OpenAI has apparently built a tool to watermark and detect writing created by its chatbot, ChatGPT.
As reported by The Wall Street Journal, the tool is quite effective with a high accuracy rate of 99.9 percent and can pretty much catch AI-generated articles that are circulating on the internet today. It would also be a boon for teachers, as they’d have a better means of identifying students who cheat and plagiarize without the risk of getting false positives and accusing honest students.
However, the company is still considering whether to make it available to the public. OpenAI reportedly has had this tool already prepared for release for nearly a year now, and that all it takes to launch the tool is essentially just a press of a button, according to a source cited by the WSJ. Internal debates are ongoing regarding the implications of the release.
- ChatGPT’s coding answers are incorrect more than half the time — study
- AI detection tools show bias against non-native English speakers
Among the concerns is the potential for discrimination against non-native English speakers who may use AI as a writing aid. OpenAI also conducted a survey among ChatGPT users, with 30 percent of the respondents saying they would reduce their use of ChatGPT if such detection technology is implemented. There are also worries that the quality of ChatGPT’s output might be adversely affected.
OpenAI previously claimed that AI writing detectors do not work. In July 2023, the company discontinued AI Classifier, its previous tool for distinguishing human text from AI-generated text, as it had a terrible accuracy rate of only 26 percent.