Stanford University launches DetectGPT to counter ChatGPT!

OpenAI's ChatGPT has astounded the globe with its tremendous information integration and conversation skills, as well as its astounding natural language processing capabilities. It was recently announced that ChatGPT passed the exams for legal and business courses at major universities in the United States, which caused students' heads to explode!


In response, a Stanford University research group has recently launched DetectGPT, which is based on the notion that text generated by Large Language Models (LLM) typically resides in a certain region of the model's log probability function in the negative curvature zone.


With this discovery, the researchers devised a novel metric for determining if a sentence was written by a computer that does not need training artificial intelligence or gathering massive text comparison datasets.

The new method of DetectGPT, known as "zero-shot learning," was developed at Stanford University and enables DetectGPT to recognise machine-written text without needing to know the source of the text. artificial intelligence tool. In contrast to other approaches, which involve training "classifiers" and real and produced paragraph data sets, its functioning does not require training.

The team evaluated the accuracy of DetectGPT using a dataset of phoney news items; DetectGPT performed better than previous AI that detects AI-generated text.

Source: IndiaTimes

Is this the end of ChatGPT's era? You may leave your thoughts at the comment section below!