Leveraging GPT-4 for Accuracy in Education: A Comparative Study on Retrieval-Augmented Generation in MOOCs [r-libre/3251]

Miladi, Fatma; Psyché, Valéry et Lemire, Daniel (sous presse). Leveraging GPT-4 for Accuracy in Education: A Comparative Study on Retrieval-Augmented Generation in MOOCs. Dans AIED 2024 - 25th International Conference on Artificial Intelligence in Education (LBR Track). New York City : Springer-Verlag, coll. « Communications in Computer and Information Science ».

Fichier(s) associé(s) à ce document :

PDF - Poster_2024.pdf
Contenu du fichier : Manuscrit soumis (avant évaluation)
Licence : Creative Commons CC BY.

Télécharger

Catégorie de document :	Communications dans des actes de congrès/colloques
Évaluation par un comité de lecture :	Oui
Étape de publication :	Accepté (sous presse)
Résumé :	Large Language Models (LLMs), such as Generative Pretrained Transformers (GPTs), have demonstrated remarkable capabilities in natural language processing (NLP). However, these models often encounter challenges such as inaccuracies and hallucinations, which can undermine their utility. Retrieval-Augmented Generation (RAG) has emerged as a promising approach to enhance model accuracy and reliability by integrating external databases. This study investigates the use of RAG to improve the accuracy of GPT models in educational settings, particularly within the realm of Massive Open Online Courses (MOOCs). Through a comparative analysis of various GPT model iterations, we observed a significant improvement in accuracy, increasing from 60% with GPT-3.5 to 80% using the RAG-augmented GPT-4. This enhancement highlights the considerable potential of RAG-augmented GPT models in improving the accuracy of content generation. Such enhanced accuracy suggests revolutionizing assessment methodologies and learning experiences, fostering an educational environment that is more interactive and tailored to individual needs
Déposant:	Lemire, Daniel
Responsable :	Daniel Lemire
Dépôt :	09 mai 2024 13:31
Dernière modification :	09 mai 2024 13:31

Actions (connexion requise)

RÉVISER