Apple’s Worldwide Developers Conference keynote has come to a close — and the company had a whole lot to share. We got our first look at the AI features coming to Apple’s devices and some major updates across the company’s operating systems.
If you missed out on watching the keynote live, we’ve gathered all the biggest announcements that you can check out below.
↫ Emma Roth at The Verge
Most of the stuff Apple announced aren’t particularly interesting – a lot of catch-up stuff that has become emblematic of companies like Google, Apple, and Microsoft when it comes to their operating systems. The one thing that did stand out is Apple’s approach to offloading machine learning requests to the cloud when they are too difficult to handle on device. They’ve developed a new way of doing this, using servers with Apple’s own M chips, which is pretty cool and harkens back the days of the Xserve.
In short, these server are using the same kind of techniques to encrypt and secure data on iPhones, but now to encrypt and secure the data coming in for offloaded machine learning requests.
The root of trust for Private Cloud Compute is our compute node: custom-built server hardware that brings the power and security of Apple silicon to the data center, with the same hardware security technologies used in iPhone, including the Secure Enclave and Secure Boot. We paired this hardware with a new operating system: a hardened subset of the foundations of iOS and macOS tailored to support Large Language Model (LLM) inference workloads while presenting an extremely narrow attack surface. This allows us to take advantage of iOS security technologies such as Code Signing and sandboxing.
↫ Apple’s security research blog
Apple also provided some insight into where its training data is coming from, and it claims it’s only using licensed data and “publicly available data collected by our web-crawler”. The words “licensed” and “publicly available” are doing a lot of heavy lifting here, and I’m not entirely sure what definitions of those terms Apple is using. There are enough people out there who feel every piece of data – whether under copyright, available under an open source license, or whatever – is fair, legal game for ML training, so who knows what Apple is using based on these statements alone.
From Apple’s presentations yesterday, as well as any later statements, it’s also not clear when machine learning requests get offloaded in the first place. Apple states they try to run as much as possible on-device, and will offload when needed, but the conditions under which such offloading happens are nebulous and unclear, making it hard for users to know what’s going to happen when they use Apple’s new machine learning features.