While ChatGPT is still far from being a reliable source of information, it’s already gained an actual use case among students – cheating on essays. The AI chatbot is capable of writing essays that would otherwise take any human hours to complete, in mere seconds. Edward Tian, a computer science student at Princeton, clearly did not like this and dedicated his New Year’s holidays to building an app that can help detect whether a text was written by a human or using ChatGPT.
Called GPTZero, the app can “quickly and efficiently detect whether an essay is ChatGPT or human written,” Tian said in a tweet. Public schools in New York City recently banned ChatGPT from school devices and WiFi networks fearing cases of “AI plagiarism.” This may be a step in the right direction to prevent cheating, but it isn’t very effective as students can still work around it by completing their essays from home.
Meanwhile, GPTZero arms teachers with the ability to tell AI-generated essays apart from human-generated ones, which is a smarter way to go about it.
Tian told CBS MoneyWatch that the app works by looking at a text’s “perplexity” by measuring its randomness. If it’s human-generated, it scores higher on that aspect since humans are more unpredictable. Meanwhile, bot-generated texts typically tend to have low “perplexity” which leads to the bot flagging them as machine-generated.
In his tweets, Tian shared a couple of videos comparing GPTZero’s analysis of a New Yorker article with a ChatGPT-generated LinkedIn post. The tool gave the article 114 points and correctly detected it as being written by a human, while giving 40 points to the LinkedIn post, and once again, correctly identifying it as “AI generated.”
ChatZero is currently in its beta stages right now, according to its creator, who says they’ll be spending the next few weeks improving the model and analysis. You can give it a go from its official website — gptzero.me.