Apple: large language models can be run on the iPhone

Apple's AI researchers have deployed large language models on memory-limited devices through flash memory utilization techniques that promise to give future iPhones more powerful Siri and other AI features. They came up with a clever flash memory utilization technique, windowing and row-column bundling, that allows models to run in limited memory. This breakthrough opens up new possibilities for AI capabilities on the iPhone, and Apple has plans to integrate AI into more apps, including Siri and other features. Apple is also working on its own generative AI model, "Ajax," designed to compete with OpenAI's GPT-3 and GPT-4.