Trustworthy Large Language Model Prompt Engineering for Risk-Free Smart Grid Management Education

The management and control of smart grids are rapidly evolving to include Artificial Intelligence (AI). Large Language Models (LLMs) are the latest solution for AI-capable Industry 4.0 management. The trustworthiness of the LLMs is an important issue in the management of such systems. In this paper, the behaviors of various locally-hosted LLMs in managing unauthorized operations within smart grids are investigated. To highlight the advantages of local-hosting, we also compare the performance of these models with online LLMs. Our findings show that locally-hosted LLMs perform better in preventing vulnerabilities than their online counterparts. This study proposes an educational platform for future generations of smart grid operators and provides guidelines for teaching the vulnerabilities of LLM prompting needed to be included in the educational curriculum of the smart grid management workforce. Furthermore, the findings of this research can provide guidelines for standardization, legislation and regulations of AI applications in smart grids.