Edward Tian is worried about the rise of AI-generated did ai write this. The Princeton University computer science major has been teaching himself coding for the last year, and he created an app called GPTZero over the course of a few weeks in December to help teachers weed out students who might be trying to pass off artificially generated assignments as their own.
The free tool scans text for clues it could be written by an AI, including its “perplexity” and whether the words sound too “burstiness,” which means they’re too predictable and stray from existing patterns in human writing. GPTZero quickly became popular, and it grew so large that Streamlit, which hosts the site, had to give it more memory and resources to handle the demand.
Unveiling the Writer’s Best Companion: Exploring AI Content Detectors for Quality Assurance and Plagiarism Detection
Several other AI detection tools are available online, too. Glenn Gabe, who has studied AI for 27 years, says Originality AI has a reputation for being more accurate than others in detecting GPT and ChatGPT texts, which are often recognizable by their patterns. The tool can analyze an entire paragraph in seconds and offers five responses, ranging from very unlikely to likely that it’s AI-generated. It also has a Chrome extension that lets you test content right within your browser.
Still, no one tool is foolproof. And it’s important to always have a human read over any piece of work before posting, no matter how pristine it looks. Editors catch mistakes that machines might miss, and they can double-check facts to make sure they’re valid. That’s especially true when it comes to sourcing data and statistics, which can be easily manipulated by an AI.