Princeton Senior Develops GPTZero to Detect AI-Generated Plagiarism
Princeton senior Edward Tian creates GPTZero, a new software tool that identifies writing produced by AI chatbots like ChatGPT. The tool measures text perplexity and burstiness to distinguish human writing from machine-generated content.
Princeton senior Edward Tian develops GPTZero, a new software tool designed to detect plagiarism generated by the AI chatbot ChatGPT. Inspired by the rapid rise of the OpenAI chatbot, Tian spends his winter break coding the application in a Toronto coffee shop to help educators combat AI-assisted academic dishonesty.
The software analyzes text using two specific metrics known as perplexity and burstiness. Because ChatGPT predicts the most logical next word, its writing exhibits low perplexity and consistent burstiness, making GPTZero highly effective at distinguishing machine-generated prose from the more random, varied patterns of human writing.
GPTZero gains massive traction immediately after its beta launch, amassing over seven million views on Twitter and reaching users across 40 states and 30 countries. Major publications like the New York Times and the Washington Post cover the tool, highlighting the growing need for AI detection as chatbots become increasingly advanced.