The year 2023 saw more changes than 50 years of artificial intelligence technology development.
This study categorizes the main features of GPT-4.0 into multimodal capabilities, language understanding, processing, performance, and accuracy. Based on t...
The year 2023 saw more changes than 50 years of artificial intelligence technology development.
This study categorizes the main features of GPT-4.0 into multimodal capabilities, language understanding, processing, performance, and accuracy. Based on the five factors of GPT 4.0, this paper proposes a technical/social engineering governance model to prevent vulnerabilities in large-scale language models and block penetration attacks by hackers. This study focuses on the case of jailbreaking a large language model. This study analyzed individual penetration attacks and jailbreak sites such as Bard, open-AI, and MS-bing. The source site for this study was Zebra Break Chat (https://www.jailbreakchat.com/). The results showed that AI is inherently vulnerable to bypassing penetration methods that hijack the goal because it uses objective functions to optimize. For future research, it will be necessary to study AI security functions that can robustly protect the fundamental purposes that society should pursue, such as "inherent human dignity," "humanity," and "justice and fairness."