Introduction to DetectGPT - a model to detect machine-generated text

Abstract

With the explosion of Large language models (LLMs), recent advances in natural language generation using artificial intelligence are creating a wave of opinions on social networking and artificial intelligence community. Especially, with the recent launchof ChatGPT, young people, specifically students, have begun to update and apply this model in the field of learning. With ChatGPT or similar large language models, students can complete essays with very little effort and attention. This is also a problem in the teaching and learning process of universities. Since then, a group of researchers at Stanford University have proposed a model, called DetectGPT, for the purpose of detecting texts generated by ChatGPT. The article will analyze the DetectGPT system model from an algorithmic perspective and compare the results of DetectGPT with existing models. Then, highlight the outstanding features of the DetectGPT model.