Apple details how it trained its new AI models: 4 interesting highlights

Apple has unveiled new on-device and cloud-based AI models, releasing a technical report detailing their development. The local model is split into two blocks for improved efficiency, while the cloud model uses a custom Parallel-Track Mixture-of-Experts architecture. Multilingual representation increased by 275%, with data sourced from web crawling, licensed content, and synthetic generation. These advancements aim to enhance Apple's AI capabilities across its products and services.