Using the Classifier
Users can paste the text into a box, click a button, and receive a classification of "very unlikely," "unlikely," "unclear," "possibly," or "likely AI-generated." However, there are some limitations to keep in mind, such as a minimum of 1,000 characters and the potential for mislabeling both AI-generated and human-written text.
Caveats and Limitations
The classifier needs to be more reliable and should not be used as a primary decision-making tool. It was primarily trained on English content written by adults, so it may need help with text written by children or not in English. The classifier may also "incorrectly but confidently" label human-written text as AI-generated. OpenAI states that the tool is still a "work in progress."
Testing the Classifier
In tests, the tool labeled AI-written text as "likely AI-written" 26 percent of the time and gave false AI detections 9 percent of the time. It correctly marked most ChatGPT responses as "possibly" or "likely AI-generated."
OpenAI's Focus on Education
OpenAI is focusing on using this detection technology in education. It is "engaging with educators in the US" to understand the impact of ChatGPT in the classroom and is soliciting feedback from those involved in education.