Researcher Ciptakan DetectGPT Demi Perangi ChatGPT Di Dunia Pendidikan Yang Semakin Menkhawatirkan

JAKARTA - Recently, many are worried about the emergence of a chatbot based on Artificial Intelligence (AI), including Google and how it affects the world of education. Now, a solution has emerged with DetectGPT.

According to a research team at Stanford University, DetectGPT is one of the first tools to combat text generated by Large Language Models (LLM) on popular AI text applications such as ChatGPT in the world of education.

Such a method, based on the notion of text generated by LLM, usually resides around a given region using local curvature of the log model probability function.

That is, the model will recognize patterns that may indicate a piece of AI-generated text, based on what is presented thanks to the zero-shot method. Knowing these patterns, DetectGPT should be able to detect if there is a piece of text generated by AI.

DetectGPT operates in stark contrast to other methods that require classification of training and datasets from real and generated parts.

In addition, the research team tested DetectGPT on a data set of fake news articles and outperformed other zero-shot methods to detect machine-generated text.

In particular, they found the DetectGPT improved the detection of fake news articles generated by the 20B GPT-NeoX parameter from 0.81 AUROC for the strongest zero-shot baseline to 0.95 AUROC for DetectGPT.

The zero-shot method, which does not require additional data or training, makes it an efficient and effective tool to identify machine-generated text.

As the use of LLM continues to grow, the importance of appropriate systems to detect machine-generated text will become increasingly important, as quoted by Neowin, Monday, January 30.

DetectGPT is a promising approach that can have a significant impact in many fields, and further development can be beneficial for many fields.