The most frustrating part of iOS today, for me, is the voice recognition in iMessage. It feels like something from a decade ago. I hope they use a real LLM to improve accuracy. According to reports they will use an on-device large language model. Can a good-enough LLM run on the phone?