Thread by @Linkletter

In one week, on April 4th, Turnitin will release a product that does not work: AI text detection. All current customers are affected, and no one can opt out.

Turnitin confidently purports that this tool will only have a 1% false positive rate. This claim is 100% false.

Turnitin admits that their AI text detection tool was only built with GPT 3 and GPT 3.5 in mind. But ChatGPT and Bing use GPT 4 now! It's outdated and hasn't even been released yet.

What is the false positive rate on GPT 4, or other LLMs? They don't even know. No one does.

This technology is positioned to cause lots of harm to students, and it will. Instructors have been told it is highly accurate: it's not. Institutions have been told they cannot test it before release: that's dangerous.

Turnitin's AI text detection tool is simply unsafe.

Turnitin admits that their AI text detection tool is defeated by paraphrased text. Does anyone think students won't figure this out? Turnitin's AI text detection is only free until January 2024, and their paraphrasing detection tool will cost - and isn't even out yet!

Turnitin *must* halt the rollout of its AI text detection tool until institutions can test it themselves and turn it off if they so choose. Institutions have paid for a tool that shows its work, oh wait I mean other people's work. Turnitin is now a black box, and that's not OK.

Stop the rollout. Let us test. Let us turn it off. Prove that it works on GPT 4 and other LLMs. Prove that the false positive rate is low on any AI system.

Slow down! Stop! This isn't safe! #Turnitoff