Introduction and Analysis of Improving Non-Player Character Performance Based on Large Language Models

Abstract

This paper aims to study the means of using Large Language Models (LLM) to improve the performance of Non-Player Character (NPC), since game industry has high business potential and there are few studies targeted at using LLM to intelligentizated NPC. Writers first review on the history of development of both LLM and NPC, then considered four key dimensions of NPC - behavioral realism, social intelligentization, memory continuity, and dynamic response - and concluded the advanced studies related to those dimensions. The paper also does simple testing about the three games that used LLM - Whisper from the star, Justice Mobile and GTA5 AI Mod, and found that none of them developed their NPCs to communicate in a natural way, but observed typical Chat-like performance. This paper also listed several challenges and limitations of those advanced technologies, and give some predictions about the possible future developments and applications of intelligentiation NPC.