Apple’s latest AI dictation feature for the iPhone 17 Pro and iPhone Air, part of the iOS 27 developer beta, is not activated by default. This new system, powered by the AFM 3 Core Advanced model, offers a significant improvement in accuracy, especially in capitalization and punctuation, compared to its predecessor. The model, which is a 20-billion-parameter system, operates using a sparse architecture that activates only a fraction of its parameters as needed. It is stored in flash memory, employing a method called Instruction-Following Pruning to optimize processing. Evaluations show a strong preference for this model over the previous one, with notable improvements across various quality metrics. However, due to its size, the feature is limited to newer devices like the iPhone 17 Pro, iPhone Air, and select iPads and Macs with specific hardware configurations. The standard iPhone 17 is excluded due to its lower RAM capacity. The dictation model functions entirely on-device, maintaining consistent transcription quality regardless of network connectivity. It remains uncertain whether this feature will remain off by default in the final release of iOS 27 or be enabled automatically during the beta phase.
Why Isn’t Advanced AI Dictation Enabled by Default in iOS 27 Beta?
