Implementasi Large Language Model Menggunakan Gemini 1.0 Pro dalam Permainan Role-Playing untuk Menciptakan Emergent Gameplay

Abstract

The development of Large Language Models (LLMs) has driven advancements in all technological fields, including the entertainment media dominated by video games. LLMs possess role-playing capabilities that can assist game designers in both designing and directly integrating LLMs into games. The utilization of LLMs within games is still considered under-explored, particularly in the RPG genre which has the potential to leverage LLMs to create emergent gameplay experiences. Emergent gameplay arises from the dynamic interaction between game mechanics, player decisions during gameplay, and unexpected events. This emergent nature allows games to provide more unique narratives, exploration, and surprises for players, leading to a more immersive experience. This research utilizes the game project "Rudantara," an RPG and Rogue-like game with procedurally generated maps and play areas. The implementation of the LLM focuses on a companion character that accompanies the player throughout their adventure. The companion's actions are controlled by the LLM based on prompts designed in accordance with the companion's sensor, memory, and behavior modules. Each module gathers real-time data from the game environment to provide the latest information to the prompt designer script, which then requests responses from the LLM at specific intervals and under certain conditions. This research uses the Gemini 1.0 Pro API as the LLM integrated into the game. Testing in this research was carried out by giving questionnaires to 8 users to assess the potential of applying LLM to role-playing games and whether emergent gameplay could be created.