OpenAI’s New Tool Detects Homework Done by ChatGPT

MetaversePlanet August 6, 2024

0 2 minutes read

It turns out that OpenAI, the company behind ChatGPT, has developed a watermarking tool that allows instant detection of chatbot-generated text. However, it has also been reported that the company is undecided about rolling out the tool due to the risks it entails.

Generative AI models have now reached incredible levels. Currently, you can ask chatbots like ChatGPT any question you want and get an answer almost instantly. Whatever you need, it prepares for you in seconds. Of course, this capability can be abused. For example, students might use artificial intelligence to complete their homework, and we frequently see AI-generated content on the internet.

There are many tools for detecting AI-written texts, but most of them do not provide very reliable results. However, a report published by The Wall Street Journal reveals that OpenAI has developed a tool that will allow for the detection of text written by ChatGPT.

Contents show

OpenAI is hesitant to launch the tool that allows detecting texts created with ChatGPT

According to the WSJ’s report, OpenAI has developed a tool that can identify any text written with ChatGPT. An OpenAI spokesperson confirmed to TechCrunch that this tool is in development. Such a tool can be very useful in situations like detecting if artificial intelligence has been used to complete homework.

However, OpenAI is uncertain about rolling out this technology, described as a watermarking tool. The company acknowledges that the tool is promising but also comes with significant risks. These risks include the possibility of misuse by malicious actors or producing inconsistent results in non-English languages. Therefore, it is unclear if the tool will be released. Time will tell what will happen.

So how does this tool work?

Let’s unpack the tool a bit. According to the information, it will be very different from current detection methods and will provide better results. The tool will only work on texts created with ChatGPT, meaning it will not be effective for texts generated by other artificial intelligence models. It will make tiny changes to the words chosen by the model, effectively adding an “invisible watermark” to them. As a result, texts created with ChatGPT will be detectable by different tools.

We don’t know if the company will release this tool. However, considering that artificial intelligence is becoming widespread and such uses are increasing, developing a method that gives truly accurate results would be very useful.