Apple is planning to analyze user data to improve its large language model (LLM) software while upholding user privacy.
The company has been using synthetic data to train its artificial intelligence (AI) models but has found that method to be ineffective, Apple wrote in a Monday (April 14) blog post.
Now, Apple will still use synthetic data as a starting point but will compare the generated text to a sample of emails from participating user to determine which generated output best lines up with real-world messages.
“Only users who have opted-in to send Device Analytics information to Apple participate,” Apple said in the blog post. “The contents of the sampled emails never leave the device and are never shared with Apple. A participating device will…