Lyon: 50% of its students cheat with ChatGPT
This university professor is already confronted with the ChatGPT phenomenon.
It didn't take long for ChatGPT to become a problem in French universities. A recent example is provided by our colleagues from Progrèswhich relate the bewilderment of a teacher. This handicapology teacher in Lyon had indeed given a written assignment to be completed at home to his Master's students.
Most teachers are concerned
Reading their copies, he quickly realized that something was wrong. If they were not strictly similar, they all used the same demonstrative structure, and used very similar examples. He then asks a student who tells him the sad truth: 50% of students have used ChatGPT to write their homework.
The teacher is nonetheless surprised by the quality of the work rendered, especially for a bac+4 level. While ChatGPT is not strictly prohibited, and the copies were worth between 10 and 12.5, he decided to award the mark of 11.75 to all students who used artificial intelligence. These works are now in the hands of the university management, which is carrying out an investigation.
According to the teacher, his story is anything but an isolated case, and talking to his colleagues, he understood that three quarters of them were facing the same phenomenon. This issue would also massively affect high schools, which ultimately raises the question of the relevance of homework.
France is not the only one to be concerned. We recently learned that schools in New York have decided to ban the use of ChatGPT on school networks and devices. According to them, the AI is not reliable enough and it poses obvious cheating problems because students can attribute a text to themselves without even understanding it.
OpenAI tries to reassure
For its part, OpenAI has understood the danger. In particular, the company wants to add ” a secret and imperceptible signal “ to its texts in order to show that it was written by an AI and not by a human.
Company CEO Scott Aaronson explains: “Empirically, a few hundred tokens seem to be enough to get a reasonable signal that yes, this text is from an AI system. In principle, you could even take a long text and isolate the parts that probably come from this system and those that probably don't.”.
But this process is not entirely convincing. the world, and some experts believe that this device could easily be circumvented by modifying a few elements of the text.