The impact of local and cloud based intelligent non player characters on the user experience in virtual reality
Despite technological advances, interactions with non-player characters in games and virtual reality often fall short of user expectations. This master's thesis compares two approaches to building an intelligent, voice-interactive non-player character: a local one using a machine learning model for semantic sentence similarity, and a cloud-based one using large language models via the OpenAI API. Both were implemented in a prototype VR restaurant game and evaluated in a within-subjects study with 19 participants, where the cloud-based approach achieved significantly higher perceived believability.