WAICF 2024, World AI Cannes Festival, 8-10 February 2024, Cannes, France
EULER (EUrecom Language model for Education and Research) aims to solve the socio-technical challenges that will bring access to personalized education. In many institutions, the high student-teacher ratio limits the level of personal interaction and individual instruction adapted to the learner, further hindering the students’ access to equitable education. Our goal is to empower students to self-learn through an interactive journey that ensures an understanding of basic concepts, hence improving their comprehension skills and fostering critical thinking. This digital tutor would act not as a direct problem-solver, but rather as a guide encouraging students to come up with solutions themselves. First, we leverage the capabilities of open LLMs, such as Llama 2 and Mixtral, through a series of improvements aimed at transforming this language model into an effective digital tutor. Next, we aime to make use of Reinforcement Learning from Human Feedback (RLHF) and fine-tuning methods to align the model with this directive approach, and to rely on Retrieval Augmented Generation (RAG) to leverage a vast array of multimodal teaching material (slides, videos, handouts, textbooks) that already exist within our educator community.
Type:
Poster / Demo
City:
Cannes
Date:
2024-02-08
Department:
Data Science
Eurecom Ref:
7595
Copyright:
© EURECOM. Personal use of this material is permitted. The definitive version of this paper was published in WAICF 2024, World AI Cannes Festival, 8-10 February 2024, Cannes, France and is available at :
See also: