How Apple Fumbled Siri’s AI Makeover

Wayne Ma (article):

Apple’s decision to delay new AI features in Siri came after years of dysfunction and clashes between the AI/ML and software engineering groups. Many blame Siri’s poor leadership for its slow progress.

Hartley Charlton:

Apple apparently weighed up multiple options for the backend of Apple Intelligence. One initial idea was to build both small and large language models, dubbed “Mini Mouse” and “Mighty Mouse,” to run locally on iPhones and in the cloud, respectively. Siri’s leadership then decided to go in a different direction and build a single large language model to handle all requests via the cloud, before a series of further technical pivots. The indecision and repeated changes in direction reportedly frustrated engineers and prompted some members of staff to leave Apple.

[…]

Apple’s AI/ML group has been dubbed “AIMLess” internally, while employees are said to refer to Siri as a “hot potato” that is continually passed between different teams with no significant improvements. There were also conflicts about higher pay, faster promotions, longer vacations, and shorter days for colleagues in the AI group.

Apple AI chief John Giannandrea was apparently confident he could fix Siri with the right training data and better web-scraping for answers to general knowledge questions. Senior leaders didn’t respond with a sense of urgency to the debut of ChatGPT in 2022; Giannandrea told employees that he didn’t believe chatbots like ChatGPT added much value for users.

[…]

The report claims that the demo of Apple Intelligence’s most impressive features at WWDC 2024, such as where Siri accesses a user’s emails to find real-time flight data and provides a reminder about lunch plans using messages and plots a route in maps, was effectively fictitious. The demo apparently came as a surprise to members of the Siri team, who had never seen working versions of the capabilities.

Ryan Christoffel:

It’s pretty shocking that Siri’s new design was, per this report at least, the only new feature that was ready for testing within Apple last June.

I’m still surprised that they shipped the new UI with the old Siri, thus confusing people and also missing their chance to get people to give the improved version a fresh look.

Also, Ma closes with a fascinating tidbit: Federighi has reportedly already changed one big AI policy for its work on Siri.

Whereas Apple engineers previously had been directed to build features only using internal models, now using open-source third-party models has been given the green light.

Previously:

• Siri Product Knowledge
• Apple Shifts Siri From Giannandrea to Rockwell
• Rotten
• Apple Delays “More Personalized Siri” Apple Intelligence Features
• Turmoil Behind Siri

Update (2025-04-11): John Gruber:

What Ma describes is a scenario where Walker missed the fact that the whole forest sucked and didn’t work, while focusing on one or two nice trees.

[…]

I’m not even sure eliminating the requirement to use the verbal “hey” prefix was a win at all. It’s purely anecdotal and personal, but I think I get more unwanted invocations now than I did when “Hey Siri” was the required prompt.

Same here.

Hundreds of engineers for a machine learning team outside Apple’s AI/ML division sounds like the definition of dysfunction.

Federico Viticci:

My read of this part is that Federighi might have instructed his team to use distillation to better train Apple’s in-house models as a way to accelerate the development of the delayed Siri features and put them back on the company’s roadmap. Given Tim Cook’s public appreciation for DeepSeek and this morning’s New York Times report that the delayed features may come this fall, I wouldn’t be shocked to learn that Federighi told Siri’s ML team to distill DeepSeek R1’s reasoning knowledge into a new variant of their ∼3 billion parameter foundation model that runs on-device. Doing that wouldn’t mean that iOS 19’s Apple Intelligence would be “powered by DeepSeek”; it would just be a faster way for Apple to catch up without throwing away the foundational model they unveiled last year (which, supposedly, had a ~30% error rate).

Tim Hardwick:

Apple’s current struggles with Apple Intelligence and Siri began in early 2023 when AI head John Giannandrea sought approval from CEO Tim Cook to purchase more AI chips for development, according to a new report from The New York Times.

Cook initially approved doubling the team’s chip budget, but CFO Luca Maestri reportedly reduced the increase to less than half that amount, and instead encouraged the team to make existing chips more efficient.

The lack of adequate GPU resources meant Apple’s AI team had to negotiate for computing power from providers like Google and Amazon.

At the time, Apple’s data centers had about 50,000 GPUs that were more than five years old – far fewer than the hundreds of thousands of chips being purchased by competitors like Microsoft, Google, and Meta.

M.G Siegler:

Honestly, reading these reports, even if just some of the details are as bad as they sound, it’s hard to see how Giannandrea and Walker can stay at Apple. Again, it’s undoubtedly not as cut and dry as it may seem on paper, with Mickle getting intel that Giannandrea had long ago asked for the ability to buy up more GPUs to get their AI work done, only to have it blocked by CFO Luca Maestri – someone who is, incidentally, no longer in that role.

[…]

But really, this all speaks to what is clearly a larger issue: a lot of this falls – or should fall – on Tim Cook.

There’s just no sugarcoating it. He’s the CEO. The buck literally stops with him – if the CFO overrides him on the GPU request, he should veto that, as he apparently had already signed off. If the CFO changed his mind, okay, but why? Just cost? That would indicate a colossal misreading of the situation.

Stephen Hackett:

For a company that says it doesn’t like looking back at its own history, very often, Apple makes decisions like it’s the late 1990s and the company is on the verge of failure.

Only for core technologies like Macs, high-end processors, QA, and Siri. When it’s the car or Vision Pro or Apple TV+ the sky’s the limit.

Apple Apple Intelligence Artificial Intelligence Craig Federighi iOS iOS 18 Mac macOS 15 Sequoia Siri Tim Cook

3 Comments RSS · Twitter · Mastodon

---

---

---

Leave a Comment

Name

E-mail (will not be published)

Web site

Δ