After Apple had to delay its most important Apple Intelligence feature, it seems the company is catching up on what it needs to do to deliver advanced AI capabilities.
According to Bloomberg‘s Mark Gurman, Apple will start to analyze customers’ data on devices to improve the Apple Intelligence platform. The journalist explains that Apple currently trains its AI models with synthetic data without personal details.
As Gurman explains, the company creates fake inputs of what customers could search, write, or do. With the new approach, Apple will compare a user’s email, for example, with its fake input, to improve its AI algorithm on-device.
Gurman says these insights “will help the company improve text-related features in its Apple Intelligence platform, such as summaries in notifications, the ability to synthesize thoughts in its Writing Tools, and recaps of user messages.”
The journalist writes that these features are only for users who “are opted into device analytics and product improvement capabilities.” With that, it will track how the model responds when multiple people have made the same request.
It’s unclear if this change will fundamentally improve how AI models are trained. However, we might start giving more precise input on users’ requests. Besides many other differences, what sets OpenAI and Google apart is that they collect data for their LLMs based on everyone’s information. If Apple succeeds in this more privacy-first approach, it will finally deliver on its promise of almost a year.
In the past few weeks, Apple has had to address the previously announced Siri feature for on-screen awareness, which has been postponed indefinitely.
Since then, BGR has reported on a strategic shake-up for several Apple teams. Siri now has new bosses, and Apple is rushing to deliver the on-screen awareness capabilities as soon as this fall—even though the company might have to delay them until 2026.
Since this Siri feature was the last one missing from the original Apple Intelligence announcement, the company has nothing else to promote. In addition, the already available features aren’t anything groundbreaking. Writing Tools barely worked, and Summarize Notification got an update to stop summarizing news articles’ headlines as it was getting the information wrong.
