More trustworthy generative AI through hallucination reduction

• Main Author: He, Guoshun
• Other Authors: Alex Chichung Kot
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Hallucination; Generative AI; Benchmark
• Online Access: https://hdl.handle.net/10356/177162
• Institution: Nanyang Technological University

In the current wave of rapid development in artificial intelligence, AI is being widely applied in various industries. In this process, the reliability of artificial intelligence is receiving increasing attention. In current research, people largely focus on hallucination to study the reliability of multimodal large-scale language models. The existence of the hallucination problem can lead to the output of misleading information for users, as well as serious problems such as causing security risks and even a decrease in trust in large models. This report explores the impact of hallucination on the reliability of multimodal large-scale language models, starting from the perspective of hallucination, and employs a set of evaluation criteria to compare the reliability among different large-scale language models.