Bias in Large Language Models in Higher Education

Team Name: Unaite

Use of AI Tools: We used Claude Design and Eleven Labs to edit our presentation video. The AI ​​tools allowed us to save time on source research and writing.

External Contributions and Feedback: We were primarily inspired by Anthropic's YouTube channel, which features many experts in AI and ethics. We are also members of an association that organizes numerous conferences on research in mathematics and computer science in companies like Hugging Face and Meta, so we are constantly learning about the subject. Finally, the work of CREST and the work of Grégoire Borst (Professor of Developmental Psychology and Educational Neuroscience at Paris-Descartes University) caught our attention.

Initial Contribution: Mapping teaching style spectra in LLMs : an initial contribution

Final Contribution:

A reflection on the process (important)

Explain how your contribution evolved during phase 2:

The feedback helped us refine our thinking, particularly regarding the practical aspects such as decision support. For example, we found Bruno DE LIEVRE's comment on the possibility of conducting a pilot comparing two different uses of LLMs for revision particularly relevant, as well as the warnings about the need to clarify our evaluation criteria for "cognitive debt" or "psychological support," which can indeed be vague concepts based more on the (albeit real) impressions of users and, in my case, on the construction of the model itself.

What changes did you make as a result?

We sought to develop a practical LLM selection grid adapted to the specific and varying revision needs of students, in order to inform and guide these practices in universities and make access to AI equitable and sustainable.

How did this process help improve your final proposal?

This process allowed us to realize that moving from science to concrete political and educational decision-making is not easy. And that consultation with the public or other researchers is important to produce a result that is understandable to everyone and, in a way, effective.

Video presentation

https://drive.google.com/drive/folders/10a0J7ugzP01ZJf2dwFTGfKn6YSww5_5_