New developments with AI writing detectors (they still don’t work!)

Some interesting recent developments in AI writing detection:

• Last week, OpenAI quietly ended access to its AI writing classifier. According to them, “the AI classifier is no longer available due to its low rate of accuracy.”

• In May, Turnitin admitted their AI writing detector isn’t as accurate as they previously claimed. They revealed that “Since our release, we discovered real-world use is yielding different results from our lab.” Shocking, right?

• In April, the AI writing detector GTPZero concluded that the U.S. Constitution is “likely to be written entirely by AI."

As I and others have previously written, low accuracy is a problem with AI writing detectors across the board. They don’t work reliably, and they’re doing real harm to students, who are being falsely accused of cheating.

I commend this move by OpenAI, while also recognizing that it was never a source of revenue for them. They didn’t have much to lose by shutting it down. Many other AI writing detectors, including the for-profit ones mentioned above, are still out there.

Spread the word - no one should rely on them.