A new resource is available to teachers who are concerned that their students will submit essays written by a popular AI chatbot.
Tian, who is majoring in computer science and minoring in journalism, spent some of his winter breaks making GPTZero, which he says can “quickly and efficiently” tell if an essay was written by a person or ChatGPT.
He made the program to stop what he sees as a growing problem: AI programs that copy other programs. Since the ChatGPT language model came out at the end of November, there have been reports that students have used it to turn in AI-written assignments as their own.
“There is a great deal of ChatGPT fervor currently.” Could it be that AI has written this and that? “As sentient beings, we must have access to this information.” In a tweet, Tian explained GPTZero.
After Tian posted his bot online on January 2, he received a lot of positive feedback from teachers who tried it out.
Just one week after its release, GPTZero had already been tried by more than 30,000 people. The app failed due to overwhelming demand. Tian has been given additional memory and resources by Streamlit, the free platform that hosts GPTZero, to deal with the increased web traffic.
Functioning of GPTZero
GPTZero uses “perplexity” and “burstiness” as indicators to determine whether an excerpt is written by a bot. A text’s perplexity indicates its complexity; if GPTZero has trouble understanding it, it’s highly complex and likely written by a human. In contrast, the text is more likely to be generated by AI if it is more familiar to the bot (due to its training on such data), in which case it will be of low complexity.
The sentence variations are compared independently using the burstiness metric. People tend to write more quickly, which is clear from the fact that the same text can have both short and long, simple and complicated sentences. Most sentences produced by AI are fairly standard.
Tian showed off the app’s ability to analyze a New Yorker article and a LinkedIn post by ChatGPT in a video demonstration. It was able to tell the difference between human and computer-written text.
Because of the issues that have been reported by some users, Tian has admitted that his bot isn’t perfect. He promised that he would keep refining the model until it was perfect.
However, the app’s design contributes to Tian’s overarching goal of bringing transparency to AI by illuminating the differences between humans and computers.
“AI has been a black box for so long that we really don’t know what’s going on inside,” he said. And I hoped to make a stand against it all with GPTZero.