TokenVerse is a research-driven game-based learning environment that helps players understand how large language models work and how they affect human–AI interaction.
The game takes place inside Ditto, a fictional company that operates a large language model service. As players move through the company's interfaces, departments, and internal systems, they gradually uncover how language is processed, how responses are produced, and how technical and organizational decisions shape the behavior of an AI system.
TokenVerse is developed as a series of connected chapters within the same company. Each chapter explores a different layer of large language models through its own environment, narrative, and gameplay mechanics.
The first chapter begins at Ditto's user-facing service interface. Players process user requests and encounter human–AI interaction risks, including emotional dependency, self-harm-related conversations, and the reinforcement of delusional beliefs. By moving between the company's normal operations and an overloaded system state, players investigate how user needs, model behavior, safety policies, and corporate priorities can come into conflict.
Future chapters will lead players deeper into the company, introducing foundational concepts such as tokenization and transformer-based language processing. These technical processes will be represented through playable spaces and systems, allowing players to explore how language is broken down, interpreted, and transformed before a response reaches the user.
Built in Unity, TokenVerse combines game design, interactive storytelling, and educational research to make the hidden systems behind large language models visible, experiential, and open to critical reflection.
