2-10-LLMs Roles and Limitations in Education 

LLMS' ROLES AND LIMITATIONS IN EDUCATION

 

Richárd Ádám Vécsey¹^,² *

 

¹ Hungarian Academy of Sciences, Veszprém Regional Committee — Subcommittee on Economics, Law and Social Sciences — Working Group on Social Science Research into AI and Creative Industries, Veszprém HU-8200, Hungary
² Independent Entrepreneur, Hungary
*Correspondence: vecsey.richard@gmail.com 

---

Abstract 

The widespread integration of large language models (LLMs) into the educational landscape necessitates a comprehensive analysis of their multifaceted roles and inherent limitations. This paper first explores the significant opportunities these models present for students, teachers, and researchers, including the facilitation of personalized learning pathways, the streamlining of curriculum development, and the acceleration of AI-assisted research. However, it also rigorously examines the core challenges that arise from the technology's fundamental design. These critical issues include threats to academic integrity, such as plagiarism and overreliance; the inherent risks of data bias and cultural misrepresentation; and the persistent problem of hallucination, where models generate factually incorrect or entirely fabricated information. To mitigate these risks, this paper advocates for a multi-layered approach centered on a robust verification methodology that combines both manual and algorithmic methods. It further argues that greater transparency from developers regarding model capabilities and limitations is essential for promoting the responsible use of these tools. Ultimately, the aim of this work is to equip all educational stakeholders with the knowledge required to ethically navigate this evolving AI environment, thereby ensuring that these technologies enhance academic integrity and critical thinking rather than undermine them.

 

Keywords:  artificial intelligence, large language models, education technology, learning methodology, teaching methodology, prompting, hallucination, bias, benchmarking, plagiarism