In response to critiques about the performance of its AI features especially notification and email summarie Apple has outlined a new strategy focused on improving its AI models without compromising user privacy. The company announced that it will use synthetic data alongside a technique known as “differential privacy” to fine-tune its systems.
Rather than relying on real user content, Apple creates synthetic messages that mimic the format and characteristics of actual data. These messages are then transformed into “embeddings,” which capture key aspects such as language, topic, and length. A select group of user devices only those with analytics sharing enabled receive these embeddings and compare them against local data to assess accuracy. The feedback helps Apple improve its AI without ever collecting raw personal data.
This method is already being applied to enhance Apple’s Genmoji models and will soon extend to features like Image Playground, Image Wand, Memories Creation, Writing Tools, and Visual Intelligence. Email summaries are also set to benefit from this private, synthetic data-based approach.
For more tech updates, visit DC Brief.
