Assoc. Prof. Dr. Bor Gregorčič (U Uppsala): How understanding large language models can help inform their use in physics education

Datum objave: 2. 5. 2024
Ponedeljkov fizikalni kolokvij
14.15 - 15.15

In late 2022 the public release of OpenAI's ChatGPT caused many people to start thinking about the implications of the so-called generative artificial intelligence. ChatGPT is a representative of a specific type of generative artificial intelligence - large language models (LLMs). LLMs are already disrupting how we do knowledge work. The list of affected areas includes education at different levels and in most disciplines, including physics. Physics, with its specific ways of reasoning and problem solving approaches, is an especially interesting context in which we can explore the capabilities of LLMs, consider their limitations, and potential for their educational use. In my talk, I will attempt to explain in non-technical terms, how LLMs work and how understanding of their functioning, even on a basic level, can help us understand their performance on introductory physics tasks, and vice-versa, how studying their performance on physics tasks can inform our understanding of how they function. I will present some examples of ChatGPT's responses to physics problems and explore how its performance depends on the way we prompt the model. Finally, I will invite the participants to engage in a discussion of the possible roles LLMs could play in the future of physics education and physics as a scientific discipline.
Cookies, tea