Weeks after ChatGPT’s November 2022 release set off alarm about AI-written homework, a Princeton University senior named Edward Tian published GPTZero on January 1, 2023. He had coded the detector over winter break; his very first test fed it writing by the New Yorker author John McPhee, who happened to be his writing professor. A demonstration tweet went viral on January 9, drawing millions of views and tens of thousands of early users.
GPTZero estimated whether a passage was machine-written using two statistical signals: perplexity, a measure of how surprising or unpredictable the word choices are, and burstiness, which looks at how that surprise varies from sentence to sentence. Human writing tends to be more variable; early language models produced flatter, more uniform text. Tian framed the mission as “preserving what’s human in an AI world,” left his PhD plans, raised about 3.5 million dollars, and shipped a Chrome extension called Origin alongside the company.
GPTZero became the public face of the AI-detection industry that sprang up in 2023 - and, like its competitors, it ran into the same hard limit that led institutions such as Vanderbilt to abandon detection: the tools produce false positives, and the underlying problem grows harder as models improve.
Why business readers should care: GPTZero shows how fast a single motivated builder can ship a product into a panic-driven market. It also shows the trap of building a business on a detection problem that may be unsolvable in principle - early traction is easy, durable reliability is not.