On-device intelligence (ODI) is an emerging technology that combines mobile computing and AI, enabling real-time, customized services without network reliance. Researchers have proposed methods such as federated learning (FL) and transfer learning (TL) to balance AI training needs with device limitations and privacy concerns. However, these paradigms struggle to balance privacy and performance constraints. To address this, the researchers from IEEE introduce Privacy-Preserving Training-as-a-Service (PTaaS), a robust paradigm offering privacy-friendly AI model training for end devices. PTaaS delegates core training to remote servers, generating customized on-device models from anonymous queries to uphold data privacy and alleviate device computation burden.