Edward Tian, a 22-year-old IT student at Princeton University, announced just before the new year that he had created a bot that detects using ChatGPT.
ChatGPT has been causing problems for Norwegian educational institutions since its launch at the end of November last year.
A TV2 survey showed that only three universities detected the use of the bot in relation to exams.
Many universities, colleges, and experts believe the hidden numbers are large as there are currently no systems in place to capture bot usage.
Directorate of Education I also decided to restrict internet access during certain high school exams pending more knowledge on how to handle a chatbot.
FRESH CASE: A student using ChatGPT was caught during an exam where all aids were allowed Photo: Helge Hansen
Exam answers are usually checked with a so-called plagiarism check. The text you submit is then checked against texts already available on the Internet.
ChatGPT writes new scripts, and thus is not caught in plagiarism checks.
But perhaps a new “robot” can put an end to that.
You will “fight” artificial intelligence
The 22-year-old wrote this on Twitter:
I’ve spent new years creating GPTZero – an app that can quickly and efficiently detect whether an article is ChatGPT or human typed
– Edward Tian (@edward_the6) January 3, 2023
GPTZero is the name of the robot. According to the site NPRthe bot must measure the text by two parameters: confusion and continuity.
If the text is mistaken for GPTZero, there is a greater chance that the text was written by a human. If the text does not confuse the bot, there is a greater chance that it was written by the bot.
The second parameter was continuity. People have long, complex sentences and short simple sentences apart from each other. Bots usually only have one of two options.
The young student admits that the robot is not perfect, and that someone managed to trick the robot. However, he believes this is the beginning of a “fight” against artificial intelligence.
– For too long, AI was like a black hole in which we didn’t really know what happened, Tian says according to NPR.
Made in the afternoon
Inga Strömke is one of Norway’s leading AI experts. She believes that it shouldn’t be too difficult to create a bot that can detect ChatGPT usage.
It states that there is a machine learning model inside ChatGPT called GPT-3.5. Its construction is a secret. Microsoft owns the licenses for this model.
But the previous version is called GPT-2 is available openly. The two are completely alike. So GPT-2 can be used as a detector for ChatGPT. It already exists.
Strümke believes such a detector could be made in an afternoon if you had the experience and access to the different models.
– The way this is done is by exploiting the fact that modern linguistic models do one thing, which is say the probability of subsequent words in a sentence, Stromky explains.
– If GPT-2 gives a high probability of all words that a text contains, then there is a high probability that this text was written by a GPT variant.
Expert: Inga Strümke says that owners of GPT-3 licenses, which is Microsoft, can easily make a ChatGPT detector. Photo: Mona Hooglede
Strumky also believes that this cannot be called a “plagiarism check”.
– Detecting that chatGPT has written script is not about checking whether the text really exists, but about saying something about how likely it is that a GPT model wrote it, the researcher wrote.
– long-term
At the beginning of January, TV 2 asked the same questions to 30 of the country’s largest higher education institutions. One of the questions was “How did you discover the use of ChatGPT in the exam?”.
Almost all of them answered that they had no prerequisites for detecting bot use, because a bot would not affect the plagiarism checks in use today.
No cases: The University of Oslo has not yet detected a robot being used in exams. Photo: John Olaf Nesvold
They are aware of the existence of such bots, says Vice-Chancellor of the University of Oslo, Bjorn Steinsäcker.
We know there is technology available that can help us detect if bots are involved. It does not change our starting point. It’s not about bans, it’s about adapting to technological developments, Steinsacker tells TV 2.
The future: Vice-Chancellor Bjorn Steinsacker says the university will set up a working group. Photo: Jarli & Jordan/UiO
He also did not think it would be appropriate to use bots to review exam answers that were submitted before the New Year.
Stensaker also says that the university has already established a working group on the digital learning environment which will discuss, among other things, how the university should approach AI.
The most important thing we can do is look at the long-term goal. How can this be used in teaching and assessment in the future?
Watch the video: – Computers will never be able to compete with art
“Web specialist. Lifelong zombie maven. Coffee ninja. Hipster-friendly analyst.”
