OpenAI confirms it’s looking into text watermarking for ChatGPT that could expose cheating students
Following a bellow from The Wall Road Journal that claims OpenAI has been sitting on a scheme that may maybe well presumably do essays written by ChatGPT with a excessive diploma of accuracy, the corporate has shared slightly files about its compare into textual notify material watermarking — and why it hasn’t launched its detection methodology. In step with The Wall Road Journal’s bellow, debate over whether the scheme desires to be launched has saved it from seeing the sunshine of day, no topic it being “ready.” In an update printed on Sunday to a Would possibly well also weblog post, seen by TechCrunch, OpenAI acknowledged, “Our groups include developed a textual notify material watermarking methodology that we continue to clutch into memoir as we compare choices.”
The company acknowledged watermarking is one of multiple solutions, including classifiers and metadata, that it has looked into as half of “intensive compare on the location of textual notify material provenance.” In step with OpenAI, it “has been highly honest appropriate” in some cases, however doesn’t manufacture as nicely when faced with certain forms of tampering, “love using translation systems, rewording with but any other generative model, or asking the model to insert a obvious personality in between every be conscious and then deleting that personality.” And textual notify material watermarking may maybe well presumably “disproportionately impact some groups,” OpenAI wrote. “As an instance, it can well presumably stigmatize employ of AI as a functional writing scheme for non-native English audio system.”
Per the weblog post, OpenAI has been weighing these risks. The company also wrote that it has prioritized the release of authentication instruments for audiovisual notify material. In a assertion to TechCrunch, an OpenAI spokesperson acknowledged the corporate is taking a “deliberate methodology” to textual notify material provenance thanks to “the complexities gripping and its seemingly impact on the broader ecosystem beyond OpenAI.”
