After a 22-year-old from Toronto made a third-party ChatGPT detector tool to determine whether a piece of text was written by a human or an AI, OpenAI is itself now introducing a new “AI Classifier” for “indicating AI-written text.”
“We’re launching a classifier trained to distinguish between AI-written and human-written text,” reads the tool’s page on OpenAI’s website. The company says that while it is impossible to “reliably” detect AI-written text, it believes that good clarifiers can inform mitigations for false claims regarding a piece of text that was written by AI but claimed to be human-written.
The company makes it very clear that its classifier is not fully reliable. In its tests, the classifier correctly identified 26 percent of AI-written text, while it incorrectly labeled human-written text as AI-written nine percent of the time. “Our classifier’s reliability typically improves as the length of the input text increases,” said OpenAI. Texts with less than 1,000 words are prone to be labelled incorrectly.
OpenAI also says that the tool should not be used as a “primary decision making tool,” and instead, should complement other ways of determining whether the piece of text is human-written or not. Further, OpenAI recommends using the classifier only for English text. “It performs significantly worse in other languages and it is unreliable on code.”
The company is making the classifier publicly available to collect feedback from users, while it continues to improve the tool. Try out the classifier here.
Image credit: Shutterstock