OpenAI has released a tool to help you determine whether text was more likely written by a human or AI. However, the ChatGPT maker warns that its equivalent of Blade Runner’s Voight-Kampff test can also get it wrong.
The tool includes a box where you can paste text that’s at least 1,000 characters long. It will then spit out a verdict, like “The classifier considers the text to be very unlikely AI-generated” or “The classifier considers the text to be possibly AI-generated.”
I tested it by prompting ChatGPT to write an essay about the migratory patterns of birds, which the detection tool then described as “possibly AI-generated.” Meanwhile, it rated several human-written articles as “very unlikely AI-generated.” So although the tool could raise false flags in either direction, my (tiny sample size) test suggests at least a degree of accuracy. Still, OpenAI cautions not to use the tool alone to determine content’s authenticity; it also works best with text of 1,000 words or longer.