AI Models empowering teaching in higher education: current applications, practical dilemmas and pathways for improvement

As generative artificial intelligence continues to iterate, mainstream models such as ChatGPT, DeepSeek and Copilot are entering the teaching spaces of universities, and teaching processes, teacher–student relations and assessment practices are changing with them. Drawing on a literature review and case analysis, this paper maps the deployment landscape and application picture of mainstream AI models in higher education teaching and assesses both the effects that have already emerged and the problems that have been exposed. The study argues that the evidence is relatively clear for gains in teaching efficiency, in the accessibility of personalized learning support, and in the extension of teaching across time and space, whereas risks to content reliability caused by hallucination, cognitive reliance and uneven literacy among teachers and students, and the absence of regulation together with ethical controversy constitute the main obstacles to wider diffusion. On this basis, the paper proposes a coordinated approach spanning technology, literacy and institutional design.