With this tool from OpenAI you can recognize AI texts
OpenAI’s ChatGPT is currently making headlines. Because the AI chatbot impresses with precise texts that can hardly be distinguished from human ones. But that also harbors dangers. Now OpenAI itself offers a tool that can recognize AI texts.
ChatGPT has been on everyone’s lips for quite some time. Mainly because the OpenAI AI is able to quickly write texts that are so precise in terms of language and content that they are difficult to distinguish from human formulations.
But a new one tool from ChatGPT developer OpenAI itself now starts right there. Because the company is launching a classifier that is trained to distinguish between AI and human-written text.
Contents
ChatGPT: This is how the OpenAI tool can recognize AI texts
The so-called classifiers is a finely tuned GPT model that predicts how likely it is that a text was generated from a variety of sources such as ChatGPT.
The tool currently needs at least 1,000 characters, i.e. around 150 to 250 words, for the check. Users can simply paste the text they want to check into the appropriate field. The following applies: the longer the text, the lower the error rate.
Recognize AI texts: OpenAI tool not quite accurate
However, the tool is not always accurate: it can mislabel both AI-generated and human-written texts, according to OpenAI. Mainly because humans can easily edit the AI texts to bypass the classifier.
In addition, the tool apparently works less well when a text was written by a child. It is mainly trained on adult texts in English. Other languages also pose a problem for the software.
In a test, the classifier correctly recognized only 26 percent of the checked texts as “probably written by an AI”. Meanwhile, he misclassified human writing as AI text nine percent of the time.
ChatGPT is a problem especially in schools
AI texts from ChatGPT are apparently a problem, especially in schools. OpenAI at least states that the company is aware that the AI texts are an important point of discussion among educators.
However, the new tool is not only intended to help in everyday school life. The developers assume that the classifier could also be important for journalists and disinformation researchers, for example.
Detect ChatGPT: OpenAI classifier has room for improvement
OpenAI makes its classifier publicly available to get feedback on whether imperfect tools like this are useful. Work on recognition of generated texts continues. The company hopes to be able to provide improved methods in the future.
Also interesting:
