Securing the Supply Chain of Large Language Models as Software with Explainable AI and Humans in the Loop
Dr. Ziyu Yao
KEY INTERESTS
Natural language processing; Artificial intelligence; Human-AI interaction; Language and code semantics; Efficient machine learning
AFFILIATIONS/APPOINTMENTS
Assistant Professor, Department of Computer Science, George Mason University
ACADEMIC DEGREES
BE, Communication Engineering, Beijing University of Posts and Telecommunications
PhD, Computer Science and Engineering, Ohio State University
SECURING THE SUPPLY CHAIN OF LARGE LANGUAGE MODELS AS SOFTWARE WITH EXPLAINABLE AI AND HUMANS IN THE LOOP
Large language models, or LLMs, have appeared as the “game changers” and now beenincreasingly used in a variety of applications and domains, including the supply chain industry. In this age of “LLMs as software”, it becomes particularly important for industrial practitioners to gain sufficient understanding on the potential risks of LLM-based applications and the solutions to defend them. Notably, this is a topic that is still largely understudied. This research fills this gap by first proposing a novel investigation on the software supply chain vulnerabilities of LLMs when they are used with “prompt engineering”, an emerging paradigm only enabled by the advanced LLMs. The project seeks to then devise a novel approach to defend LLM-based software, leveraging Explainable AI and Human-AI interaction.