Thesis
Exploring the applications and limitations of large language models: a focus on ChatGPT in virtual NPC interactions
Master of Science (M.S.), Drexel University
Dec 2023
DOI:
https://doi.org/10.17918/00001888
Abstract
Large Language Models (LLMs), like ChatGPT and Google's Bard, have gained attention for their versatility and accessibility. This study delves into ChatGPT, exploring its potential to elevate virtual Non-Player Character (NPC) interactions, especially in domains such as law enforcement, mental health, and emergency services training. The primary goal was to assess ChatGPT's strengths and weaknesses through qualitative research, identifying opportunities for system enhancements. The evaluation criteria encompass parameters vital to digital media storytelling, including empathy, emotional design, and interactive storytelling. By elucidating the capabilities and constraints of LLMs like ChatGPT and their role in enhancing virtual NPC interactions, this research aims to contribute valuable insights to domains reliant on human-computer interactions and emotional engagement. Additionally, the study integrates scenarios used in training police officers into the researched application, offering best practices for configuring an NPC powered by an LLM. While the emotional parameters showed overall success, there were occasional inconsistencies and unconventional responses. While testing the application for training in verbal de-escalation, the basic scenario seemed to work better than its moderate and complex counterparts due to the straightforward nature of the scenario and since it involved only two characters, making use of the basic techniques involved in learning verbal de-escalation.
Metrics
127 File views/ downloads
151 Record Views
Details
- Title
- Exploring the applications and limitations of large language models
- Creators
- Ayaan Anand
- Contributors
- Emil Polyak (Advisor)
- Awarding Institution
- Drexel University
- Degree Awarded
- Master of Science (M.S.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Number of pages
- vi, 106 pages
- Resource Type
- Thesis
- Language
- English
- Academic Unit
- Digital Media; Drexel University; Antoinette Westphal College of Media Arts and Design
- Other Identifier
- 991021807413704721