted Siri for years now and it’s the most public facing disaster the company has had in a while. Time to start over.
4 replies →
> Your usage of Siri today (probably on an old version of iOS) frankly has nothing to do with the article we are discussing.
Yes, but isn't that infuriating? The technology exits! It even exists, as evidenced by this article, in the same company that provides Siri!
At least I feel that way every time I interact with it – or for that matter my Google Home speaker, ironically made and operated by the company that invented transformer networks.
Despite all the “Apple is evil” or “Apple is behind” (because they don’t do evil). Well, what they made with the Foundation Model is great. The fact that they build a system within the Swift language that allows you to specify structured data models (structs) to be used like any other model in a modern programming language, and you actually get back generated data in that format is great. Unlike a lot of other AIs where you might get back a well formatted JSON after a carefully crafted request, but still you never can’t be sure and need to implement a bunch of safeguards. Obviously it’s still the beginning and other tools might do something similar. But as an iOS developer that makes the usage of AI so much simpler. Especially with the bridge to external AIs that still allows you to map back to the type safe structured Swift models. I try not to be a hater, every progress, even slow or underwhelming at first might lead to improvements everywhere else.
Apple is behind. People forget that Google was shipping mobile-scale transformer-based LLMs in 2019: https://github.com/google-research/bert
By the time Apple has an AI-native product ready, people will already associate it with dehumanization and fascism.
I think it's a great move that Apple is cautious with including a hardcore LLM in everything. They are not that useful to the regular user.
2 replies →
Guided generation is called "Structured Output" by other providers?
Well partially generated content streaming thing is great and I haven't seen it anywhere else.
Sorry if I didn’t use the correct terms. Didn’t catch up on all the terminology coming from my native language. ;) But yes, I agree, the fact that parts, different parameters, of the model can be completed asynchronous by streaming the output of the model, is quite unique. Apple/swift was late with async/await, but putting it all together, it probably plays well with the ‘never’ (I know ) asynchronous and reactive coding.
An issue with this is that model quality can get a lot lower when you force it into a structured form, because it's out of distribution for the model.
(I'm pretty sure this is actually what drove Microsoft Sydney insane.)
Reasoning models can do better at this, because they can write out a good freeform output and then do another pass to transform it.
5 replies →
How do you think their implementation works under the hood? I'm almost certain it's also just a variant of "structured outputs", which many inference providers or LLM libraries have long supported.
Huh? Grammar-based sampling has been commonplace for years. It's a basic feature with guaranteed adherence. There is no "carefully crafting" anything, including safeguards.
Every time I see a paper from Apple I just feel like, OK so why isn’t my iPhone actually doing any of this yet?
Why give this to developers if you haven’t been able to get Siri to use it yet? Does it not work or something? I guess we’ll find out when devs start trying to make stuff
> why isn’t my iPhone actually doing any of this yet?
What exactly are you referring to? Models do run on iPhone and there are features that take advantage of it, today.
None of those features are in any way interesting though. Image playground is a joke, Siri is a joke, that generative emoji thing is a joke.
The AI stuff with photography sure, but that’s more like machine learning.
The photo touch up thing is… useable? Sometimes?
What is it you’ve been so impressed with?
8 replies →
Yes, but why do I have to open a third-party app to do these things when Apple, the company that primarily popularized the entire genre of mobile voice assistants, could very feasibly bake all of that into theirs?
I mean, the thing even lets me ask ChatGPT things if I explicitly ask it to! But why do I need to ask in the first place?
3 replies →
> why isn’t my iPhone actually doing any of this yet?
Probably Apple is trying to distill the models so they can run on your phone locally. Remember, most, if not all, of Siri is running on your device. There's no round trip whatsoever for voice processing.
Also, for larger models, there will be throwaway VMs per request, so building that infra takes time.
It says there’s 2 models - one local. It’s already released to app developers to use locally I think (it was in the keynote for WWDC).
3 replies →
They just launched "Private Cloud Compute" with much fanfare to enable server-side LLM processing, so between that and the fact that Siri has been server-based for most of its existence (local processing is fairly new), I don't think that's their main constraint at this point.
That said, "Private Cloud Compute" does run on proprietary Apple hardware, so availability might be a concern (assuming they don't want to start charging for it).
Apple silicone unified memory is amazing for running things like ollama. You don’t have to wait for them to release their own applications.
I know apple is methodical and don’t show their hand but I cannot help but feel they are releasing all this research because they haven’t integrated any into the phone or provided a compelling AI functionality for their users. This is their only way to say “hey we are good with AI too”.
AFAICT this is the first commercial model trying to be marketed as responsibly-sourced. Love it, but it also seems like the noise around this issue has died down. Is this for legal cover? Or more apple-privacy marketing
Stockholders are suing them over Apple Intelligence. Definitely legal cover.
"Sorry we are hilariously far behind everyone else in the industry after having made a huge amount of fanfare about 'Apple Intelligence' for years. It's just that we have shot ourselves in the knee to satisfy Bluesky posters and the NY Time's lawyers"
Do people have an issue with the smollm datasets? I guess it isn't really commercial.
Siri is literally a joke!
My son (he's 11 years old now and fairly skilled with all the main AI tools, eg chatgpt, gemini, etc) and I retry her every month or so, and this past time we just laughed. Can't handle basic questions - hears the question wrong, starts, stops, takes us to some random ass webpage, etc, etc.
"She's so jacked up!" he said.
Apple needs to get this under control and figured out, stat!
Apple can’t afford to run models, there are too many iPhones and not enough data centers.
Running on device is also risky because cycle limitations will make it seem dumb in comparison.
Looks nice. I just wish they’d improve the models behind dictation on both iPhone and Mac to have better accuracy and on the fly multiple language transcription.
I'd really like to be able to use this 3B model on my little 4GB GPU card! It looks very capable for a reasonable weight. Maybe one day on HhuggingFace
This isn’t the Apple I remember. Product integration falls apart at every seam, but don’t worry—we’ve got plenty of impressive technical documentation to compensate. I’m sure Jobs would be thrilled to see his ‘it just works’ philosophy replaced with ‘it barely works, but here’s a 50-page PDF explaining why.
The philosophy is the same, and since it was never implemented in the mythical era of Jobs, so is the practice. So he'd be as thrilled as he was back then?
What I don't understand is how this happened. What really has changed at Apple in the last decade?
As someone that was around in the days of Apple before bankruptcy, the same, there is no Jobs more around, and is getting back to Gil Amelio kind of Apple.
Tim Cook might be better at squeezing the juice, but he is not a product person.
This time around they need another solution, otherwise regardless of how much money they have, they will stay as the iOS/iPad company, given the relevance of macOS on desktop market worldwide.
I think that iOS 7 theme update caused their brains to rot.
I wonder if we'll see these models running on the phone (aiPhone) hardware in the future.
It does. You can use it directly on iOS 26 beta - without writing a line of code I can toy with the on-device model through Shortcuts on my 16 Pro. It’s not meant to be a general purpose chatbot… but it can work as a general purpose chatbot in airplane mode which is a novel experience.
https://share.icloud.com/photos/018AYAPEm06ALXciiJAsLGyuA
https://share.icloud.com/photos/0f9IzuYQwmhLIcUIhIuDiudFw
The above took like 3 seconds to generate. That little box that says On-device can be flipped between On-device, Private Cloud Compute, and ChatGPT.
Their LLM uses the ANE sipping battery and leaves the GPU available.
It would be interesting to see the tok/s comparison between the ANE and GPU for inference. I bet these small models are a lot friendlier than the 7B/12B models that technically fit on a phone but won't accelerate well without a GPU.
3 replies →
Wild to see what improvements might come if there is additional hardware support in future Apple Silicon chips.
What’s the cost of pointing it to Private Cloud Compute? It can’t be free, can it?
2 replies →
As someone mentioned, this model is available in the beta version of iOS 26; it's also part of macOS 26, iPadOS 26 and visionOS 26. Anyone with a free developer account can install the developer betas; the public beta is expected next week.
There's a WWDC video "Meet the Foundation Models Framework" [1].
[1]: https://developer.apple.com/videos/play/wwdc2025/286
> The new Foundation Models framework gives access to developers to start creating their own reliable, production-quality generative AI features with the approximately 3B parameter on-device language model. The ∼3B language foundation model at the core of Apple Intelligence excels at a diverse range of text tasks like summarization, entity extraction, text understanding, refinement, short dialog, generating creative content, and more. While we have specialized our on-device model for these tasks, it is not designed to be a chatbot for general world knowledge. We encourage app developers to use this framework to design helpful features tailored to their apps
> a ∼3B-parameter on-device model
There are even already some local AFM to Open AI API bridge project on GitHub - that lets you point basically any Open AI compatible client at the local models. Super nice for basic summarisation and completions.
I was worried "device" was a Mac mini, not an iPhone. (I already have been running models on my MacBook Pro.)
The more I think about Apple, the more I realize that Apple is so far behind. While other companies are pushing the envelope (OpenAI, Anthropic, Google ..) Apple's ambitions seem much much smaller.
And this is after they made very big claims with Apple Intelligence last year, when they had everyone fooled.
This is like watching a train-wreck in slow motion.
Apple's ambitions are actually bigger than openai or anthropopic. Only Google's ambition (surprise surprise) is similar. Apple fundamentally wants the llm to be a tool. It doesn't want the llm to be the product.
I think it's the right strategy for Apple.
They're not a model company. The risks of deploying something half-baked to their users is unacceptable. They're taking it slow and trying to do it in a way that doesn't damage/erode their brand.
Wait it out, let the best model(s) rise to the surface (and the hallucination problems to get sufficiently mitigated), and then either partner with a proprietary provider or deploy one of the open source models. Makes more sense than burning billions of dollars training a new foundation model
This is a reasonable approach, but unfortunately misses what made Apple soooo successful. Apple is the master of controlling the brand. Apple DOES NOT like to highlight their suppliers. Nobody knows who makes iPhones displays, or sensors, or RAMs.
They love to "invent" brands that they control, so that they can commodotize the underlying supplier. Hey user, it is a retina display and dont worry whether it is LG or Samsung is making it.
Apple tried this with AI, calling it "Apple Intelligence". Unfortunately that faltered. Now Apple will have to come out and say "iPhone with ChatGPT" or "Siri with Claude". AND APPLE HATES THAT. HATES IT WITH PASSION.
People will start to associate smartness with ChatGPT or Claude, and Apple loses control and OpenAI/Anthropic's leverage goes up.
Apple has painted themselves into a corner. And as I said elsewhere, it is a train-wreck happening in slowmotion.
1 reply →
They already deployed half-baked models (eg needing to disable news summaries because they were so bad), and haven't delivered on other aspects of apple intelligence. This is hard to call being cautious, this is them not being able to keep up.
Exactly. Another mobile.me moment that adversely impacts customers is worse than making something useful that works. Anyone that “needs” AI can use an app.
1 reply →
Only if you think they _must_ compete with large models on the internet.
I'm fine with Apple chilling on the sidelines for a bit.
I wouldn’t go as far as GP, but yes, absolutely, they must compete with large models on the internet. Customers are now used to being able to ask a computer a question and get something better than “I just ran a web search for what you said, here are the uncurated, unsummarized results”.
Yes, this is in fact what people want. Apple is the biggest company in the world (don’t quibble this y’all, you know what I mean) and should be able to deliver this experience. And sure, if they could do it on device that would be aces, but that’s not an item on the menu, and customers seem fine with web-based things like ChatGPT for now. To act like Apple is doing anything other than fumbling right now is cope.
6 replies →
I see it as the opposite. Apple is absolutely positioned to own "chat". I am not worried they'll soon sort things out — and eventually we'll have an LLM integrated into the iPhone; call it Siri or otherwise.
With my history encrypted in the cloud, and the trust that Apple has built around privacy ... I think they're going to come out alright.
But they have de facto admitted failure of most of the strategy if the rumours are true that they are switching much harder to OpenAI/Anthropic for upcoming LLM products.
This is the first time in 10+ years I've seen Apple so far on the back foot. They usually launch category defining products that are so far ahead of the competition, even by the time they work through the 'drawbacks' in the first versions of them they are still far ahead. OS X, the iPhone and the iPad were all like that. They are still way ahead of the competition on Apple Silicon as well.
I am not very confident on their on device strategy at least in the short to medium term. Nearly all their devices do not have enough RAM and even if they did SLMs are very far behind what users "know" as AI - even the free ChatGPT plan is leap years ahead of the best 3B param on device model. Maybe there will be huge efficiency gains.
Private cloud is used AFIAK for virtually 0 use cases so far. Perhaps it will be more interesting longer term but not very useful at the moment given the lack of a suitable (ie: non Chinese), large (>500b param) model. They would also struggle to scale it if they roll it out to billions of iOS devices especially if they put features that use a lot of tokens.
Then they've got OpenAI/Gemini/Anthropic via API. But this completely goes against all their private cloud messaging and gives those providers enormous potential control over Apple, which is not a position Apple usually finds itself in. It will also be extremely expensive to pay someone per token for OS level features for billions of iOS/Mac devices and unless they can recoup this via some sort of subscription will hit services margins badly.
To me its clear the future of "OS" is going to involve a lot of agentic tool calling. These require good models, with large context windows and a lot of tokens - this will definitely not work on device. Indeed this is exactly what the Siri vapourware demo was.
I'm sure they can potentially get to a great UX (though these missteps are making me question this). But having such a core feature outsourced does not leave them in a good position.
4 replies →
Apple is only "behind" if you think they're in the same race. They haven't shown any interest in developing frontier models or taking on the enormous costs of doing so.
Did you even watch Apple Intelligence ads? They were very much in the race, just that they got ahead of themselves a bit.
They were touting the same features that other companies are now delivering. Point the phone at something, and it'll tell you what you're looking at. Or summarize news articles etc. Instead we got .. emojithingy
I'm having trouble understanding, do you think people are going to stop buying iPhones because Siri isn't as good as ChatGPT? Do you think Apple users are going to flood over to the Pixel phone to use Gemini?
What is this train-wreck you are hallucinating?
The paper was a very nice read, and they did many creative things. It's a pity this model won't be directly accessible, only integrated in some apps.
> It's a pity this model won't be directly accessible, only integrated in some apps.
It's already accessible using Shortcuts, even to non-developers "iOS 26 Shortcuts + Apple Intelligence is POWERFUL " (Youtube) [1].
[1]: https://youtu.be/Msde-lZwOxg?si=KJqTgtWjpdNDxneh
When the Blackberry ruled the Earth, people asked 'Why doesn't Apple do a smartphone?'.
I feel like this is the most exciting news today about AI on hn. I really hope apple shows that small models can be just as capable as the bigger ones. Maybe they have the people on perplexity working on these small models.
Lol and yet, Google has AI image descriptions in their screen reader, TalkBack, before Apple. Apple is supposed to be the accessibility king. But with AI, they just can't, even if they obviously have access to ChatGPT which has vision capabilities. Granted, I don't know what model Google uses because tech news don't do Android Accessibility Suite APK teardowns, but it works pretty well, and fast too.
Hasn't Apple had AI image descriptions in VoiceOver for 5 years now? https://www.idropnews.com/ios-14/ios-14-adds-ai-based-voiceo...
The dozens of "contributors" being presented in random order is, one would suppose, an anti-poaching tactic?
It's hard to know what it isn't for certain but there are many other reasons papers list contributors in a flat structure (be it random or alphabetical order). Particularly with large numbers of collaborators.
"References" section sort of narrows the field anyway.
As someone whose last name is near the end of the alphabet, that's not the first presumption I had seeing that page.
Well meta already got Ruoming so he can obviously give them a ranked list of who to grab.
Most of his team are former Google brain so GDM knows who is good.
Not very hard to look people up on LinkedIn and figure out who the core researchers are. I think this is just a very surface-level overview paper that encompasses a bunch of different research projects conducted by different teams, and it would be difficult to order the contributors in any meaningful way.
Considering a large portion of the contributors have names originating in a script and language that has no relationship whatsoever to English’s already arbitrary letter ordering, this list configuration is as good as any.
Here is my question…
This is the first time that millions of people will actually download and run a model on their own devices.
The question is… will Apple be constantly tweaking these models, or only during OS upgrades?
I for one really like local software. Call me old-fashioned, but I enjoy when a company doesn’t switch up software anytime on the server, or phone the results home all the time in order to extract more profits from their collective users.
> The question is… will Apple be constantly tweaking these models, or only during OS upgrades?
Certainly when new updates are released--going from macOS 26 to 26.1).
They can probably push model updates between releases if necessary.
Per the PDF in this post:
> “Adapters produced by the toolkit are fully compatible with the Foundation Models framework. However, each adapter is compatible with a single specific model version, meaning that a new adapter must be trained for each new version of the base model.”
Any changes should require retraining any LoRA adapters that has been built & distributed by third party developers, so they wouldn’t update the models outside OS updates at the drop of a hat I don’t think.
LoRA adapters can be distributed via Background Assets, but the base model itself should be version-locked to the OS build (e.g. iOS 26.0 → 26.1) and updates only when Apple ships a new OS image.
1 reply →
The model is gigabytes so I doubt they will push updates frequently.
Educate me: is there any work on modifying models in a way that changes relatively few parameters, so an update is a smaller payload?
2 replies →
In the meantime, when I ask Siri to set a timer for 15 minutes, about 10–15% of the time it just says, “Here’s what I found about setting a timer for 15 minutes,” instead of actually setting the timer"