Yes, news about the AI developed by OpenAI has been coming at a frantic pace since December 2022. The name ChatGPT is on everyone’s lips. Some speak of a revolution, while others fear that this new program will open the door to all sorts of excesses. The most common example is that of a student entrusting their homework to the bot. But there are others…
Faced with these concerns, more or less justified depending on the case, reactions were not long in coming. In one of our recent articles, we introduced you to GPTZero, as well as Draft & Goal. These two online software programs are designed to detect texts generated by AI.
Today, however, we are going to talk about an “in-house” AI developed directly by OpenAI, the AI Text Classifier. Let’s see what the firm offers as a solution to cheating attempts and other fraudulent uses.
AI Text Classifier: an indicator more than a detector
The AI Text Classifier’s approach of providing estimates rather than definitive verdicts is similar to that of GPTZero and Draft & Goal. However, the AI Text Classifier uses vague expressions like “probably” and “possibly” to rate submitted content very broadly. This caution is likely due to legal reasons, and the AI Text Classifier is more sober in delivering its results compared to other similar software.
The reliability of the assessment is crucial, and it’s important to evaluate the performance of the AI Text Classifier compared to other detectors. It would be ironic if the performance was not up to par. Therefore, it’s essential to assess the AI Text Classifier’s effectiveness and accuracy to determine its usefulness in detecting fraudulent content. Without further information on its performance, it’s challenging to form an opinion on the AI Text Classifier’s effectiveness as a detector.
A reliable verdict during our tests
It’s good news that the AI Text Classifier was able to successfully detect the work of another AI when submitted a battery of texts produced by ChatGPT or humans. In particular, the AI Text Classifier was able to identify the work of an AI when asked to write about cats in 200 words. The software was also able to detect the human touch when tested with the same topic. Similar results were obtained when testing the AI Text Classifier on other topics such as the reign of Louis XVI or toasters, covering a range of different subjects.
While these initial results are promising, it’s premature to declare victory. Further testing and evaluation are necessary to determine the AI Text Classifier’s reliability and effectiveness in detecting fraudulent content. It’s important to assess the software’s performance on a broader range of topics and in more complex situations to gain a better understanding of its capabilities. Overall, the initial results are encouraging, but further testing and evaluation are necessary to fully understand the potential of the AI Text Classifier.
AI Text Classifier: a satisfactory tool, but we will remain vigilant
It’s important to note that OpenAI did not promise absolute reliability for the AI Text Classifier, and it’s likely that future versions of the software will continue to improve in precision and nuance. While the initial testing of the AI Text Classifier has shown promising results, it’s important to approach the software with a critical spirit and not rely solely on its estimates. Additional analysis, particularly by humans, may be necessary in high-stakes situations where accuracy is critical.
In summary, the AI Text Classifier can be a useful tool to confirm or refute intuitions, but its estimates should not be taken as the final word. As with any technology, it’s important to understand its