Apple VP of AI Retiring
Apple’s Senior VP of Machine Learning and AI Strategy, John Giannandrea, is retiring, and being replaced with Amar Subramanya who recently worked at Microsoft, but has been with Google for 16 years and worked on the Gemini assistant.
Apple is in a weird place with AI. They recognized early on the value of machine learning, and their strategy has been to bake it into the products to produce valuable features, without really identifying it as AI.
They have a great architecture for it. They’ve been building in AI acceleration hardware for years. The photos and camera pipeline specifically makes heavy use of machine learning. Other areas include Touch ID and Face ID, the rich handwriting recognition and math features in Notes, health features in the Apple Watch, even battery charging optimization.
The unified memory architecture that once seemed like an architectural mistake, has accidentally turned out to be exactly what you need for big LLMs.
But generative AI caught them by surprise, and then culturally they didn’t know what to do with it.
Apple has always been a very “human” company, and features that replace human creativity with AI are just culturally not aligned with who they are.
They’ve been flailing for the last 2 years. Siri’s efforts keep getting rebooted and retooled, and the main interface that people use for AI today - a chatbot - is completely missing.
They’re also paying for some bad architectural decisions with Siri. Best-in-class AI is not compatible with privacy. It’s just not. For the frontier models, you need hardware that’s bigger than what fits in a phone, so you have to either limit yourself to what you can run on a phone, or do it remotely.
Apple has tried to find a middle ground with Private Cloud Compute, a clever solution that runs the LLM in the cloud on essentially a beefy Mac that they trust with your sensitive information. Problem is, PCC is still fairly restricted compared to what top-tier LLMs run on, and until we actually get a new Siri, I’m not sure it’s being used for much of anything.
But Apple has potential.
I think Apple has more potential here than either Microsoft or Google does, because they own more of the stack. If you’re in the Apple ecosystem, you’re using an Apple desktop OS, an Apple phone OS, and Apple’s cloud services. Google doesn’t (really) have a desktop OS, and Microsoft doesn’t have a phone.
What Apple needs is a vision, and for someone to reset their culture on AI and privacy.
Microsoft has made culture shifts before. They famously missed the Internet, until Bill Gates redirected them. But they also missed mobile, and simply surrendered. I don’t know what Amar Subramanya will have gotten from his time at Microsoft, but I’m guessing his short tenure means it didn’t click with him.
His much longer tenure at Google is more interesting. Google made the right decisions with AI. They invented the GPT architecture we’re all using now. After some of their own early flailing, they have a solid, integrated direction, much better than the Copilot strategy that Microsoft seems to be taking. Both Copilot and Siri are brands more than products; Gemini is one textbox that you can use to do everything.
It’s not too late for Apple to bring all the pieces together into a coherent whole, but it’s both a technical and a cultural task inside Apple. I wish Amar Subramanya good luck in pulling it off.