A report from the Wall Street Journal says that OpenAI, the company behind ChatGPT, has a tool that can detect AI-generated tasks with great accuracy. Now, the company has shared a little more about its research in an update shared via a blog post, as well as why it hasn't made the detection method available yet.
See also: SearchGPT: OpenAI tests its own AI search engine
OpenAI says its teams have developed a text watermarking method that they are testing, but are also looking at alternative solutions for ways to prevent AI-generated text written by ChatGPT from being used in situations where it shouldn't be, such as a student who suddenly "thought" of an epic assignment.
According to OpenAI, watermarking is one solution. It can involve classifiers and metadata. It is accurate, but it can be less useful if you use a translation system or if you rephrase the text using another production model. It can also be circumvented by asking the AI to insert a specific character after each word and then delete it later.
See also: Former OpenAI researcher: Security is not a priority for AI companies
Text watermarking could also impact other users, such as non-native English speakers who would like to improve their text and use artificial intelligence to help them write.
According to the blog post, OpenAI has looked at all aspects of the text-origin issue.
See also: Why is Microsoft competing with OpenAI?
Addressing this issue is a complex one. Of course, we don’t want kids to be unable to write an essay (or any text longer than 200 characters) on their own. But there are also cases where using Generative AI to write things is perfectly acceptable and shouldn’t be such a big deal. Regulating AI is one of the challenges we face as technology evolves , and we should be looking at this more broadly moving forward.
Source: phonearena
