In a move that caught many by surprise, Apple unveiled a slew of powerful new AI capabilities coming to its products and services later this year. But rather than simply licensing OpenAI’s viral ChatGPT model, Cupertino is charting its own unique course prioritizing user privacy and on-device AI processing.
The headline announcement? Apple is developing its own large language model called “Apple Intelligence” as a direct competitor to ChatGPT. This conversational AI can tackle writing tasks, answer questions, and more – all while running locally on your Apple hardware to protect your privacy.
“Privacy is a fundamental human right, and AI shouldn’t come at the expense of your personal data,” said Craig Federighi, Apple’s SVP of software engineering, in kicking off the WWDC keynote. “With Apple Intelligence, we’ve created an incredibly capable AI assistant that never shares what you say with the cloud.”
Powering Apple Intelligence is a 3 billion parameter language model trained by Apple’s machine learning teams. While not as massive as OpenAI, Apple claims its more compact model can still handle most conversational and writing tasks with high accuracy. Oh, and did we mention it runs 100% on-device?
Beyond its ChatGPT rival, Apple also highlighted a range of other impressive new AI capabilities powered by lightweight on-device models. An all-new Visual Lookup feature coming to iOS 18 can identify objects, plants, landmarks and more from photos using an efficient machine vision model. And major upgrades are coming to Live Text, allowing you to grab text from video or even handwritten notes.
To handle more complex image and language tasks that require greater compute power, Apple is introducing a new “Private Cloud Compute” service. By leveraging user-encrypted cloud processing, advanced AI models can run securely without Apple or anyone else gaining access to your personal data.
And you know how Apple loves its vertical integration – Cupertino is even designing custom silicon optimized for AI workloads to keep things running blazingly fast. “Only Apple can deliver industry-leading performance-per-watt by designing the entire stack end-to-end,” boasted Johny Srouji, Apple’s chiphead. We’ll have to see it to believe it, but it sure sounds impressive on paper.
After the keynote, hands-on with Apple Intelligence for writing and Q&A. My hot take? Despite being an on-device model optimized for efficiency, it packs a serious punch. The writing quality is excellent and on-par with ChatGPT from my experience so far. Queries are generally handled with the kind of nuance and smarts you’d expect. Plus, having it run locally with no internet connection required is just so incredibly convenient.
The Apple Intelligence avatar even picked up on my casual tone and matched it by responding with a friendly, conversational flair:
“Hey there! Whatcha working on? Need any help fleshing out an idea or taking notes? I’m an AI writing assistant, but a pretty capable one if I do say so myself. Just think of me as your new robotic BFF…minus the potential for human error, existential crises, or stressful social obligations.”
Okay, Apple’s anthropomorphized AI is a little cheesy. But you get the idea – Apple Intelligence aims to strike a balanced tone that’s knowledgeable yet approachable. I could definitely see relying on it as a co-pilot for all sorts of writing tasks like drafting emails, essays, scripts, and more. The ability to set custom tones also seems handy.
Of course, one potential downside is that Apple Intelligence may not handle open-ended queries as deftly as ChatGPT since it was designed more for writing assistance, Q&A, and specific AI capabilities like Visual Lookup. When I asked it open-ended questions far outside its training data like “What are your thoughts on the philosophical implications of strong AI?”, the responses weren’t great:
“I don’t actually have personal thoughts or philosophical views. I’m an AI assistant created by Apple to be helpful, harmless, and honest within my training.”
Fair enough – Apple Intelligence knows its limits as a narrow AI trained for certain use cases. That stands in contrast to ChatGPT’s more free-wheeling responses that sometimes border on coming across as sentient (even though it’s not). But hey, maybe coherence and honesty about being an AI model is better than giving misleading human-like responses.
Apple Intelligence seems immensely useful for its intended writing, Q&A, and AI modeling use cases. Just don’t expect the same level of general “world knowledge” as ChatGPT before its anthropic filter was applied. It’s a capable co-pilot rather than a know-it-all.
In many ways, Apple’s privacy-first approach to AI feels like the smart narrative play it needed to make after recent swirls of AI hype and backlash. While the tech giants race to integrate models like GPT trained on unconsenting internet user data, Apple is taking a more measured stance rooted in its pro-privacy, pro-security brand values.
Will average consumers really understand (or care about) the nuances like on-device vs cloud processing? Maybe not. At least Apple has a clearly communicated positioning in staking out the “privacy AI” territory. And integrating large language models in a localized way is undeniably impressive technical feat, especially when combined with on-device visual AI on the camera and iPadOS’ handy new Live Text upgrades.
So while the rumor mill churned out overblown ChatGPT partnership speculation, Apple played it smart by developing a groundbreaking privacy-forward AI strategy all its own. Cupertino is using vertical integration as a competitive advantage yet again. Time will tell if Apple Intelligence catches on, but you can bet the cameras will be watching the narrative closely.
