Internet users who want quick responses to a variety of questions and tasks have grown increasingly fond of the AI chatbot created by OpenAI, called ChatGPT.
Educators are extremely concerned about students turning in papers that were produced by ChatGPT because of this tendency, though. As a result, Stanford University researchers developed a tool called DetectGPT to assist teachers in identifying content produced by ChatGPT and other comparable large language models (LLMs).
OpenAI has stated that it is developing a detecting tool after New York forbade public school pupils from using ChatGPT for academic purposes.
DetectGPT is a tool designed to recognise texts published by ChatGPT or other chatbots of a similar nature. These AI-generated sentences “occupy negative curvature regions of the model’s log probability function,” according to the academics who created them.
By examining the log probability, this information aids DetectGPT in determining whether a text was created by a particular language model. According to the researchers, DetectGPT is accurate 95% of the time.
Machine learning and artificial intelligence both employ the mathematical idea of log probability functions. It is used to calculate the likelihood or degree of confidence that an event will occur. Particularly in applications like natural language processing and language synthesis, the logarithmic transformation of probability facilitates comparison and result interpretation. This enables a more precise assessment of the text produced by the AI models.
Educational institutions have developed a tool called DetectGPT to assist in catching students who use programmes like ChatGPT to write their papers and other assignments. Regrettably, the creators of DetectGPT have not yet provided much information about their product, and nobody can currently use it.
Yet, the creation of such tools is a positive step towards ensuring that students submit their work in an honest and reliable manner.