by 
OpenAI is releasing tools to detect ChatGPT-written content
OpenAI has created an application that may detect cheating students by using ChatGPT to complete their assignments however. Adequate reports from The Wall Street Journal the company is mulling whether or not to release it.
the OpenAI spokesperson said that they are examining the watermarking of text method that was as described in the Journal’s report. But added that it’s using the “deliberate approach”. Because of “the complexity of the process and the likely impact it will have on the larger ecosystem that isn’t OpenAI.”
“The method for watermarking texts we’re experimenting with is technically sound but it’s also a risky proposition. We’re looking at as we investigate alternatives. Like the possibility of being hacked by malicious actors. And the potential to significantly affect groups such as non-English native English speakers,” the spokesperson said.
This is a different method than previous efforts to recognize AI-generated texts, which were largely unsuccessful. In fact, OpenAI itself has shut down its prior AI detection of text in the year 2000 because of the “low degree of accuracy.”
Text Watermarking
With the use of text watermarking OpenAI will concentrate on detecting text from ChatGPT and not from other models from other companies. It will do this by making minor changes to the way ChatGPT chooses words. Effectively creating a watermark that is invisible in the text that can later be spotted by a different software.
After the publication of the Journal’s story OpenAI also revised the blog post in May on its work in the detection of AI-generated material. The updated post states that text watermarking has proved “highly precise and effective against localized tampering for example, paraphrasing” but it’s “less resilient to the global tampering process. Which includes with translators, rewriting with a different model as well as asking the machine to put a specific character between each word, and then delete the special character.”
In the end, OpenAI states that this technique is “trivial to be circumvented by bad actors.” The update by OpenAI also confirms the point of view of the spokesperson regarding non-English speakers. Saying that watermarking text could “stigmatize the use by AI as a helpful software for writing tools that are not available to native English language users.”
Thanks for Reading this article for more information please visit. www.techychunk.com
Q&A
What is OpenAI doing with detection tools?
OpenAI is releasing tools to detect content generated by ChatGPT.
How is OpenAI approaching this release?
OpenAI is taking a cautious and gradual approach.
