Why Apple Still Hasn’t Cracked AI
Mark Gurman and Drake Bennett (Reddit, Hacker News, 9to5Mac, Dithering):
As for the Siri upgrade, Apple was targeting April 2025, according to people working on the technology. But when Federighi started running a beta of the iOS version, 18.4, on his own phone weeks before the operating system’s planned release, he was shocked to find that many of the features Apple had been touting—including pulling up a driver’s license number with a voice search—didn’t actually work, according to multiple executives with knowledge of the matter.
[…]
“Steve just didn’t believe in customers going to try to find things,” says someone who worked with him. “He believed that Apple’s job was to curate and show customers what they should want.” That belief, like many of Jobs’, shaped the company long after his death. In the mid-2010s, Apple explored the idea of placing a search bar at the top of the iPhone’s home screen, rather than burying it behind a swipe gesture. But Apple’s design team vetoed the idea.
[…]
The main technical issue is that Apple essentially had to split Siri’s infrastructure in half, with the old code underpinning legacy features such as setting alarms and the new code underpinning requests that draw on personal data. The kludge was considered necessary to bring the new features to market as soon as possible, but it backfired, creating integration issues that led to delays. Individual features might look good, employees say, but when code is merged so the pieces can be tested together in Siri, things begin to fall apart.
[…]
When [Rockwell] joined Apple in 2015, he proposed that Siri be much more capable and central to the user experience: a sort of always-on life co-pilot. “He would rant about how important Siri is and how it will be the most important way people will interact with their phone,” someone who knows Rockwell says. At the time, Rockwell succeeded mainly in getting the company to upgrade the assistant’s voices by hiring expensive actors and opening high-end recording studios.
There are so many interesting details here. Federighi couldn’t be convinced of the importance of AI as far back as 2014; Cook and Rockwell showed more interest. The neural engine came from the car project. Giannandrea wanted to integrate Gemini instead of ChatGPT, citing privacy concerns with OpenAI, but lost to Apple’s corporate development team. He still doesn’t think consumers want chatbots. Apple Zurich is working on a monolithic LLM Siri.
As the saying goes, “there’s no such thing as a free lunch”. But that is literally true at Apple. Even as the rest of their tech peer group dished out lavish lunch options to employees, Apple held firm in this regard. Yes, even Steve Jobs paid for his own lunch. Or at least it was true, until the team rushing to get Apple Intelligence out the door clearly needed some sort of extra incentive to keep them working beyond their normal workloads to ship something of vital importance. Something which they still largely failed to do. If there is a better metaphor for this fiasco than this lunch situation, I’ll eat my hat.
The knives are now clearly out (of the cutlery drawer – sorry, I’ll stop) at Apple, as Gurman’s report is the third or fourth or fifth or more, depending on how you want to count them, bit of detailed reporting about the failure, to date, of AI within Apple. In a way, this one feels more like an all-encompassing summary of the situation, and an extension of a couple other reports last month by Wayne Ma of The Information and Tripp Mickle of The New York Times. Putting these three together, it feels like we may have enough data points now to triangulate fairly well what may have actually happened.
And, unsurprisingly, there’s not one thing or one person to blame, it seems. Though I still might argue that this is big enough of a clear and present shitshow that Tim Cook himself ultimately should have stepped in long before he did to clean up some obvious issues.
The mess is so profound that it raises the question of whether Apple has the institutional capabilities to fix it.
[…]
This isn’t like hardware where Apple has successfully entered a category late and dominated it. Hardware plays to Apple’s design and supply chain strengths. In contrast, the rapid iteration of AI models and apps is the antithesis of Apple’s annual OS cycle. It’s a fundamentally different approach driven by intense competition and fueled by billions of dollars of cash.
Presumably, this means that users in the EU will be able to set options like Amazon Alexa or Google Assistant as their default voice assistant on Apple devices.
Apple is working on this change in response to expanding EU regulations, the report said.
It would also be good for iOS and Mac users outside the EU to be able to choose a default voice assistant. Apple probably sees this as a threat, and so they won’t allow it, just like you can’t change the default maps app outside the EU. But there’s an alternate reality where Apple went all-in on making its operating systems platforms for AI, just like they’re platforms for apps. If you can’t make Siri the best, make iOS the best place to use all the assistants. I would have liked to see that.
I had assumed the DMA already covered default virtual assistant, but it seems that none were designated gatekeepers. I can imagine how difficult it will be for third-party services to act as a drop-in replacement for Siri, too.
[…]
There is a long way to go from this to a full Siri replacement, but I will be hugely envious of those who will be able to take advantage of changing the default.
Some Apple executives are now reportedly pushing to turn Siri into a true ChatGPT competitor. A next-generation, chatbot version of Siri has reportedly made significant progress during testing over the past six months; some executives allegedly now see it as “on par” with recent versions of ChatGPT. Apple is also apparently discussing giving Siri the ability to access the internet to gather and synthesize data from multiple sources, just like ChatGPT.
Apple is likely to keep discussion of Siri to a minimum at WWDC 2025 as it focuses on other Apple Intelligence enhancements[…] Apple will apparently focus on improving existing Apple Intelligence capabilities and adding some new ones, such as an AI-optimized battery management mode and a virtual health coach. Google Gemini is also on track to be added as a ChatGPT alternative for Siri in iOS 19.
Apple will make its artificial intelligence models available to developers to use in their apps, reports Bloomberg. The company plans to introduce a new software development kit (SDK) in iOS 19 that will make it easier for app creators to add AI features.
Interesting timing on this news, given that it hit the wire almost exactly one hour before Google I/O is set to kick off. OpenAI and Microsoft already played their hands, so now we have the trifecta[…]
[…]
This will start with the smaller, on-device models. And that’s smart. Those models could be compelling to developers because they’ll be able to run locally on the iPhone (or iPad or Mac) and thus, much faster than any model in the cloud. But they’ll also undoubtedly be far more limited than any larger “flagship” LLM.
Previously:
- AI and Google Search Volume in Safari
- How Apple Fumbled Siri’s AI Makeover
- Apple Shifts Siri From Giannandrea to Rockwell
- Leaked Apple Siri Meeting
- How Apple Could Help With AI and LLMs
- DMA Compliance: Default Maps App in EU
- Rotten
- Apple Delays “More Personalized Siri” Apple Intelligence Features
- EU iOS Envy
