Teachers are lining up for a ChatGPT detector created by this 22-year-old in Toronto
Edward Tian, a 22-year-old from Toronto, created a ChatGPT detector.
Published Wednesday, January 25, 2023 1:57PM EST
Last Updated Wednesday, January 25, 2023 1:57PM EST
As awe-struck Internet users obsessed over the wondrous abilities of ChatGPT, a 22-year-old in Toronto was feverishly crafting a tool to detect its misuse.
“ChatGPT is an incredibly cool innovation,” Edward Tian told CTV News Toronto.
“But it’s like opening a Pandora's Box.”
He would know. The Etobicoke native is a computer science major at Princeton University and spent the last couple years studying GPT-3, artificial intelligence that produces human-like text, just like ChatGPT.
The interactive chatbot is powered by machine learning. ChatGPT essentially swallowed massive swaths of the Internet, learning language patterns in the process that it can recreate in response to a human prompt.
As ChatGPT landed in the hands of the public in late November, Tian played around with the technology alongside friends. They asked the program to write poems and raps.“Wow this is really good,” Tian remembers thinking. “This is better than something I could write myself.”
That high-level of skill was raising alarm bells for educators, who began fearing that their students would hand in essays generated by a machine and they would have no way of knowing or confirming suspicions. Immediately, Tian became aware of this too.
“Everyone deserves to know the truth and everyone deserves a tool at their fingertips that can determine whether something is human or machine generated,” he said.
Luckily, he had time on his hands during winter break and sat down at a coffee shop in Etobicoke to do something about it. The result: GPTZero, an app that can decipher whether something was written by a machine or human.
First, a user copy and pastes text into the app. An evaluation begins, measuring the perplexity, creativity and variability of the writing. Then, GPTZero delivers a score, which leads to a result: either the text was generated by ChatGPT or a human.
On Jan. 3, the app went public. More than 300,000 people tried it and over 7 million people viewed it on Twitter.
“It was totally crazy. I was expecting a few dozen people,” Tian said
In particular, teachers were noticing GPTZero worked in detecting if their students were writing their papers, or not. Now, Tian is building a tool specifically designed for educators. Already, 33,000 teachers have signed up on the product waitlist.
“No one wants to be deceived if something they are reading is misrepresented as human,” Tian said.