OpenAI presented an AI that can identify texts written by other AIs.
Having created the “problem” OpenAI now also provides us with the solution. The company presented a tool capable of sgamming the texts generated by ChatGPT.
OpenAI , the leading company in the research and development of Artificial Intelligence, has announced that it has created the AI Text Classifier, a tool accessible via the web that allows you to find out if a text has been written by a human being or by an AI.
This tool was trained on text samples. A huge database of texts actually written by men and others generated by artificial intelligence, such as the now famous ChatGPT (also by Open AI). The database includes several Wikipedia entries, as well as texts taken from WebText and other sources.
The tool provides five different grades of assessment, ranging from “Very Unlikely” to “ Probably AI Generated ”. Those who have tried it explain that to date the tool is rather inaccurate and it is actually very easy to deceive it and pass a text written by an AI as authentic. For example, it would be enough to ask the artificial intelligence to rework the texts it has previously written.
Open AI itself specifies that AI Text Classifier should be used as a last resort and cannot replace a thorough investigation. “Use it together with all the other criteria that are normally used to determine the authenticity of a text”, its developers are therefore preparing to specify. The fact that the classifier is imperfect today, however, does not mean that it cannot become a much more reliable tool in the future.
The popularity of ChatGPT has created a new need, especially within schools and universities, where the chatbot is used more and more frequently by students to cheat and produce term papers and papers with a simple click. There are already numerous schools around the world that have strictly forbidden the use of ChatGPT and any other generative AI.