OpenAI pulls back AI-generated text detector, cites “low rate of accuracy”

OpenAI has quietly removed its AI classifier tool which detected whether or not text was generated by AI.

OpenAI has quietly removed its AI classifier tool which detected whether or not submitted samples of English-language text were generated by artificial intelligence, due to the tool’s “low rate of accuracy.”

While users could previously try out OpenAI’s classifier for free and submit their own text for analysis, clicking on the link now takes the user to a page removal notice.

A statement on the web page for OpenAI’s classifier said, “As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy. We are working to incorporate feedback and are currently researching more effective provenance techniques for text, and have made a commitment to develop and deploy mechanisms that enable users to understand if audio or visual content is AI-generated.”

The Hindu previously tested OpenAI’s classifier tool when it was still available and found that it generated false negatives, such as incorrectly labelling ChatGPT-generated text as “unlikely AI-generated” or even “unclear.”

(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)

OpenAI also admitted on its website that the classifier delivered false positives, or mistakenly flagged human-generated text as being AI-generated 9% of the time during evaluations.

The boom in generative AI technology and chatbots in 2023 led to fears that educators and publishers would be flooded with dishonest work generated by free services such as ChatGPT.