AI Detection Accuracy for Chinese LLMs: AI or Not vs ZeroGPT

Chinese LLMs such as Deepseek, Kimi, Qwen, and Sonoma Sky Alpha were used to test the accuracy of AI detection tools. The comparison between AI or Not and ZeroGPT analyzed how well they detect AI-generated text. Through 95 different writing examples, AI or Not demonstrated a 95% accuracy rate, while ZeroGPT showed a 56% accuracy rate.

The advancement of AI and LLMs necessitates understanding the implications of using machine-generated text in academic, professional, and research settings. As Chinese AI models rapidly advance, the ability of AI detection tools to accurately identify outputs from Chinese-based models becomes crucial. This has real implications for students, teachers, researchers, and corporations.

AI detectors compare AI-generated and human-written text to find patterns. They are trained on a variety of content types and models, providing a probability percentage of AI generation. AI or Not and ZeroGPT are trained on diverse texts from numerous models.

95 AI-generated prompts from Chinese LLMs were used to test accuracy and potential false positive rates. This study involved models like Deepseek, Kimi, Qwen, Sonoma Dusk Alpha, Sonoma Sky Alpha, and Moonshot. China provides open-source AI technology, allowing developers easy access.

AI detection tools failing to spot China-based LLM outputs create issues for students, researchers, teachers, and professionals. Incorrect text identification leads to unfair penalties for students and misidentification of AI-generated content as human work. AI or Not aims to address these issues by training against diverse content sources.