Improving Privacy and User Experience of Self-Regulated Learning Using Local Large Language Models (LLMs)

Recent years have seen a growing usage of artificial intelligence in the education field, especially for self-regulated learning with formative feedback. This evolution has dramatically transformed how educators teach and how students learn. This paper introduces a platform for self-regulated learning for students with formative feedback that enables teachers to generate content and interact with a large language model (LLM). The motivation lies in providing a reliable self-regulated learning tool to students and teachers that offers better security and privacy, as well as improved flexibility and control on the models used. The main contribution of this study lies in the development of an offline platform to offer increased security and privacy to the user. For this purpose, the platform exclusively uses large local language models. In addition, it offers greater flexibility by allowing educational staff to choose between various LLMs that fit better their needs and standards. Compared to existing solutions, the work presented here offers increased security, privacy, and model flexibility. When using a human evaluation, it gives similar quality in terms of student feedback when using Mistral Instruct. When evaluating the system with a ROUGE score, it shows that Llama 3 generates a feedback closer to the reference feedback from the LEAP platform. Despite a long response time to student feedback, it offers a novel solution for educational institutions willing to focus on privacy, security, and stricter control over the tools they use.