OpenAI will not watermark ChatGPT text since its users may get detected

Open AI has a text system created by the cat waterproof and a tool to detect the waterproof for about a year, reports the Wall Street Journal. But the company is divided internally on the opportunity to release it. On the one hand, it seems to be the responsible thing to do; On the other, it could harm its results. Openai's watermark is described as adjusting how the model predicts the most likely words and sentences that will follow those previous, creating a detectable model. (This is a simplification, but you can consult Google's deeper explanation for the watermark of Gemini's text to find out more). Openai would have a Ki text detector

OpenAI Reportedly Has an AI Text Detector

According  Openaai has discussed the publication of such a tool in the past two years. The text of the recognition of the text has also been ready for about a year. The report said that they have quoted anonymous people who know the question. One of the sources informed the publication that the publication of the tool was as simple as to press a button.

Such a tool could help educators and other similar institutions in which people have used content generation such as tests and research with AI. At the beginning of this year, it was found that a scientific article published in Elsevier's Surface and Interface Journal was published by AI by AI and was then removed after having read online. This AI-powered plagiarism has become a major problem for academia because there is no reliable method to detect the use of AI.

Providing a way to recognize AI-written material is a potential boon for teachers trying to discourage students from submitting writing assignments to AI. The Journal reports that the company found that watermarks do not affect the quality of its chatbot’s text. In a survey commissioned by the company, “people around the world supported the idea of ​​an AI detection tool by a ratio of four to one,” the Journal writes.

After the journal published its paper, OpenAI confirmed that it was working on text watermarking in an updated blog post spotted by TechCrunch today. In it, the company says its method is highly accurate and resistant to "manipulation, such as paraphrasing." But it says techniques like rephrasing using a different model make it "trivial to circumvent by malicious actors." The company also says it's concerned about stigmatizing the usefulness of AI tools for non-native speakers.

But it seems OpenAI is also worried that using watermarks could turn off surveyed ChatGPT users, nearly 30 percent of whom apparently told the company they would use the software less if watermarks were implemented.

Despite this, some employees reportedly still believe watermarks are effective. However, given the nagging user sentiment, the Journal says some have suggested trying methods that "may be less controversial among users but unproven." In its blog post update today, the company said it is in the "early stages" of exploring metadata embedding. It is "too early" to know how well it will work, but because it is cryptographically signed, there can be no false positives.