Apple AI Learns From Fake Emails While Keeping Yours Safe

Apple plans to train its AI models using synthetic emails despite delaying its Personalized Siri feature until next year. According to Bloomberg, company employees believe the feature might launch earlier than announced. The tech giant detailed its training approach in a Machine Learning Research blog post that explains how synthetic data helps develop Apple Intelligence.

Reports indicate Apple trails competitors in AI development because its synthetic data strategy faces limitations. These include difficulty understanding trends for summarization tools that need longer sentences or complete emails. The company addressed this challenge with new technology, comparing synthetic data to samples of recent user emails without compromising privacy.

The process creates various synthetic messages like "Would you like to play tennis tomorrow at 11:30 AM?" without accessing actual user emails. The system generates embeddings that capture key message dimensions such as language, topic, and length. These embeddings go to devices opted into Device Analytics, where they match with small samples of user emails through differential privacy.

Apple says this method helps understand overall trends without gathering personal information. Bloomberg reports the company will release this technology in the upcoming iOS 18.5 and macOS 15.5 beta versions. Users can find additional information about the process on Apple's website.
 

Attachments

  • Apple AI Learns From Fake Emails While Keeping Yours Safe.webp
    Apple AI Learns From Fake Emails While Keeping Yours Safe.webp
    51.5 KB · Views: 17

Trending content

Latest posts

Top