In recent years, significant advancements in educational technology, particularly the rise of online learning platforms, have transformed the way students engage with educational content. This study developed an intelligent digital human-based teaching system that leverages the capabilities of large language models to provide personalized, interactive, and real-time support.

This research article presents an intelligent digital human-based teaching system designed to address limitations of traditional pre-recorded courses—such as lack of real-time interaction, personalization, and emotional support—and bridge gaps between pre-recorded flexibility and live course engagement. Developed by a team from Chinese universities, the system integrates fine-tuned large language models (LLMs) for resource understanding and teaching script generation, advanced computer vision/audio synthesis for personalized digital instructor creation (replicating human appearance, gestures, and voice), automated lecture video synthesis, and an interactive Q&A module with empathetic, profile-based responses.

Two case studies (on resource understanding and primary school programming education) validate its efficacy: the fine-tuned LLM outperformed general LLMs (e.g., Llama3) in content accuracy and pedagogical quality, and digital human instruction improved student engagement (85% vs. 65% finding lessons interesting) and learning outcomes (average posttest score 87 vs. 75) compared to traditional methods. The article also discusses future directions, including broader educational applications, ethical considerations, scalability, and cross-disciplinary tech integration, aiming to advance personalized, accessible digital education.

The work titled “Advancements in Digital Humans for Recorded Courses: Enhancing Learning Experiences via Personalized Interaction”, was published on Frontiers of Digital Education (published on September 22, 2025).
DOI: 10.1007/s44366-025-0072-9