Apple found itself playing catch-up as the age of generative AI dawned with the release of OpenAI's seminal ChatGPT 3.5 in November 2022. It took Apple more than 18 months to respond, significantly longer than fellow tech giants like Google and Amazon, and handset makers like Samsung.
This became the key announcement for WWDC 2024, Apple's annual conference for developers who create apps and services for its software platforms, now held at its spaceship-like campus in Cupertino.
While Apple is often fashionably late to the party, this time there were concerns that it wasn't just late but ill-prepared to respond. Generative AI seemed to be outside of Apple's comfort zone. Apple's answer is its own take on AI, dubbed Apple Intelligence. It is a less technical approach to large language models based on transformers, focused on delivering seamless, responsive, and private intelligence through Apple's devices.
What Is Apple Intelligence?
Apple had been quietly working on this and unveiled a 3 billion parameter model that runs natively on the iPhone, iPad, or Mac in tandem with its private cloud compute servers powered by custom Apple Silicon. These servers have no storage, only RAM, and are audited by security researchers for every change Apple makes to ensure privacy is maintained. It even verifies the software that is being run and if the verification fails it will not execute the query. It is all very sophisticated from a security perspective, something that could only be achieved by a prodigiously talented team tasked with making the most secure server architecture on the planet with constraints like a budget being non-existent.
As if this wasn't enough, Apple has partnered with OpenAI, providing a layer of ChatGPT 4o integration that the revamped Siri can tap into and suggest to users. This also happens privately, with queries anonymised and no OpenAI account login required. Users with existing ChatGPT premium subscriptions will be able to log in and unlock more advanced features of the language model.
Apple isn't stopping there. As Craig Federighi, its Senior Vice President for software, indicated in closed-door briefings with international media, more third-party foundational models like Google's Gemini will be added, giving users choice. This effectively turns Apple's platforms into a kind of AI-neutral ground.
How Is Apple Intelligence Different?
Apple's approach is unique. A 3 billion parameter model is modest, even compared to Microsoft's new Phi 3. It's not even a large language model, but rather a small one. However, when implemented correctly, its advantages are significant. It will be fast, especially on Apple's devices – whether smartphones, notebooks, or tablets. Apple's A-series chips pioneered the idea of a neural engine and machine learning accelerators on the CPU back in 2017.
Its M-series chips have set the benchmark for the notebook market since late 2020, only now being challenged by Qualcomm's Snapdragon X Elite, Intel's latest Meteor Lake, and upcoming Lunar Lake chips. But Apple's latest M4 is already here in the newest iPad Pro, capable of 38 trillion operations per second.
Microsoft and its chip partners Qualcomm, Intel, and AMD may claim we're in the age of the CoPilot+ PC or AI PC, but even on these powerful devices, much of the processing is offloaded to the cloud. Google's Tensor chips in its Pixel phones and even Qualcomm's Snapdragon 8 Gen 3 in the latest Galaxy S24 models rely on the cloud for 90% of their AI capabilities. While Microsoft and Google are both cloud computing powerhouses, neither are as well-established or trusted as consumer brands, nor are they known for their privacy standards.
Apple is introducing several AI-powered features to make using its devices easier and more efficient. The AI can automatically summarise long emails and documents, intelligently manage notifications, suggest email replies, help rewrite messages, clean up handwritten notes, and even allow searching for specific content within photos. These AI enhancements aim to streamline common tasks and boost productivity without extra effort. And these features work across Apple's own apps like Mail, Photos, Notes, Pages, Keynote, Safari, and are also being opened up to third-party apps via APIs (application programming interfaces).
These are capabilities we've come to expect from large language models. But Apple is integrating this at an operating system level, not just for the iPhone, but for the iPad and Mac as well. Three of its biggest product categories will be transformed by this, with a significant portion happening on-device. This means it will be faster, more personal, and still private. The added benefit is Apple's unique Private Cloud Compute, which is being offered at no cost.
And when that's not enough, ChatGPT will be privately embedded into all these operating systems later in the year. This is reminiscent of the early days of social media, another area where Apple perhaps didn't excel. It enabled system-level Twitter and Facebook integration in iOS 5 and macOS Lion. Of course, those integrations have since been removed as Apple has taken an anti-social media stance, prioritising user privacy.
In fact, in the early days of the smartphone wars, the iPhone became neutral ground for the best apps from internet giants like Google, Microsoft, Yahoo, Facebook, and Twitter. This is what propelled the iPhone to become the single most important phone model in the world. Apple could very well angle the iPhone as the neural ground for third party LLM models with Windows being a domain of the GPT-40 powered Co-Pilot and Android phones largely gravitating towards Google’s Gemini Nano model.
Why Is Apple Perceived To Be Behind In AI?
Apple has always been the maker of our favourite gadgets. It develops technology to deliver meaningful features, not the other way around. The realm of generative AI is more research-driven. OpenAI was founded for AI research. Transformer-based large language models were developed at Google as part of research efforts, and Google sat on the technology for years until it suddenly found itself behind OpenAI. Developing these models requires vast amounts of data and computing power.
Apple's strong stance on privacy never allowed it to harvest publicly available data or user data the way many suspect OpenAI has, or the way Google does to a certain extent. Apple is also not a player in cloud computing, instead leveraging Google and Microsoft's cloud services for its own needs. It never stockpiled Nvidia's Grace Hopper AI accelerators like cloud computing companies or the Internet did, so it couldn't rapidly create LLMs because they are essential for training these models. Apple also had many of its resources tied up in projects like the Apple Vision Pro, launched last year, and Project Titan, its self-driving car project that was shelved earlier this year.
With Vision Pro released and the car project cancelled, and with investor pressure mounting, resources have been shifted to its generative AI ambitions. In 2018, Apple hired John Giannandrea, former head of Google Search, to lead its AI and machine learning strategy. However, Apple's culture of perfecting products and having teams work in silos didn't help its AI efforts. Giannandrea's team wasn't able to collaborate effectively with the rest of Apple due to a cultural clash. The launch of ChatGPT 3.5 was a turning point, with both Giannandrea and Federighi acknowledging that Siri, the pioneer of virtual assistants and something Steve Jobs last highlighted on the iPhone, had become a caricature in the assistant space despite handling over a billion and a half queries per day.
Now, Apple is leveraging its immense capability in designing semiconductors and deploying its own Mac Studio or Pro models in its private cloud. Its chips were always ahead of the curve for on-device processing, and its vertical integration across platforms and hardware allows it to access personal data privately at high speed, unlike any other player. It took Apple 18 months because it is a large company, and like a large ship, it takes time to change course. And it will do so at a massive scale across device types. Microsoft is doing it on PCs, Google and its OEM partners are doing it on tablets and smartphones, but Apple is doing it comprehensively, designing the software platform, semiconductors, cloud, AI, and the end devices. With WWDC 2024, it's clear that Apple has made the right turn, but this AI war will be a long one, and only time will tell how it plays out.
Sahil Mohan Gupta is a Contributing Editor for BW Businessworld. He writes on auto and tech developments.