Amidst criticism regarding the lackluster performance of its AI solutions, particularly in notification summaries, Apple announced on Monday its plans to enhance its AI models by leveraging user data while maintaining privacy through synthetic data techniques.
Apple intends to implement a strategy known as “differential privacy.” This involves creating synthetic data which mimics the characteristics of actual user data, yet does not encompass any real user-generated content. By generating a broad range of synthetic messages covering various topics, Apple is able to construct representative datasets. Each synthetic message is then transformed into an embedding—a representation capturing key aspects such as language, topic, and length.
Once these embeddings are formulated, they are dispatched to select user devices that have opted into sharing device analytics with Apple. These devices subsequently compare the embeddings against a sample of actual emails, enabling Apple to determine which embeddings are the most accurate and valuable for enhancing its AI models.
This innovative method is particularly significant for Apple’s Genmoji models and will soon extend to other applications, including Image Playground, Image Wand, Memories Creation, and Writing Tools, as well as Visual Intelligence. Furthermore, Apple plans to gather insights from users who agree to share device analytics using this synthetic data approach to refine email summary features.
Overall, Apple’s commitment to using synthetic data aims to bolster the effectiveness of its AI offerings while respecting user privacy, addressing ongoing critiques, and ensuring that future developments in AI technology are both innovative and user-centric.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
