Li Li

Institute of Higher Education, iFLYTEK Co., Ltd.

Abstract：

Since 2024, global Large Language Models (LLMs) have demonstrated a "step-change" breakthrough in parameter scale, multimodal fusion, and deep reasoning capabilities, providing a replicable technological foundation for higher education. In August 2025, China's State Council issued the "Opinions on Deepening the Implementation of the 'AI Plus' Initiative," explicitly listing "AI + Higher Education" as a key direction for developing new quality productive forces for the first time. Based on the practical implementation of domestic large models like iFlytek's Spark, DeepSeek, and Qwen3 across 27 "Double First-Class" universities, this paper systematically examines the technical pathways, training methodologies, and governance experiences of Domain-Specific Large Models (DSLMs). It proposes a novel five-element synergistic model encompassing "Computing Power - Data - Model - Scenario - Evaluation." This framework aims to provide a self-sufficient, mutually reinforcing (between teaching and learning), and data-driven Chinese pathway for university discipline construction and the cultivation of top-notch innovative talent.

Key Words：

domain-specific large model; generative Artificial Intelligence; higher education; talent cultivation; iFlytek spark