Department of Mathematics and Computer Science

Large Language Models (LLMs) are increasingly becoming part of everyday life, yet interacting effectively with them requires users to understand that their behavior is not fixed. Through prompting, feedback, and customization, users can steer an LLM towards different outputs and communication styles. At the same time, communicating one's preferences to an AI can itself be challenging.

How can this knowledge be conveyed in a playful and memorable way?

This thesis investigates how LLM customization can be transformed into an interactive game. The game should allow players to experience how different instructions and customization choices influence an LLM's behavior. Rather than explaining prompting through a tutorial, the goal is to create a playful experience in which players discover relevant principles through interaction.

The exact learning objective and game concept are deliberately open. Possible topics include understanding that the same LLM can communicate very differently depending on how it is instructed, learning strategies for expressing preferences more effectively, exploring different levels of human and AI control, or understanding that LLM outputs can be steered but remain inherently imperfect and non-deterministic.

The final game may be digital, physical, or hybrid and should be suitable for use in public science communication settings. Previous research has shown a broad range of digital, hybrid, and unplugged approaches for playfully communicating concepts related to artificial intelligence.

Procedure

First, relevant literature on playful learning, prompting, and Human-AI interaction is reviewed. Based on this literature, a concrete learning objective is defined: what should players understand about interacting with LLMs after playing the game?

Next, different game concepts are developed through an iterative design process. Concepts may draw on existing research on LLM customization and novel prompting interfaces, but creativity in the choice of game mechanics is explicitly encouraged.

A selected concept is then implemented as a functional prototype and refined through iterative testing. Finally, the resulting game is evaluated with users. The evaluation examines both the player experience and whether the intended learning objective is successfully communicated, for example through observations, short interviews, or pre- and post-game measures.

The thesis results in a playable prototype as well as design insights for communicating principles of LLM interaction and customization through playful experiences.  

References

• Gennari, R., Melonio, A., Pellegrino, M. A., & D'Angelo, M. (2023). How to Playfully Teach AI to Young Learners: A Systematic Literature Review. Proceedings of CHItaly '23.

• Gmeiner, F., et al. (2025). Intent Tagging: Exploring Micro-Prompting Interactions for Supporting Granular Human-GenAI Co-Creation Workflows. Proceedings of CHI '25.

• Riche, N., et al. (2025). AI-Instruments: Embodying Prompts as Instruments to Abstract & Reflect Graphical Interface Commands as General-Purpose Tools. Proceedings of CHI '25.

• Zhang, R., Ringland, K. E., Paan, M., Mohr, D. C., & Reddy, M. (2021). Designing for Emotional Well-being: Integrating Persuasion and Customization into Mental Health Technologies. Proceedings of CHI '21.