Are you planning to have your thesis or correspondence done by ChatGPT from now on? Hold off on that for a while, otherwise you’ll be glowing with GPTZero.
The beginning of a revolution
In the month since ChatGPT was released, AI has caused quite a stir around the world. For example, the chat client appears to be used en masse by students to write papers. People know how to write their e-mails a lot neater, which is very positive. And the art world is also not happy with the highly realistic imitations that openAI’s Dall-E 2 can make of artists.
For example, a comparable program won an important art prize with digital art, while the jury did not realize that the creator was not a human being, but a computer. This problem inspired the creator of GPTZero.
ChatGPT still very unreliable
Dangerous, because ChatGPT regularly makes stupid mistakes. You should not think that, for example, the prescriptions of medicines, the manual of an advanced device or a court judgment are written by ChatGPT and contain a gross error that can destroy human lives.
Fortunately, Edward Tian, a student at the leading American university Princeton, has found a solution. At least, for now. His online web tool GPTZero can immediately check whether a certain text comes from a human or from an AI such as ChatGPT.
How does GPTZero work?
It turns out that AIs use rather stereotypical language and rarely combine unexpected words. We humans do, because we understand what we are talking about when we say something. And therefore also being able to use words that rarely occur in combination with other words, but do convey the meaning.
An AI like the one behind ChatGPT only looks at the frequency with which words occur in combination. That is also the way in which you can discover whether an AI is writing a text. GPTZero uses this method.
GPTZero not quite 100% accurate
The disadvantage of GPTZero is that this method is not entirely reliable. In about one in five cases, the tool gives a false alarm. Funny, of course, if your piece of work or article that you’ve been working on for a long time is wrongly rejected in that way. It could even cost you your job. Also, people who use artificial intelligence, for unethical reasons, will of course invent all kinds of methods to avoid this detection method. Consider, for example, the arms race between spammers and anti-spam filters, which is becoming increasingly sophisticated.
The last word has probably not yet been said on this subject and we can expect many new developments. Until that moment, actually always, you better have fun with ChatGPT and try to avoid the protections.