ChatGPT creator reveals new tool to spot AI-written homework, admits it's not perfect

OpenAI has released a tool to help educators spot texts written by artificial intelligence.

ChatGPT, a large language model by OpenAI, was released to the public around two months ago. Since the beginning, it has raised concerns, especially among teachers, about cheating since the AI-driven program can write an essay in a blink of an eye.

To help educators adapt to the new reality where ChatGPT has become students' best friend, OpenAI released a new feature called AI text classifier that allows users to check if a student or AI wrote an essay, article, or any other text.

"Really excited about this tool as a starting point for dialogue. As a part of this release, we've been engaging with educators about the responsible use of ChatGPT and AI text classifiers for teaching and learning," Lama Ahmad, policy research director at OpenAI, tweeted.

OpenAI admitted that the new tool isn't fool-proof.

"While it is impossible to reliably detect all AI-written text, we believe good classifiers can inform mitigations for false claims that AI-generated text was written by a human: for example, running automated misinformation campaigns, using AI tools for academic dishonesty, and positioning an AI chatbot as a human," it said.

It noted that the reliability typically improves as the length of the input text increases.

"In our evaluations on a "challenge set" of English texts, our classifier correctly identifies 26% of AI-written text (true positives) as "likely AI-written," while incorrectly labeling the human-written text as AI-written 9% of the time (false positives)," OpenAI said.

Recently, another anti-cheating app created by a 22-year-old Princeton student went viral. Edward Tian created GPTZero, which checks if a bot or a human wrote a given text.

Using ChatGPT for school cheating is not the only concern raised by the new technology. A recent investigation by the Cybernews research team revealed that an AI-based chatbot could provide hackers with step-by-step instructions on how to hack websites. After only 45 minutes of chatting, researchers got all the necessary information for exploiting a website's vulnerabilities.