ChatGPT, een vloek voor het onderwijs?

Find your favourite podcast on VRT-Max. I'm a little bit of a fan of Zenoo. It's the fastest growing internet platform. It took about nine months to reach this point. Instagram needed two years. If you don't use the software, you'll have to look for a different way to learn. I have a friend of mine who asked for his promoter. He could use it for his research project. His promoter said he couldn't use it. I panic in the teachers' rooms and universities, because students can write full-time letters through a chatbot. Your mentor is looking for a student who uses ChatGPT to write his paper. Maybe ChatGPT has become unmistakable for you to code, write poems or inspire people. But what is a useful tool for the one is a nightmare for teachers. Today, we have three guests, Amber, Robbe and Arnaud. Amber Steijard, hi. You'll talk to the teacher. -Hello, hi. That's right. -And you'll talk to Les. I'm Leidmeen, the teacher and the teacher. You'll talk to the teacher and the paper. There are a number of students who think they've used ChatGPT. Robbe Wilgaard, hi. You're a teacher in artificial intelligence. Yes, that's right. I teach engineering, programming and artificial intelligence. I teach my students how to work with ChatGPT. I also teach a faculty at the University of Gent. You can answer all our questions on ChatGPT. Arnaud, hi. You're a student at U-Antwerpen. Yes, that's right. I'm a student at Informatica at U-Antwerpen. And I think ChatGPT itself is a great tool that can definitely help with tasks. Okay, so ChatGPT is the end of papers and homework. And how does it have to deal with artificial intelligence? You'll hear that in this episode of "Snappt je mij nu?" a podcast where I'll talk to you about the things we're talking about. Welcome. I'll start with you, Robbe. For the background, I want to talk about ChatGPT. What exactly is that? Can you explain that in human language? I'll try. You asked me. I'm teaching, so my job has to be to explain that in human language. How old? I teach 11, 12 years until I'm 5th grade. So we should understand that if you explain it in detail. Hopefully. Otherwise I'll be wrong. We'll do our best. What do we do best? - Yeah. So ChatGPT is, if you look at the word after the letter, GPT is a generative language model that generates things. In this case, it generates text. So pre-trained, pre-trained. It can study a whole post on our language. Just like teachers do for language, you have to study it a bit. Now, we have millions of models. Millions of web pages, articles and other materials. He studied it and at the moment he can. You can see that when you use the tool, you take something in, you ask, you write an article about history, about the Middle Ages, and he'll try to write your article. So it's an AI system that, through the middle of the training, on all the data, on all the texts, has learned exactly how to write the texts on command. And that is apparently very reliable or handy. Okay, it seems very clear to me. How does that system work exactly? What is the difference with Google, where you also get something in-depth and a result? Google is a search machine. It's also a word in our language. I'm going to Google something, it's really a term. That might be short, I'm going to GPT something. But in this case, if you ask something about a text algorithm, what it does, it has a kind of wolf in its head and it plays an association game. For example, when I say a word in Koning, you think of... Philip. Yes, so Philip, he links Koning to Philip. That's because those words are in our language. For a computer, it works a bit differently, but it actually draws a line, a vector, a arrow between those words. I'm actually going to look at which words are or have the great chance to be close to each other. Koning with Philip, Konigin with Mathilde. If you ask him a question, he writes my answer about the King's House, he will actually start generating those texts word by word. So he generates new texts. And he will always look at which word is the next. And which word is the highest chance to be the next. And you see that too. If you look at it, you really see it word by word on that screen. Because that's how the calculator model works behind it. So you see the text, you see the attitude to roll out there, but on the background, it's pure wisdom. We just heard that a student... We suspect that he has actively used Chat GPT for writing, for writing. So the language teachers are a little bit with rough-faced poopers. But I sometimes give programming to my students, when I take in my programming training, almost all of my students can use Python codes, they can use it and the codes also work. If you give a difficult task, it's a little less. But my students can use that to zoom in, or to fraud, within my programming field. So we're also not safe. Arno, I'll give it to you. Do you zoom in with Chat GPT? Of course, zoom in is a big word. The situation, as you describe, if you give a letter effectively, you type it in and answer it literally when you send it. Of course, that's not the intention. Chat GPT can in some cases compare it to Google, because you don't want to type it in and give it as a result, literally copy-paste and send it. It's a tool that you can use as a personal assistant, where you ask or get extra ideas. And then there's the question of when something is missing, when you can use it as a personal assistant, without doing it with an annular goal. You don't take full action with that. No, because it doesn't work in a second master. How do you use Chat GPT in your homework? I use it a lot. For example, last week I had a presentation of someone and they said a word that wasn't included, and I said, "We can explain it in simple terms." During the presentation, you can follow up with a very precise summary. During studies, you can use it in a similar way, but you can also use it to generate questions about the teacher. Specifically in tasks, there are also many ways to use it. For example, for presentations or papers, you can do inspiration through structure. And sometimes you can also make small adjustments. For example, today I saw a lecture of 50 students and I wanted to put all the ideas between assignments. You have online tools, but you also have to look for them. So you can ask Chat GPT, "Hey, can you put these ideas between assignments and don't do it?" Do you already know how to use Chat GPT? My students are not at my level. I have a friend of mine who asked for his promoter, if he could use it for his research project. His promoter said that he could not use it, probably because there is no clear legislation for it. Yes. Ambrie, your students are probably using it. That's right. - Why would they use it? How do I imagine it? It changes a bit. We have one student who we suspect that he wrote a large part of her paper with it. Why do we suspect it? Because there are 40 or 50 references that he used for her paper. 40 out of 50. - 40 out of 50. And what Chat GPT does, I can ask him to write a paper with references. Robben has been working in a language software, so he will understand the structure of a reference and he will write an author and a year and a title. But the reference does not exist. That's why we saw that it was very suspect. We suspect that one student knows that it goes around very quickly, her other students. We also know from other software that, for example, they bought a group license from that software, so there are a number of softwares that students use together. The longer they exist. - The longer they exist. For example, it's Killbot, that paraphrases you. It's a bit like Speak-O-One. There are a number of other students who we suspect. It was a time when they also had to send us a round for feedback. The first time was before December. Chat GPT was only launched in December. We could see the difference with the papers that they uploaded before, and then their end papers. And then there was a mega and a beautiful piece, where they thought there was still a lot of work to do. Then there is a doubt, but it's very difficult to do. We'll get to that later. Do you use it yourself? Yes, I use it myself. What for? I use it in a non-professional context, especially when it comes to laughter. I've asked myself a story about living in a Christmas house with my mom. Yesterday, I had a friend who was looking for a travel destination. And we said, "Oh, I'll give you 10 suggestions for original travel destinations." But I also use it in a non-professional context to create something together with my own work. For example, I wrote a paper to ask questions like, "Can you write a conclusion?" and then inspiration. So that's important. I think that's an important nuance, so I'll never take it lightly. I use it sometimes as inspiration, but then it's still... There's always a reflection needed to make something of yourself. There's a lot to do online, too. I scrolled through Twitter a bit. I found a tweet of Tipi. "Teachers are becoming so stupid and incompetent by the chat GPT." "That's the nightmare of every teacher." Sorry for the language break. Arno, are you the speaker of the chat GPT? I'd like to say that you're getting more productive. You get more work done in the same amount of time. But eventually you can do a lot more. You can go deeper in your research. So they certainly don't become stupid. But at that level, if you're a first-year student, you often have basic programming exercises, simpler invoices and so on. And that is very dangerous for the chat GPT because they don't have enough time to copy and paste. Amr, you probably have courses that are not perfect. What do you do when you get a student who thinks you didn't write that? How do you quote that? How are you going to deal with it now? We're going to deal with it in the same way we would do if we had a student who's in charge of plagiarism, so something that can be taken over from other papers, where we already have a good software to check. So that's a bit different with a plagiarism checker. So a plagiarism check software can be easily seen, because it's all taken over from existing papers. With chat GPT, it's much more difficult to do it at the end, as I said. Because you have a number of softwares that could say, "It's probably not for a person to write, but it can never be 100% certain." And also in different articles, the reliability of that software is already in question. And is that already widely discussed with the agent? Are you talking about teaching, or what needs to be done? Oh, yes, in the meantime I'm at every lunch. Yes, it's going to be a lot of fun. We're here today, and we also received a mail from the agent, which was a sort of first stand point that they're talking about, but also a number of tips about what we have to do with it. I would appreciate if they give you tips on how to use it in your practice, and what you can do with it, and what are the advantages of it. And also try to make students aware of the possibilities as the dangers of chat GPT. On the other hand, they also have a number of tips about evaluation forms. Because now something that really lives within the agent, but I think within other universities as well, I think the big guideline is that if chat GPT could make it, you can better understand another evaluation form. So, for example, we have a blog next semester. Students have to write a blog about a certain subject, so I think it's very easy for you to ask chat GPT. So we're probably going to have to go to another format, where the students will reflect more in the class, with each other in the discussion, instead of leaving it at home in a blog form. So you're trying to make it really active, that it can be used. Yes. Do you believe that students will be taught by the horizon? Or do you really believe that they will be taught by the horizon? I think it can. It depends a lot on how they use it. For example, I wouldn't be able to do my lesson, I can ask students to look at the abstract of a paper and ask chat GPT to make a simpler version of it. If you don't understand the abstract, you can also ask chat GPT to look at what that means. So in that way, I think students can be smarter. On the other hand, I think that if they really use it, even if we suspect that one of our students has used it, to write a paper for them, I think you can be the dumbest one, because it's not a software that gives you more explanation, or that explains the way you did it. For example, a software that I often play with students is an academic writing tool, where they're actually not going to say, "This is the fault of grammar, this is the recording." But they're also not going to explain why that is a fault. I think if you use that, you still have that reflection as a student, and I think that's a mistake with chat GPT. You said, hang on, how did you use the tool? I got a speech from a school teacher at the 4th middle school, who used chat GPT for almost all of the tasks. Personally, I really use chat GPT very often for school. Mostly for the Netherlands, where I write reviews or summarizing, about certain texts or books, but also to write my emails or to improve my texts. Because I make a lot of grammar mistakes, pen mistakes and so forth. At my school, teachers didn't keep saying that I use chat GPT, but I think that's also part of it, because I really make sure that this is not recognized by chat GPT. I let it be written more normally by various AI robots, so that it's actually not recognized. Yes, that's right, Roque. Probably already a number of people. Robbe, I hear two things here, right? I hear the fact that they make a lot of spelling mistakes and grammar mistakes and improve that. On the other hand, they can also do it secretly, so that the teachers don't recognize it by various AI bots. Robbe, I follow you on Twitter. I saw that you had a test with your students, in which you can explain yourself, in which you can describe yourself or write chat GPT. Yes, that's right. So the teachers had a lesson about the Turing test. What is the Turing test? It's a little bit of philosophical proof. In the 1950s, it was long before there was a talk about artificial intelligence or chat GPT, when I thought we might be able to do a test with a jury, one robot and a person, and they lead a chat conversation. If the jury doesn't know who the robot was, I would assume that it was the robot intelligent. I thought, okay, but if it's written together, and it seems quite relevant to say, "Look, you can divide yourself into three groups." One part of the class wrote it for yourself, and the other part of the class wrote it for themselves. One part did it in the 1950s, just like the teacher you just heard in the quote. And one group did it all in the same way. Because doing a Turing test or doing some lessons for a longer period of time via plagiarism, taking things from Wikipedia and that sort of thing. They did the Turing test, and I went to a detective to find out who did it. I had all of them, except one. Except one. Only one. So I had all of them, and I was able to classify them by putting a lot of work into it. One of them was just like the teacher you just heard in the quote. So he wrote the text, and then he worked on the following alternating rules. He worked on it himself. And he also added his own examples. We saw this in that lesson. We saw that in that lesson last year. So he did go to battle with that output. And have you heard of the AI detection tools? Probably. Yes, so the AI detection tools don't work that well. When I put my own text in there, he said it was really written by an AI system. And above all, I don't need computer chips, but I did that myself. When I put a text in there of Ovidius, it also turned out to be written by an AI system. I don't think Ovidius was in the history of the AI system at that time. Yes, that's for sure. No, certainly not. I also think, if you get out of it, that we're going to try and approach this through all kinds of detectives. This is a cat and mouse game that you don't want. The virus will get better, the antivirus will get better, the vaccine will get better, the virus will get better. You're in an endless loop, you can do that. But I don't think that's the best way to approach it. I have a question. I've been asking myself for a long time. Would it be possible for the GPD to put a kind of tracker on your text so that they can recognize it themselves? Yes, but it's the same if you ask for example, a watermark can be put in a photoshop to capture images. Maybe in the metadata, it's not literally visible. There will be someone who will find a tool to get the watermark out. And then we're at 11. So that will work for a day, two days a week. And then they will find a way to go around it. And then you stay behind the facts every day. And as an evaluator, as an expert who wants to give feedback, it's an endless fight. I would rather stick to studying students with that tool. How do you use it? How don't you use it? And maybe also look at our own evaluation methods if your question isn't robust. So let's say, as simple as the students and students say, "Ask that robot, sir." Maybe we should put our question mark higher. I want to go further on the other part of the speech of the schoolteacher. She said, "I'm particularly bad at acting, and grammar, and I've improved my writing skills a lot." Of course, without saying what's wrong and why it's wrong. That might be a stretch, but do you think there's a link? You've heard a lot about the bad language at schoolteachers. That they're getting worse in the Netherlands. Do you see a link with technological development? I think that you can draw a link that you don't know at all. Our Netherlands is already very angry at it. Even with the students who are now at the university, such as Jacky Pettyn, who was launched in December, so it wouldn't make sense to put it that way. The other part of the speech is that she said, "I'm not so good at language. I use my most common language." I don't see a problem at all. If there's a teacher who says, "I'm an individual businessman, I have very serious dyslexia, I'm sorry, but writing without fault will always be incredibly difficult." There are situations where the teacher uses this with a writing problem. The school just says, "You're not allowed to do that." But that's the equivalent of taking me or Arnaud's glasses off. That's not fair. Something else, Arnaud. How do you like the system? Did you use it for work, for research, for coding? I think it's very good. It's true that it achieves its goal very well. It gives very human-realistic texts, but it certainly has its shortcomings. The truth is, it can give false sources. It can just give false information. I even made grammatical mistakes, which is very strange. But people do it too, so that's logical. But on the other hand, it can also come with very old-fashioned solutions, which you can't think of yourself. For example, if you think that you need an idea to get more followers on social media, that can give you perfect inspiration. Do you use it for inspiration? Do you make mistakes? Yes, of course. If you say that you make mistakes, you make mistakes, but most of the time you don't use that input to make mistakes. You take out the right ideas to find more of them. But do you feel like you're learning something by taking out those mistakes? If I generate something, I ask myself a question. It's much easier to be an interlector and take out things from zero with an idea. Do you feel like you learn a lot if you have to improve a code or a text? If you have to create, how do you react to that? If you want to do that, taking out those mistakes, it's obvious that you know something. So without knowing a thing, without knowing what it's about, you're nowhere to be found. And then you can only experience the output of that system. If the GPT system or other GPT system writes code and that code doesn't work, you can take out the mistakes you know from programming. And I don't see you either. Do you have an agreement? Yes, I have an agreement. I think we were wrong about that. I don't think a student even learns a lot by writing a paper by ChatGPT if they go through the whole process themselves. Because the paper itself is not... In our department, for example, the paper is the end product. What is it? It's called Scientific Communications on Rural Development. It's about how you write a paper. So the whole process. It means that they learn from how you read the paper. That's the beginning, because they have to base their text on paper. So how do you read the paper? How do you make a critical reflection of that? But for such a department, ChatGPT is a power, Marie. Yes, for sure. The basic skills you have to learn from it, how do you generate them? Yes. I have an agreement that, as Robben said, it's very important to make students very clear that they know why we're doing that exercise. And if it's not true, if they keep using that software, they have to find a different way to learn the competencies themselves. Robben? What is often said is that it would be a sort of equalizer. Everyone can use it. Everyone can actually write well as other students, but that's also not true. Equalizer. It's now or it will be paid for by everyone in short. That seems to me also a free lead for inequality. I think it's not going to have that much effect on paying. Because what I understand now is that paying is going to be a priority in the future. But when it's paid, there are also alternatives. ChatGPT is not the only one. And if I don't forget, Google will also have its own tool. And it's going to play competitive competition at a very high level. Who will be the best tool that will win the most of the users? And will be the paid model, which probably won't be available to everyone. Google is still free after all those years of searching and seeing. Because they just have a different business model. Maybe you can get ChatGPT ads for products in short. Maybe it can be a model to keep everything free. Is there a clue? Maybe a little bit, but I think it's also a challenge of the bot itself to make that cloud as small as possible and give unanswered questions as well as possible. Do you think that the bot will take care of it? That it will be in there? Or that people at Google will be in there? We are currently in 2023. I think over the years, as long as people continue to use it. If more people use it, Google will definitely be in there. Julie, you're wrong. They say in touch, ChatGPT says we want to be a software for everyone and we want to make it easier for everyone. I don't think so. Why would they sit there with me? I wonder if they would get better of themselves. But I do think so. Even with the digital cloud, we try to solve it by providing schools and seniors. I think that with ChatGPT, it's the same. We have to bring that one to the lower school. And two, people who are not yet at home. We have to give them options to go to school. There's only chat about ChatGPT in the media for the last few weeks. Why is there suddenly a lot going on? Those models have been around for a long time. I've been teaching for two and a half years. I've been teaching for three years with GPT models. My students have done everything they can. I've worked with colleagues and they have come this far. What's striking is that with GPT2, the parents of GPT3, 9 out of 10 results were very bad. This is GPT3.5. It's not official, but it's 3.5. So 9 out of 10 results were bad. One out of 10 results were a bit usable. Now you have 9 out of 10 results that are usable. And one out of 10 results that are hopelessly spread out, a tic-fault or a grammatical fault. It's much better now. Plus, you used to have to have a website or you had to run something on your computer. You had to install something. You had to have knowledge of it. Now it's a chat conversation on your smartphone, your surf website and you have the price. What OpenAI has done great with GPT, but also with Dali, the system where you want to take a bus on the moon on a bike, and sign it for you, is to make it available for everyone. Democracy is available. Just say, here's a search bar. Have fun with it. And everyone was able to use it. And they went from 1 million users in five days. Now they're talking about the 100 million unique users. It's the fastest growing internet platform. It took 9 months to reach this point. There were two years needed on Instagram. So the chat GPT has apparently reached a zenith. In a big month, they came across some unique users where other big names took up space for two or three years. And nightmarish for the signers and the artists. Yes, but I also think that many writers, if you ask a chat GPT, write a story about Harry Potter in the style of J.K. Rowling, and he begins to write about it. I also do the tank training with the students. We make a new book, and then we put it on a standard book handle for 20 euros. What do we do with it? You tell me. J.K. Rowling never gave the impression that a robot, hair writing style, can copy. So that's a problem. You're also going to the slaughter with something that someone else, an author, really wants to own. That student gets 20 euros for it. For that book written with chat GPT. How do we divide that 20 euros? Is that 20 euros for my student? Is that 20 euros for me? Because I did that. Is that 20 euros for chat GPT itself? Because it's always written in the EI. 20 euros for J.K. Rowling. That's a question that we still don't have a legal solution. And here we're going to have to do something about it. Do you think it's a curse or a blessing? Amber. In my job, as it is now, I would say it's a curse. But I think my job has to adapt to the software before that. How? Actually, it's an evolution that we're working on now. By working more with other evaluation forms, by also evaluating students more, for example. By evaluating more in the meantime. That's an evolution that has been working for a longer time, where you deliver one task at the end, which takes place at all points in more than a year. So I suspect that we'll be able to do that in a short time, thinking about how we can do that better. Back to pen and paper. No, I don't hope so. I hope that we'll think about it sooner, about a new, innovative method that will go back. Because I've already heard from Prof. who said, "Okay, back to all my students, all with pen and paper." Or mondial exams. Or all mondial exams, I don't think so. Yes. Arno, is it a blessing for you? Yes, it is a blessing. I fully understand from the evaluation perspective that it's a blessing, but it's up to the future to let people do more, to be productive, so that everyone has their own personal assistant. It's really a blessing that we can use it a lot for the future. You're going to keep using it. Certainly. I see a dystopian future for me, where the students in the lesson learn how to deal with Dalí instead of getting a drawing lesson, where we make music with AI instead of forming pianists. Do you see that for yourself, Robert? Always welcome to come to the lesson. Next week we'll talk about Dalí. We effectively draw with an AI system. I'm not saying we let an AI system draw. We draw together the most. So I see those things as a challenge. There are a lot to stay in, in one form or another, whether it's better or worse. I think we should make the students and the lesson warm, or at least learn how to use it correctly. Don't you see a lot of jobs to disappear? I think there are a number of jobs to be done in a different way. You're drawing on the artists, writers, artists, musicians. I understand, and we do that as well. If you have a drawing style, and you're still alive, and you can ask a robot, I want something in the style of "up, up, over me". And you take that in, that it's a bit of a stretch, that the system can do that. That the system can do that with Van Gogh. I don't think Van Gogh did that on his drawing. Now, it's not possible. If you say, "I use the drawing program myself." I'm working on making a children's book with something like that, a youth literature. So I couldn't do that before. I was a painter, but I didn't need to draw. But by integrating those goals in the rest of my arsenal of software packages and skills, I can get on with that story, in a different way, depending on the material. So I think you can do things with that, but I understand that sometimes you feel a kind of danger. You feel, "Look, I feel my rights and my skills." People win jobs. That's an extremely talented job. Yes. Do you think that's a romantic picture of the future? Maybe you don't have to ask the two IT guys, but... No, when I hear you talk like this, I think, "Oh, that's a bit scary." I think we always need artists, writers, and artists. And software has its limits, too. At some point, I suspect that software has to be implemented. If there's no original content, if there are no people, no more things to write. If it's going to be a loop, I'll keep writing. I hope it will never happen again, but I haven't seen that software yet. I've seen it myself, but I thought people can't do this better. Let's hope for the sake of all the artists, journalists and writers. That evolution doesn't come too quickly. Thank you for sharing your expertise and your stories. Merci, Amber, Robbe and Arno. Where do you try to convince your friends? Where are you tonight from Bakker? Send me your ideas through the Instagram page of VRT News or send me a message on Twitter via Orillie, Laagstreepje, Buffet. And maybe you're the guest in my next episode.