An Efficient Finetuning Method for LLM generated text detection in Power Grid

The increasing prevalence of Large Language Models (LLMs) in various sectors, including the power grid industry, has raised concerns regarding the detection of AI-generated content, particularly in expert technical documents. Existing detection methods, while effective in some cases, often struggle with the diverse and complex nature of power grid documents. This paper introduces a novel approach for detecting LLMgenerated text in electrical engineering documents by efficiently fine-tuning the ChatGLM-6B model using Low-Rank Adaptation (LoRA). Our method leverages the rich domain-specific knowledge embedded in the LLM, while minimizing computational overhead and memory usage. Experimental results demonstrate that our approach significantly outperforms current state-of-theart methods, achieving $91.47 \%$ accuracy on the Kaggle dataset, 97.23% on the LLM-Detector dataset, and 87.65% on the power system dataset. This work provides a robust and efficient solution for LLM-generated text detection in the power grid sector, offering high generalization ability and scalability.