Expressing Educational Content with Extensive Language Models: A Question of Context

In the context of Generative Artificial Intelligence and large language models (LLMs) applied to education, this study focuses on the challenge of adequately contextualizing these technologies for their effective application. The central problem lies in the tendency of LLMs to generate decontextualized responses, which compromises their reliability and usefulness. To address this problem, a methodology has been adopted that combines Knowledge Injection (KI) and the Retrieval Augmented Generation (RAG) technique. This methodology involves the integration of relevant texts and vector databases with pre-trained language models, mitigating LLM hallucinations and allowing them to access a richer and more up-to-date context. The results obtained, particularly in the case study of the contents of the subjects Web Projects I and Databases at La Salle Campus Barcelona, Ramon Llull University, allow us to propose ways to structure the text of educational contents according to the specific tasks to be executed by the LLMs. This approach not only improves the relevance and accuracy of the models’ responses, but also facilitates the interaction with the contents of the subjects, adapting to the specific educational needs. This research emphasizes the importance of a contextualized approach to the use of LLMs, proposing a research path for their practical and effective application in education. The application of these techniques opens up new possibilities for the personalization and improvement of the educational experience through the advanced use of generative artificial intelligence.