Large language models for teaching and learning in higher education: opportunities, challenges, and future directions

Abstract

Large language models (LLMs) are rapidly reshaping the technological landscape of higher education, extending from conversational assistance and personalized learning to assessment, recommendation, and institutional services. Yet their educational value cannot be inferred from generative capability alone. What matters is how these models are incorporated into educational practice and institutional systems, and what forms of human oversight accompany their use. This editorial introduces the special issue “Harnessing Large Language Models for Teaching and Learning: Challenges, Opportunities, and Future Directions,” comprising five contributions that examine LLM-enhanced course recommendation, real-time prompting support, automated essay scoring, verifiable online education, and knowledge-augmented automation for higher-education services. Taken together, these contributions illustrate how LLMs are increasingly being used as components of educational systems rather than as stand-alone generative tools. Across the collection, the potential benefits are clearest where LLM capabilities are tied to specific educational or institutional needs. At the same time, the collection exposes a persistent gap between technical performance and evidence of educational effectiveness. Many of the questions that matter most for practice, particularly those concerning learner development, use in authentic settings, and institutional responsibility, remain insufficiently examined. We argue that future research should move beyond demonstrating what LLMs can accomplish toward establishing when, for whom, and under what pedagogical and institutional conditions their use produces meaningful educational value. This shift calls for research grounded in educational theory and tested in authentic settings, with explicit attention to how responsibility is distributed between people and AI systems. The technically diverse contributions in this special issue provide a foundation for this broader research agenda.