Content area
Full text
Abstract: With the advent of Large Language Models (LLMs), there are becoming a larger part of people's everyday lives - in their work, personal life or learning. Especially for programmers and software developers, learning how to best utilize LLMs as part of their work is becoming a crucial skill. This is especially important to students and educators have duty to prepare them to best tackle all obstacles and best utilize AI as a tool in their programming arsenal. Research into this normally focuses on the use of LLMs as tools for teaching and evaluation. This research takes another approach presenting the results from integrating LLMs as a central concept of project-based learning (PBL) semester projects for students from multiple grades from 5th semester bachelor's to 10th semester masters. All projects develop interactive systems both traditional and virtual reality and encompass a wide variety of contexts that utilize AI as a central mechanic. We show the attitude of the participating students towards utilizing LLMs, their understanding before and after the projects of AI systems and their overall satisfaction with utilizing relatively new and open technology like LLMs. To our knowledge, this is one of the first such meta-analyses of long-term effects of utilizing LLMs in students' work. We demonstrate the positive impact of utilizing LLMs on students' motivation and learning and propose several best practices to avoid some of the pitfalls associated with using these tools.
Keywords: Large Language Models (LLMs), Interactive Systems, Project-based Learning (PBL), Student Education, Programming
1. Introduction
Students increasingly use LLMs for brainstorming, writing, editing, design, development, and programming. Educators must guide their use to prevent shallow learning and over-reliance while maximizing their potential to enhance STEM education, creativity, and project development (Wu, Duan & Ni, 2024). LLMs are also much versatile tools that can be employed as parts of interactive applications for making them more robust, reliable, and boosting the possible visualization, interaction and generational possibilities with traditional procedural algorithms (Kapania et al., 2024). Harnessing their power requires creative and outside-of-the-box thinking, as literature on human-computer interactions with LLMs is still being developed and refined (Pang et al., 2025). The best way for developing such skills is by building applications that utilize AI and facing all the decisions, problems and...