Princeton Student Builds App to Detect ChatGPT-Generated Essays
A Princeton senior creates GPTZero, an app that helps educators identify AI-generated writing by analyzing text perplexity and burstiness. The tool gains rapid traction as schools worry about students using ChatGPT to cheat.
A Princeton University senior named Edward Tian creates an app called GPTZero to help teachers determine if an essay is written by a human or generated by artificial intelligence. As ChatGPT grows rapidly in popularity and valuation, school districts like New York City block the AI tool on their networks to prevent students from using it to cheat on assignments.
GPTZero works by analyzing two specific variables in a piece of writing: perplexity and burstiness. Perplexity measures the randomness of the text, as human writing typically features more unpredictable word choices compared to the highly familiar and predictable patterns produced by AI language models.
The app also evaluates burstiness, which looks at the variance and inconsistency of sentence structures throughout the text. Over 20,000 users already test the beta version of the software, and educators from around the world reach out to Tian to thank him for creating a tool that confirms their suspicions about AI-generated homework.