9 Dec 2024 | Roberto Gallotta, Graham Todd, Marvin Zammit, Sam Earle, Antonios Liapis, Julian Togelius, Georgios N. Yannakakis
This paper provides a comprehensive survey and roadmap of the current state and future directions of large language models (LLMs) in the context of games. It highlights the growing interest and potential of LLMs across various game applications, from player assistance and non-player character (NPC) development to game design and analysis. The authors identify several roles that LLMs can take within games, including player, NPC, player assistant, commentator/reteller, analyst, Game Master (GM), game mechanic, automated designer, and design assistant. They discuss the challenges and limitations of these roles, such as memory capacity and hallucinations, and explore promising future directions for research. The paper aims to serve as a foundation for groundbreaking research and innovation in the intersection of LLMs and games.This paper provides a comprehensive survey and roadmap of the current state and future directions of large language models (LLMs) in the context of games. It highlights the growing interest and potential of LLMs across various game applications, from player assistance and non-player character (NPC) development to game design and analysis. The authors identify several roles that LLMs can take within games, including player, NPC, player assistant, commentator/reteller, analyst, Game Master (GM), game mechanic, automated designer, and design assistant. They discuss the challenges and limitations of these roles, such as memory capacity and hallucinations, and explore promising future directions for research. The paper aims to serve as a foundation for groundbreaking research and innovation in the intersection of LLMs and games.