Apple Intelligence is here

Apple's OS 18.1 developer beta is out. As usual, they focus on user experience and want to build AI into the workflow. This is the way to go. 

There's a waitlist but what it includes Personal intelligence integrated throughout your apps and experiences to simplify and accelerate everyday activities. On top of this new tools for writing and images. This is a real challenge for tools like Grammarly,   Jasper, Copy.ai, Writer, Writesonic and WordAi.

But the big news is a massive upgrade to Siri with a more natural, contextually relevant and personal to you.

And as we know, there will be groundbreaking privacy protection with on-device processing combined with Private Cloud Compute, to protects your privacy at every step. What will be interesting will be the balance between what is done on device versus the cloud.

To run things locally it looks like there's one large foundational model and they use adapters to allow local processing. Damn clever...

Size isn't everything

Apple's entry to the market is yet another example of competition upping everyone's game. They're using their own foundational model plus GPT. This is moving fast and is not all about who has the biggest model, although, if the claims made about GPT5 are right, it may blow everything out of the water. 

They have introduced their foundation language models, the one's that power Apple Intelligence features. These include a ∼3 billion parameter model optimised for efficient on-device operation and a larger server-based language model designed for Private Cloud Compute. Both models are engineered to perform a their AI features efficiently, accurately, and responsibly.

They are up front on model architecture, data used for training, the training process, optimisation for inference and evaluation results.

For a longer introduction, here's a fuller video:

I'm seriously thinking about going back to the iPhone from my Google Pixel.