Tech Tonic | If Apple Intelligence doesn’t annoy anyone, that’ll be a big win

There won’t be a need to open a chatbot or a separate app (the closest to that is invoking Siri, which is also one voice command away), but systemwide knowledge will give it more to build context.

In my understanding, nothing as detailed and integrated has made its way to Android, Windows, or any other platforms of its size, as yet. There are scattered pieces, but Apple’s implementation can be referenced as a “suite”. Apple’s foundation models are trained on the AXLearn framework, an open-source project released in 2023. Consider the scope.

Rewriting emails or even drafting responses such as RSVPs, changing the tone of text you’ve written, recording phone calls and transcriptions, improving handwriting in notes, picking actionable pointers from web pages, image generations in Animation, Illustration, or Sketch styles or Genmojis as a language of conversation (the cool kids will adopt this quickly, and even Instagram will likely add support soon).

What I’ve mentioned here is just a part of the overall capabilities drawing on Apple On-Device and Apple Server models. That’s before OpenAI’s GPT-4o can be invoked for specific queries that may require expertise (such as curating a recipe for you; Apple doesn’t want to get into that race, for now).

The contours, and how they’ve been designed, make for interesting reading. The way I see it is, that Apple senses AI as the best chance to get you to upgrade your iPhone, iPad, or Mac. That’s the catch: Apple Intelligence won’t work on any current iPhone other than the iPhone 15 Pro and iPhone 16 Pro Max.

In all likelihood, iPhone 16 would, and my opinion is a lot of iPhone users will upgrade to the next generation non-Pro iPhone to use Apple Intelligence.

The other side of the coin has iPad Pro and iPad Air, both supported by Apple M1 chip or newer. Macs also need the M1 as a baseline while the Mac Pro should have the M2 chip.

Apple doesn’t say why this is the case with the iPhones in particular, but there could be a hardware limitation with neural processing on the previous generation A16 Bionic chips for the iPhone.

I checked, and these are the differences I confirmed – the A17 Bionic in the iPhone 15 Pro and iPhone 15 Pro Max has a 16-core neural engine, with up to 35 trillion operations per second alongside more memory to work with and more cores, while the iPhone 15 and iPhone 15 Plus’ A16 Bionic has a 16-core neural engine with up to 17 trillion operations per second. I’ll try to get some clarity on the exact reasons but till we have specifics on whether technicalities played a role, it seems more of a business decision.

Worth noting here, that Apple will not layer any of the AI behind a subscription paywall. Including the rather delightful Math Notes feature, which will delight many parents and children (Clearly, I went to school in the wrong era!). Unlike Samsung and Google, which have already given hints that the end of 2025 is the end of the free run for Samsung’s Galaxy flagship phone users. My suspicion is, thatit is not Samsung but Google, which will begin charging for the Gemini smarts.

Just behave, that’ll be a priceless innovation

I said it before, and I’ll say it again – it may have seemed (over the past 12 months in particular) that Apple had fallen behind in the AI arms race. Google, OpenAI and everyone else around them took enough steps forward to convince everyone that was indeed the case. It may very well have been, had the pretenders to the throne, not stumbled far too often. Time and again, the AI models made a mess of things, annoying someone or the other at some point. Case in point, Gemini’s image generation debacle earlier this year, something the company had to immediately apologise for.

For Apple, the checklist of what not to do became more and more populated as all this unfolded. Generative AI in particular, only really works if it has a lot of data to work with. Even then, there’s no guarantee it will be error-free – something we’ve seen time and again. The tightrope Apple is walking with AI involves success with the complexity of tasks they promise, layered context to hold, and personalisation without compromising data privacy, whilst minimising hallucinations, false positives and generating content that may be offensive to human beings.

Apple executives made clear to me that Apple will not use user data to train models but training and context will be built with broader activity trends. One of the ways to do that is to try and compute as much as possible on the device, which negates the need to send user data to the cloud. The way ChatGPT, Gemini and Copilot do, most of the time. And when complex queries do need to be sent, Apple’s built something they call the Private Cloud Compute servers that run the same M-series chips as iPads and Macs and therefore have similar protection mechanisms enabled.

“What’s happening there is your request to our servers is actually first of all anonymized, so your IP address is masked. Then it talks to a server that has no permanent storage, cannot log, but most importantly is running software where the image is publicized for security researchers to audit or your iPhone won’t even talk to that server is a clever kind of blockchain issue,” details Craig Federighi, Senior Vice President for Software Engineering at Apple, in a briefing session of which HT was a part.

In a technical post detailing the models, Apple says, “We never use our users’ private personal data or user interactions when training our foundation models, and we apply filters to remove personally identifiable information.”

No user query will be retailed by Apple’s models. So much so, that they’ve got OpenAI to toe the line for the GPT-4o integration for the optional call-up by users for certain tasks. To be specific, OpenAI confirms that when accessing ChatGPT within Siri and Writing Tools, requests will not be stored by OpenAI, as well as users’ IP addresses will be obscured.

The unpredictability of generative AI, and the users whose hands it’ll be in, will pose Apple a sort of challenge they perhaps haven’t tackled in many years. But one they are fully ready to do because compromises with user data or privacy are an absolute no-go.

Vishal Mathur is the technology editor for the Hindustan Times. Tech Tonic is a weekly column that looks at the impact of personal technology on the way we live, and vice-versa. The views expressed are personal.