There have been several major reported departures of late from Apple’s AI team. Today, a new report from The Information delves into the internal drama, including the story of Apple almost open-sourcing its AI models and why that didn’t happen.
Behind the scenes of Apple’s recent AI exodus
Aaron Tilley and Wayne Ma write at The Information:
Earlier this year, the Apple team working on the company’s artificial intelligence models wanted to release several of them as open-source software. Doing so would have shown Apple’s technical progress in AI while also harnessing the help of outside researchers to improve the models. Still, it also would have revealed to the public how dramatically the performance of the models dropped when Apple shrunk them to fit on iPhones, compared to versions for more powerful PCs or data center computers. Apple’s software chief, Craig Federighi, didn’t want to go the open-source route, according to two people with knowledge of the matter. In an email, he told Ruoming Pang, head of Apple’s foundation models team, that there were already plenty of open-source models from other companies to incentivize research, the people said. While releasing the models as open source would show how the software underperformed certain rival models from Alibaba and Google, Federighi said he was more concerned the public would believe Apple was making too many compromises to get the software running on iPhones, the people said.
This is the intro to Tilley and Ma’s report, and sets the stage for why several members of Apple’s AI may have departed recently.
It seems researchers within Apple believe the company’s on-device approach to AI is truly holding its models back in substantial ways. So much so that Craig Federighi reportedly said as much to departed foundation models chief, Pang.
When Apple launched Apple Intelligence last year, it did so with a heavy emphasis on user privacy.
‘On-device first’ was a key tenet of the approach. And one that, it seems, may put more limitations on the company’s AI features than it’s publicly let on.
Other interesting details from the report include:
Apple’s AI researchers being completely caught off guard when Siri’s feature delays were announced, as they had until then received nothing but positive affirmation about their work.
Also a surprise was the recent report of Apple considering using third-party AI models rather than its in-house work.
On that note, the report claims Apple has had conversations with OpenAI, Anthropic, and Google about using their LLMs to power an upgraded Siri.
The remaining foundation models members have been told Apple is “reevaluating” compensation to potentially offer more reasons to stay.
You can read the full (paywalled) report here.
What do you think of these peeks behind the scenes of Apple’s AI efforts? Let us know in the comments.
Best iPhone accessories