Researchers have developed a new system called MobileGPT that uses large language models (LLMs) to automate mobile tasks with human-like memory. This innovative approach allows for more precise and efficient learning of task procedures by breaking them down into smaller, reusable sub-tasks. According to the study, MobileGPT can automate and learn new tasks with 82.7% accuracy and adapt them to different contexts with 98.75% accuracy. The system also reduces latency and cost compared to a GPT-4 baseline.

Source: https://dev.to/mikeyoung44/llm-augmented-with-human-like-memory-for-cost-effective-mobile-task-automation-29m6