Edward Tian was sleeping when his bot crashed the website.
The 22-year-old Princeton senior spent the winter break at his local coffee shop creating GPTZero, an app he claimed would be able to “quickly and effectively” tell whether an essay was written by a human or ChatGPT OpenAI. When he uploaded it to the Streamlit app development and hosting platform, he didn’t expect it to happen that a lot of attention.
“I expected at most a few dozen people to try out the app,” Tian told The Daily Beast. “Suddenly, the number of users was insane as over 2,000 people signed up for the beta in a matter of hours.”
GPTZero eventually saw such a massive influx of users that it even suspended the platform that hosted it. “I’m amazed that it blew up and went viral,” he added.
When OpenAI released ChatGPT on November 30, 2022, it unleashed a digital Pandora’s Box on the world.
From high school teachers to college professors to journalists, everyone feared that the powerful AI chatbot had ushered in a new era of bot-generated essays and articles some had dubbed “AIgiarism.” Some teachers have already started reporting cases of their students using ChatGPT to create essays from the whole and finish written assignments.
While OpenAI said it eventually plans to implement “watermarking” to check if something was created by ChatGPT, there is still no official method to do this – which could cause a giant bot-sized headache across all sectors such as education and journalism.
Tian, who studies computer science and journalism, was concerned about the ethical dilemmas posed by chatbots, as well as what he described as a “black box” of large language models like ChatGPT. The opaque nature of the models causes people to fundamentally misunderstand them and thus misuse them.
So, despite being on the verge of graduating, he decided to spend the winter break building a tool that could help people find out if a text was probably written by a bot.
“People deserve to know when handwriting is not human” – Tian sid. “There’s been so much hype surrounding ChatGPT generation and AI lately that people deserve to know the truth.”
GPTZero uses two different metrics to judge whether a text was written by a bot: embarrassment and tearing. Texts placed in the application will have a number assigned to both metrics. If the number is low, it is more likely to be created by a bot.
OpenAI’s impressive new chatbot isn’t immune to racism
Embarrassment is a measure of randomness in a sentence. If the sentence is structured or uses words in a way that surprises the app, it will score higher in the confusing category. Tian said he used the free and open source GPT-2 to help train his app for this metric.
Crack is quality generally randomness for all sentences in the text. For example, human handwriting tends to have sentences of varying complexity. Some are simple. Some might give James Joyce a run for his money. On the other hand, bots tend to generate sentences of relatively low complexity throughout.
“There are beautiful features of human written prose that computers can and should never co-opt,” explained Tian. As a student of journalism, he was inspired by classes with the American writer John McPhee, who taught him these beautiful features of human handwriting.
Tian used McPhee’s essay New Yorker as part of his demo for GPTZero:
Despite building the tool, Tian is not against AI. He believes there is a time and place for them if they are used ethically and with permission. Hell, he even used AI programs like CoPilot to “handle most of my coding.”
“I’m not against using AI to write if it makes sense,” he said. “But people deserve to know when handwriting is not human. There’s been so much hype around ChatGPT and AI generation lately that people deserve to know the truth.”
Faced with the hype and concerns surrounding ChatGPT, a tool like Tian can prove extremely useful in a variety of sectors, from teachers wanting to check if their student has plagiarized an essay, to job recruiters wanting to check if a cover letter has actually been written by the candidate. As such, it can also be extremely lucrative for the right investors – some of whom have already contacted Tian with dollar signs in their eyes.
“Over the last day, several VCs popped up in my Twitter DMs,” Tian said, including A16Z, Menlo Ventures, and Red Swan. But he’s not done with GPTZero yet. He wants to further refine and develop the app, and even plans to make it more transparent with “explainers and detection methodologies.”
And at the end of the day, he’s a college senior. He has final exams coming up and homework and essays made by people have to worry about. At the moment, this is a much bigger problem than digital Pandora’s box or VC investors
“I’ll be answering all calls, but for now,” he said with a laugh. “But I’m just a college student focused on finishing school.
Read more in The Daily Beast.
Got a tip? Send it here to The Daily Beast
Get the Daily Beast’s biggest news and scandals straight to your inbox. Sign up now.
Stay informed and get unlimited access to the Daily Beast’s unparalleled reports. Sign up now.