GPTZero Fights Online AI Disinformation, School Plagiarism

For those worried about AI creep — the insidious influence of artificial intelligence over everything from school classwork to career aspirations — Princeton University undergrad Edward Tian has an app for that. Tian has received $3.5 million in funding for an invention called GPTZero, which analyzes text to identify the work of generative AI. The 10-person team claims the tool has a 99 percent accuracy rate for human text and can score at about 85 percent when it comes to text written by AI. The company is now talking to media leaders about partnerships for AI detection and analysis.

The 22-year-old computer science and journalism major invented GPTZero to deter students from trying to pass off AI work produced as their own. Specifically, he targeted  classroom misuse of ChatGPT, which has amassed 1.2 million registered users since launching in January.

Now, with the help of his new funding partners, he’s rolling out a new browser plug-in called Origin, designed to save journalism by flagging AI-generated disinformation online.

The seed funding round co-led by Neo Capital and Uncork Capital, with investment from tech luminaries including Stability AI CEO Emad Mostaque and Lattice co-founder Jack Altman (brother of OpenAI CEO Sam Altman) has paved the way for conversations with executives including former CEO of The New York Times Mark Thompson and the BBC.

“GPTZero analyzes the randomness of text, known as perplexity, and the uniformity of this randomness within the text — called burstiness — to identify when AI is being used,” Bloomberg reports, noting that “the company also sees its technology for use in fields of trust-and-safety, government, copyright, finance, law and more.”

Tian says that because the field of detection is so new it is not getting the industry-wide attention it deserves. “We believe we can get the smartest people working on AI detection in a room together,” Tian told Bloomberg.

ChatGPT-maker OpenAI has created an AI text classifier designed to identify machine-generated content, but Bloomberg suggests “it’s far from foolproof,” with a success rate of “only 26 percent” for AI-written text, “while incorrectly labeling human-written text as AI-written 9 percent of the time.” The classifier scores “significantly worse” in non-English languages and is “unreliable” with code and shorter texts.

When GPTZero launched in January, “over 30,000 people tried out the tool within a week of its launch, crashing the servers,” according to Fast Company, which says 1.2 million people have since registered for the app. Origin, which launched this month on Google’s Chrome Web store, “allows users to highlight text on the Internet, and a browser extension automatically checks if the content is AI-generated.”

“In the future, it will also verify the source of information,” explains Fast Company.

With Seed Funding Secured, AI Detection Tool GPTZero Launches New Browser Plugin, Forbes, 5/9/23

No Comments Yet

You can be the first to comment!

Leave a comment

You must be logged in to post a comment.