We have tested two tools that detect content generated by ChatGPT and other artificial intelligences: GPTZero and Draft & Goal. We can review them together because both perform very well. Let’s explore the contexts in which they can be useful. We will also reveal our favorite because, although both online software are reliable, one of them stands out in particular.
Why use AI text detection software?
These two software tools, accessible from a web browser, are valuable for controlling content generated by ChatGPT and other AI systems. One of the primary concerns raised by the next-generation AI of ChatGPT is the possibility of cheating in schools and universities. For several years, teachers have already had to be wary of the internet, which is a fertile ground for plagiarism.
With its ability to provide square, complete, and fluid answers (even though its limits and flaws are apparent), ChatGPT will inevitably make things even more challenging.
Beyond academic areas, access to such specialized text generation capabilities can compromise the authenticity of web articles. Web editors need to be extra vigilant when entrusting a mission to a content creator. It would be a shame to pay someone to copy and paste content produced by an artificial intelligence, wouldn’t it? Fortunately, detection tools were quickly put in place. What are their strengths and limitations? Is this the ideal solution?
GPTZero, Draft&Goal: two tools that convinced us
To form our opinion, we have tested both detectors (GPTZero and Draft&Goal) using some sample texts
We have good news: the results are in line with our expectations! Both programs consistently succeeded in identifying whether the text was machine-generated or human-edited.
• GPTZero clearly displays its operation using the “perplexity” system, which measures the extent to which the text was generated by an AI. In other words, the more nuanced and verbally complex the content, the more likely it is to have been constructed and shaped by an individual.
• Draft & Goal provides less precision regarding the detection criteria. It looks for an imprint left by artificial intelligence, which the program detects.
In any case, regardless of the method used, the results were reliable. Of course, we cannot guarantee that it works every time, as extensive testing would be necessary, and there may always be some doubt. However, fooling the detector is not easy, which is excellent news.
Two quality solutions, but the winner is…
We stand by our earlier statement that both text verification tools impressed us. However, GPTZero is our favorite for two reasons. Firstly, its method is highly transparent, and we know precisely why the program determines whether the text was generated by a robot or a human.
Secondly, while Draft&Goal currently offers a French interface, it cannot yet process texts in the French language. This may change in the coming weeks.
We now invite you to discover these two online tools, which have a reassuring aspect: complete replacement of machines by humans is not yet on the horizon 😉