Large language models in a Chinese University: Navigating the Opportunity-Challenge Nexus

• In Event: From Adoption to Governance: Implementing AI and Digital Competence in Higher and Vocational Education Worldwide

Proposal

This article examines engagement with large language models (LLMs) in Chinese higher education, focusing on their use in teaching and research in early 2025. A case study based qualitative research approach, focusing on LLM users at a Chinese university recognized for its teacher education, was supported by a review of contemporary LLM related literatures. Researching the issue of the ‘danger/promise paradox’ of a rapidly evolving technology involved adapting Celia Lury’s concept of ‘problem spaces’ within a wider geo-political economy framework.The primary research suggests a relatively ‘unmediated’ relationship between users and the new technology, reflected in calls from interviewees for more guidelines and support both nationally and institutionally.Without ‘progressive mediation,’ it is argued that the positive potential of LLMs could be reduced and dangers amplified. The final section suggests that increased strands of mediation activity could be key to a ‘recomposing’ of the LLM problem space.

To guide the research, the process and analysis were framed by an overarching question and three sub-questions. Specifically, in what ways is the LLM challenge/opportunity nexus reflected in teacher perceptions in the case-study institution?
a. How are LLMs being used in teaching and research in this Chinese university across various disciplines?
b. What are the factors affecting the adoption of LLMs by educators in the context of China-specific and wider global trends?
c. How far is the application of LLMs being affected by mediating factors such as external and institutional guidelines and the provision of resources and training?

The analysis draws upon two complementary frameworks. First, Celia Lury’s concept of “problem spaces” conceptualizes LLM adoption as a dynamic, evolving constellation of actors, challenges, and contexts, allowing for exploration of emergent tensions. Second, a political economy perspective, adapted through the 45-degree mediation model, examines vertical forces, horizontal forces, and the mediating mechanism. Together, these frameworks illuminate how LLMs both challenge and reconfigure the ethics and practices of higher education.

The study employed a compositional methodology to capture the complexity of the rapidly evolving LLM landscape. This included:Review of contemporary literature on LLM development globally and in China;Institutional case study in a first-tier Chinese university known for teacher training;Qualitative data collection: 20 semi-structured interviews conducted in two phases (January and March 2025), supplemented by a focus group discussion in April 2025. Participants represented education, technology, and engineering faculties and were selected for their active engagement with LLMs and Iterative analysis guided by problem-space and political economy perspectives, incorporating reflexive use of both Chinese and Western LLMs as research tools. The research adhered to BERA’s (2024) ethical guidelines, ensuring informed consent, transparency, and data anonymization.

The research has four findings. First, awareness and usage,awareness of LLMs was widespread across disciplines, though patterns varied. Engineering and technology staff engaged with LLMs for debugging and algorithm testing, while education faculty emphasized lesson planning and student engagement. A major turning point was the “DeepSeek Moment” of February 2025, when a Chinese-developed LLM overtook ChatGPT in popularity, symbolizing growing self-reliance in AI capacity. Second, Applications in Teaching and Research,LLMs were widely applied in teaching,like lesson preparation, case generation, personalized student support, and research,such as literature reviews, text refinement, and translation. Some educators used LLMs as critical thinking partners, having students evaluate AI-generated responses to stimulate discussion. Such practices suggest pathways toward peace-oriented education, where AI fosters dialogue, reflection, and equitable intellectual empowerment. Third,Challenges,participants highlighted significant risks:technical (hallucinations, unreliable outputs), ethical (plagiarism, academic misconduct), and equity-related (subscription fees, uneven digital literacy). A key concern was overreliance, particularly among students, which could erode trust in authentic scholarship and destabilize the foundations of academic peace. Finial,A striking finding was the near absence of institutional guidelines, training, or systematic support. Most staff learned LLM use independently or informally, leaving them exposed to uncertainty and risk. Participants consistently called for structured mediation through clear policies, ethical standards, training programs, and communities of practice.

Reflecting this situation, research participants were anxious to see more guidance concerning the application of LLMs, both generally and in relation to teaching, learning and research. There may be several reasons for this apparent paradox. At the national level it may reflect a growing political confidence of the potential of Chinese generative AI globally and nationally and thus a reduced appetite to stifle this innovation. There could also be factors, unique to the institutional context. Or it could be the result of a research snapshot in time in which the technology has been accelerating and with regulation, national and institutional, running to catch up. Only further research can cast light on these apparent paradoxes.Potential factor affecting LLM adoption and application – technological, political, cultural, institutional and professional –were built into the analytical framework. Of all of these, the factor that stood out in early 2025 was the emergence of the Chinese LLM DeepSeek. As an advanced and open-source model, the data suggest that it changed the local LLM adoption landscape while, at the same time, popularising the powers of Chinese technological innovation. However, there is little evidence thus far that DeepSeek radically affected the application landscape. Evidence from the interviews and focus group suggests that the research participants possess sufficient expertise to make a significant contribution to the development of institutional guidelines, shape training programmes both for themselves and their colleagues, and to consider the wider intellectual and educational implications of this promising but problematical technology. This latent capability points to the role of ‘holistic mediation’ in rebalancing relations in the LLM opportunity/challenge nexus. At the same time, the wider Chinese regulatory and innovation landscape is evolving with the 2025 goals of the ‘New Generation Artificial Intelligence Development Plan’ (S, including a focus on the promotion of responsible and ethical AI development. In view of this rapidly changing landscape, the LLM ‘problem space’ should be regarded as open, with opportunities to further research tensions between an evolving technology, its political and ethical contexts and professional action.

Authors

• rong liying

• xinyue cui, 首都师范大学

• yanchun chen, Capital Normal University