ChatGPT in der modernen Lehre

Now I would like to warmly welcome you all to a very spontaneous event that started on Twitter. The gpt3 chat is on everyone's lips at the moment and I thought I would like to know more about it and then a Twitter user asked me about Ms Wessels, which I immediately accepted and will now spend the afternoon with us. I warmly welcome you. It is an event that I organized together with the Center for Didactics and University Teaching of the zhdl under the direction of Dr. Helge Bad on the Landau campus, i.e. TU. Kaiserslautern and the University of Landau, which was previously part of a double structure with Koblenz, will now grow together, i.e. from January 1, 2023 the TU will shine and the University of Landau will be under one roof.

The Technical University of Rhineland-Palatinate with the Kaiserslautern campus and the Landau campus and that is why I was especially happy that we did it together in collaboration with the university didactics of the Landau campus and of course with our E-Teaching Service Center under the direction of Monika Haberer he also participated and invited other people. Who is sitting in the audience? You can tell by the names. I saw that some of our students and other interested people were also present. from all types of institutions and we are very happy that it is. I would like to come with our main guest: Professor Dr.

Wessels from the Kiel University of Applied Sciences. She specializes in NLP and project management and she was faster than all of us. Most of us are definitely faster than me, says a visionary who had already realized long before that these AI systems can be used in teaching. In August 2021 she published the first publications on AI and university teaching, for example the guest article in Die Zeit, my book. machine co-author since August 2021 along with Peter Salden on artificially intelligent writing in Dutz, this is certainly a magazine that most of us know and for the magazine's research and teaching for our undergraduates, tantalizing tools for the plagiarism and AI production in universities and not in vain came with this work and this preparatory work Project on the topic of student academic writing in the era of AI, an exploratory pilot study at the universities of Schleswig-Holstein and, If I understood correctly, this led to the founding of the virtual competence center for writing and teaching writing, teaching and learning with artificial intelligence.

Tools and techniques developed for education and science. We're so glad you're here with us today and you're discussing with us and we're excited to see you. what you will give us There are some rules for the discussion, as already mentioned, please ask questions mainly in the chat, that is the large number of people. If necessary, we will ask you again and ask you to open the microphone. We will record the event for those who came because many interested people could not come and then we will upload it in a suitable format, probably on YouTube, there is the boss in the house rules, we have journalists who are present, of course we do. welcome to talk about the event, including the content, but without assigning names, so please don't.

Professor XY asked. The student Müller wanted to know, as long as he does not have the approval of these people, I think I speak for Mrs. Wessels. saying that, of course, in principle we can both be quoted, but we would also ask for the quotes to be sent to us by email, but I think that today we are both public and therefore we can be quoted. Yes, what is our goal? a little in the chat, for me personally it is first of all a better understanding of the risks but especially the opportunities of gpt3. In Germany we always tend to talk a lot about risks, especially when it comes to AI.

It's also not entirely innocent in recent years, but I think above all we have to see what we can do with it because we can't put it back in the box. I think we are all looking for new ideas. For teaching and learning formats and exam formats. Personally, I think as quickly as 300 people gathered here and said that's exactly my topic, that maybe we should also think about whether we can form a long-term community. People who are interested in the social aspects of software, especially teaching, in this sense, I hope we all do. I have a lot to think about today and I hope everyone has fun and now I'll hand it over to the teachers.

Thank you, Mrs. Zweig, also for the very nice introduction. I felt a little embarrassed when I was announced as a visionary. Until now I haven't felt that way. I haven't seen it at all. and my observation period is from 2018 to 2022, so for four years I couldn't leave this topic alone, so I must admit that I was actually so fascinated by this topic and this fascination of it has not diminished, but in fact, has continually increased and I am actually a mathematician by training and you are trained to solve problems, so you are always given a task and you would like to solve the problem, demonstrate clarity, create order in a topic and with this topic and I think that is the special appeal to me, it's such a tough nut to crack that it really challenges me in every discussion.

In fact, I just came from the University of Applied Sciences here in Kiel, behind me you can see Kiel and behind me of course, you can see like this. Leave aside the Kiel University of Applied Sciences. I was driving from the east bank to the west bank and had a master's seminar with students on the topic of AI and ethics. In fact, this semester too. He brought up the topic of language models and one of the students said yes, who? I thought that at the beginning of the semester we started innocently, so to speak, and we were very happy to have discovered the playground in GPT 3 and now at the end of the semester we are there and we talk about BT Czech and we are already talking about completely different things.

Well, I thought that was pretty interesting before, the three-month period where students saw and then realized that the world at the end of the semester felt like a completely different world than at the beginning of the semester, so that's too much Preface So as a Virgo at least I've come to this topic, I love this topic and I also love being plagued with questions, I'm pretty sure she can't answer all your questions because we have top-notch experts with her and I'm very happy to have campaign companions from my virtual friend. or our virtual competence center for learning to write and learning with AI, so my campaign colleagues are also welcome, so if you want to add something to me and correct it, you can do it at any time, so don't do it.

It doesn't bother me. Not even through questions, but of course there is a certain order with such a large group, but Mrs. Master will be worried about that, I'm pretty sure, so I will start with the topic and that is the motto here today , namely the question How should we really treat GPT Chat in teaching? This question is actually not trivial, so I chose a context here with a lot of question marks and I thought of a very, very simple agenda that starts with something like this a little bit of classification of this topic TV because I post in a lot of the posts and what you read, sometimes I think that maybe you have some misunderstandings that have crept in, something was not fully understood, maybe you too have broached the topic again, certain things are not questioned at all, so I would like to classify it a little because perhaps there are some in this large group who are, so to speak, and I would do it. then show some examples, for example assuming they can do it in this group.

I still see examples because we get examples every day. I don't know if they can stop me at any time and say no. , no more, we already know enough and I have gathered some material for points 3 and 4. Does this mean something to think about, but perhaps also some suggestions, some ideas that I have discussed here or that we have at least discussed together here, some of which we have also published. It's actually today, I just saw when I got home, a new post of mine appeared exactly on the topic in research and teaching. I don't have the link ready. so fast, but it's today at noon, it must have been posted this morning, which wasn't there, so the teacher's current research and post that was posted today exactly about this.

On the topic of jet skis, I have, so to speak. , my most recent thoughts there, on the other hand, the background and that is very, very important in the GPT chat is, of course, or there are, of course, here the language models or the open organization language model in The foreigner Francisco Kiel is, by the way, San Francisco's twin city, very proud of this sister city and there in San Francisco in 2015 this open ai organization was founded, so the promoters of fishing intelligence are very prominent Elon Musk and Microsoft Elon Musk has withdrawn from this operating business there due to conflicts of interest. other business skills and I still have many, many lectures at the top written in bold and in red on top, without copy and pests of the construction of the existing text.

I think that's one of the big misunderstandings that you run into over and over again and then you say yes, actually that was copied together, it's nothing new and maybe you can do something with the Plag software solution. I'll take a more concrete stance on this later, but very, very importantly, it's a very miniaturized regeneration or remix. but not in the way that you can really talk about texts and pages like that and what is also very, very important from my point of view is to understand how extremely dynamic this development is and you can see this very clearly when you look back.

This very limited observation period, in this case from 2019 to today, became GPT 2 or, if you publish it in English, back then and I find it quite interesting looking back, people were worried that this comrade this system could. Fake news on the subject is too dangerous and of course we see that even today it is even more noticeable, so this candidate back then could only complete texts in English with the push of a button, so in reality existed. only this function continues, but even back then, in my opinion, it was a great revelation for all of us who used the tools worked there and yes, today we will say that it is relatively boring, but back then it was something completely new.

Here comes the bottom line: This is an artificial neural network with deep learning structures. If you have to imagine something like that, to put it bluntly, try the software to recreate the way our human brain works as far as you believe. We understand what happens in our heads and what is relevant to this is the interaction between neurons and synapses, the connecting elements, and these synapses here are, so to speak, the weights or the parameters for these networks of artificial neurons and that They rotate here and this. The number of parameters as synapses in the human brain was 1.5 billion parameters in the first version of the two versions, so 1.5 billion is of course a considerable number when you see and hear this number.

It is clear that we also have a black box problem here. , even with full disclosure, such a number of parameters is of course not something that can immediately open up where we can immediately recognize large connections, and the development was very, very fast in March, that is, in the spring of the following year . it was Corona State, so to speak, GPT was released for free and you can see that the data corpus here was quite considerable, it was also very considerable here on the left side and I marked the extra money, so here on the left side we had it in When I made a focus like this a little easier, it was about English texts that could be completed and here, in the spring of 2020, we already saw a system that can calculate, that can generate software code, that we can complete texts , that with tables.

Of course, the work can still be done in its early stages, but there are a variety of applications that we of course have not seen in the form on the left side and, very importantly, we have already increased from 1.5 billion parameters to 175 billion parameters, that is, a hundred times and it is still applied today that the level of performance is positively correlated with the number of parameters, which means an increase of more than a hundred times, but very important at that time, the organization non-profit almost became a for-profit organization, so to speak. Because Microsoft was also a sponsor of the whole thing and obtained an exclusive license for GPS, which has not yet been revealed and the source code is not always It's publicly available, so it's actually a very rapid development that we've already seen over the years.

Everything is characterized by randomness in the formation of words or in the generation of text. This is an internal feature of the system, which means that it is actuallyOsterwalde business model canvas and we let the students complete it and I thought Try that. I also think it's a demanding task, so I'll do it here first. Generate startup ideas so that the system is also solid. I am a woman who wants to start a business and two issues of sustainability and. Craftsmanship must be found together. Generate five business ideas for me, I'll get another five and so on.

We have a wonderful chat feature here, we can ask questions, we can get more answers, that's how I thought. Now comes the next exciting step, I will choose an idea, yes I will accept it. Simply put, I like the number five. Please find me knowledge of Osterweiler's business model for this suggestion and I haven't described it at all. , I entered it exactly the same way, so that was my message, nothing more, nothing less, and what does the system actually generate for me? Here I can also show some text afterwards, i.e. this is how it looks, the business model can have new elements and the tool or system actually does the following on the results screen.

Each of these new segments is specifically filled with the business idea I have here. I chose that, I didn't expect that, I thought about it a little. what's coming, but being able to present it like this without more information I think has great potential. I don't think I need to show it now because that's what blogging is about. I think the example is relatively unspectacular, but. Well, I have it too. We have a small digital Advent calendar on our competence center on the LinkedIn page and I described it there as well, so that's also something quite relevant in the economy that we use.

I can generate business ideas very quickly. and efficiently with the push of a button and knowing which approaches can also be completed, so I have a question and I think it is also relevant, so how did Chat GPS know about this? Oh, I'm sorry, I just have something, I'm sorry. I just didn't have the wrong one, the wrong one, I don't think that was your problem, I think most of us look at the PowerPoint, everything should be fine and there they show a picture of how Chat GPT free knew what was in it So what is exactly what is being asked and I think someone also asked if it will be shown on the graph, which it certainly isn't exactly?

I just thought that maybe not everyone knows this business physics at all, so I'll go If you're interested, yes, that's the question: How do I know about such a system? Obviously, through the training data, there is so much training data about this model that the system can handle this model proficiently. You can't see that from the outside, but in this case, of course, it's very common and that's why it's like that, I can also show you that, I had already opened it before, so we're right into it. away, this is what it looked like, so here I have the text that we saw before.

That was my contribution, then these five business ideas that were generated here, then I selected number 5, like this and like that, that was my message like this and then the key partners come, there it also suggests block platforms, sustainable craft companies, key activities, no, what you can do. Not only can you see that here too, very good per segment, there are 9 segments that are lined up one after the other and a pretty sensible professional segment and also several entries, yeah, so I didn't think it was a bad thing, so structuring a little The costs become a little slimmer, but the performance is surprising and that must be the quality of the training data and the many documents that obviously deal with it.

Yes, there is another example on the topic, namely math homework. I have a real, existing mat task from one of my colleagues in Kiel, a student exam task that worries me here is a business. sales sense and price functions and that's actually P1 of the model that deals with it in some way and the result is yes, but something like the total sales function that works excellently when it comes to something like contribution margin and then something like fixed costs are these 300 , which should not be taken into account because as a rule the system fails, so I have made several attempts here, it is very likely to fail, but they are very simple stories, but also what I also use here is point c, so when you multiply things like that and you have to combine the same thermal baths and so on.

It works incredibly well, that means we have and that has also been relatively evident in just a few days, I think the first weekend, that must have been It was the first weekend of December after the chat, there was already an American who has this system for the American entrance exam who studies for American universities and took this SAT exam and since the system has already worked relatively well, that means that you can see that for many exam tasks in school and in the university context, the system has a reasonable average performance here, eh, I see so many questions that now I don't know if I should answer the questions in some places and a woman, I do it for her, I look at something a little, what matters, for example, Is the quality of the results in other languages, for example in English, better than in Germany, I think German is now fantastically good, but is it even better in English?

So for me, I always thought of it like that, because I knew that before I only worked in English, for me it's more of a German reference point, so if it's there. If it works well then maybe it's a benchmark. little more convincing than if it worked in English. It's always rumored that it's even better in English. I really don't think that quality is when you notice deficiencies is that then certain terms are translated incorrectly because in the background the model is currently. more than 90% are trained in English texts, which means that when I go in with the German area it always runs in the background and translates from German to English and then from English again, you have to take that into account of course.

Of course, errors of misunderstanding can sneak through the translation, but otherwise I guess it has to be the same, no, yeah, that was a screenshot of the results, so you can see that here Um, so no , that will be the The total sales function u of x1x2 gives the total sales depending on the Mac, so that is the direct output, the direct output of this system and it can handle it very competently, even in the representation of the equation. Unfortunately, this is not always the case. correct That's the problem now, but at first glance it looks pretty clear and I would say they are very good formulas with x1 and x2 strange, but the current deficits are very impressive because I think that is also very, very important.

The deficit mentioned that there is this translation problem. but the first thing is of course the quality and timeliness of the training data, the model, the current model was last updated in June last year, which means that the later world does not know and therefore , Chat GPT no, um. , I know I haven't read it yet when the next update is supposed to run. We're stuck right now, so they're late and more than a year or a year and a half, um, I don't know. , maybe someone in the group here would also be grateful for that. The second point here is that words are very important, I have marked it in red.

Fictional texts and no facts. Open AI has been around for I don't know two years. now and you can also look at Web GPT, which is a research prototype that is supposed to be an extension of GP3 and then it also has an Internet search, which means that here we have exactly the step that helps us in many places, that is missing reports even if you then have some water yes, I'm just an AI language model and I can't search the Internet at the same time, so now we often get feedback and this shortcoming would not be significantly reduced and also Bet on the fact that with GPT 4, which is coming soon, we will have that feature there, i.e. that web search is built in there, that would be my suspicion, so implications, what does that really mean that I have that again The important thing here is that We always have to distinguish between two variants and two areas of application.

We have just talked very intensely about the first point, so it is precisely with these tools that new texts and these new texts are generated. , which I hope I have emphasized enough, are unique and not plagiarized. Now comes number two: AI-backed writing tools and GPT chat can also do this and can simply rewrite the existing text in an excellent way, that is, modify and distort it. to any extent, but if the structure remains the same in the structure, then it would still be structural plagiarism. We cannot claim that what comes out of these tools is always unique, it depends on the input, so who and what do I know if I now post one from Mrs.

Second in advice and say it please rewrite it quickly no then and I would put it under Post my name then it would be if I didn't put a lot of effort into it it would be a structural plate after, yes, if you can trace it, solve the findings, the second story, but you can also use the tools for this purpose, you can also assume that this purpose. This existing text was something else that was then widespread and capable. of plagiarism, which will probably lose its importance because it is now possible to generate something new so easily, so two loses its importance in favor of case 1 because 1 is so elegantly possible.

So what does that mean and examples from educational areas? essays taken from school that were a very good example a Norwegian student writes individually with gpt3 that means that one that has excess access per class is enough here you can do the same task 20 times generate 20 variants like that and it sold, I know it means that you make $100 or something and the good thing is that you still appreciate others and you are still successful in business and we can also assume that of course today I also received comments from a student who told my cousins ​​that the teenagers then talked to their mother when she was not there, so they acted innocently and pretended to be ignorant, but then with their aunt, when their mother was not there, they naturally said that they knew the common tools, so we can assume that too, so that's what what does it mean.

It is not only used in schools for rehearsals, but it is what we saw before. Others are actually all school subjects, say sports, but almost all school subjects are already covered in some way today and, in the medium term, I found this story very interesting for students. , very interesting. The Swiss attempt is to write a high school diploma paper, so that is also possible. I mentioned this American college entrance exam before, so if a system can do that today, I think something like that is certainly not possible. Perfect here in many cases, but we see systems here that are clearly different and with this performance cleanup we are moving in this direction, so I think the education sector is where we have it every day, which is why we held the event today. to address the question at allCan homework still function and in what form?

In what types of jobs? Final theses still work and can continue to be useful or survive in the medium or long term. And that is an extremely difficult question. I am relatively lucky because I am in a difficult situation. technical college At the University of Applied Sciences, the transfer of practice is always very, very important for us. For example, I don't accept a thesis that doesn't have this transfer of practice, so I don't want to accept anything purely literary. -Work based on, in my opinion, that doesn't fit at first. I think the type of university is a university of applied sciences, but that is too risky for me.

That means we have one. There are a lot of question marks here, but we are discussing them very intensely at the moment and I thought I would rather do point 4, solutions and food for thought, and I have something here. So I have compiled some questions that, as key questions, maybe can help us find a solution. The first is the question of what meaning academic writing really has, how we can convey this meaning to schoolchildren or, in this case, students, and how we can continue to do so. level of motivation, so I think we've all experienced that students weren't particularly motivated to write something scientific because maybe they couldn't relate to it at all, so the question here is what really is the goal of education?

I'm very, very important, so how can this writing process be better integrated into learning scenarios? At the moment it is more of a process that takes place at home. Maybe we should also think about integrating it more into teaching, that is, developing new teachers like that. Better support this process and gain better insights? I have in today's post on the research teacher. I have brought these two dimensions into discussion for the position taking process, that is, we can evaluate the technical methodological design in thefuture. but we also evaluate the tool design, what tools were selected and not only the final result, at the moment we are evaluating only the result, the finished document and the elabarat in the future.

I think we should also refer to the quality of the evaluation of this result. about the creation process and in which From my point of view, the creation process has two dimensions, so to speak, technically methodical, what tools or what procedures were chosen there and also the design of the tool, that is, any that uses AI. Supported TOOLS should do so with confidence and that should not be penalized. Rather, you can be rewarded if you have very competently selected the right tools that have then produced and helped shape this result, so the question that is always discussed should hide the fact. that the tools exist or I should use them proactively, that's where we are also from experimental writing or from our competence center, we no longer ask this question, but of course we have to use it proactively.

As a university, we do it too. pioneers of digitalization, we have to be the vanguard, we may be lagging behind, which means we are left behind No one has an alternative path forward is the motto, which also means that of course there will be a permanent need for training of all staff in schools and universities, so we simply have to have a good overview of what is happening and the tools available, and this is a market that of course is growing very quickly, there is a huge wealth of tools that dovetail with these big language models, yeah, that's also very important to understand that there are many, many writing tools with AI history if you look at them.

Sometimes you don't even realize that somehow there is AI in the background. We also don't realize that these background AI writing tools are a gpt3 or soon through calls through an interface through an API application program in the interface, but that means we know that too. So we have to know what happens in the background with these tools and I have been doing workshops since the first time on May 20 21 last year and I did it deliberately in an exploratory way, of course I tried it in a playful way, I said it These and the tools I presented to you said it at full steam, I have given you a theme and now we are trying to get what we can out of the tools, the motto is too funny, just don't write it yourself, let it be written and then we analyze it in consequence among all the possible questions and I think it is a way of approaching the topic.

I have also learned a lot from it, even from the students' comments, because we also have to identify with the text. Sometimes we have students who have writer's block, which of course can be reduced with this, but of course, we also saw great dangers here and I think media literacy has been promoted a lot if you yourself have produced false stories here, of course . Read or approach your media consumption in a completely different way, so there's a section here to prioritize and evaluate. What is the importance of literary work? Many of us had to paraphrase for years.

Paraphrasing, we don't need much more important transfer tasks yet. Creative and independent individual achievements should not be in doubt and there are oral examinations, possible examination interviews, such small disputes, defense, I think it will take up much more space and. What we need in any case is the last point: we finally need a legal clarification of copyright in Germany and with it come new forms of authorship and also new forms of declaration of independence and I really believe that. It is tragic that we do not address this issue. The copyright issue surrounding text generation is quite intense.

We do it with images, so they are multimodal areas, but we don't show you such a positive example here with text. This caught my attention yesterday. I know Michael doesn't tip. about Augsburg, but he wrote very, very well here in this YouTube video there is GPT chat for the correct search for how you can shorten scientific texts, for example, how you can then discuss the results, so here are nice and beautiful examples where You can actually do this in a very effective way, increasing quality but of course it can also be used to increase efficiency so here Michael Kipp I can also post it in the chat like anything else you should know about the tools or I can also recommend Elizabet. .org is open source based on jpt3, actually it is hardly an older version, but here you can't just search literature, so the first page appears at the bottom, you think it's just here, enter a research question and then this tool uses a semantic marker. in 200 million documents is very robust, but it also has an area called tasks and I have selected some exemplary tasks here, so you can also rewrite them, you can change the question, you can show those dependencies, compare them. with independent variables, you can summarize something, etc., so there are a lot of useful tasks here.

In fact, it is very useful to learn this tool here. Jenny is in her childhood. Anyone who has used it once, I think, had unpleasant things. , restless nights later because the tool also shows us the dangers very, very clearly. You enter a line, press the return key once or a shift key once like that, and then the tool types the port regardless of whether you have selected it or if you had selected it. You also like to have fonts built in, the tool itself or set fonts independently, meaning you will be taken to the clickable passage of text that has been generated;

On the other hand, the tool suggests suitable sources, you just need to click on it. and it does it automatically and that's almost a humiliating process when you type with it when you only have to press one key at a time. Yes, yes, yes, yes and then you produced and I think that's also a danger if you don't do it. you have a lot of time, you don't have free time, you don't identify with the task, you do it, so that's quite scary, here again such a positive example, this is a development from Kiel. Malte Hecht is a very good tool for Generate questions painted with Scholz and the special feature here is that the questions are even taken care of with correct answers and false answers artificially generated by AI, that is, the so-called tractors, if you have built them, you know that This process takes a long time.

Consumption So, for example, I always find it very difficult to generate incorrect answers and here you can get help with question eight, which makes everything very intuitive in one go. We also set up a quota sensitive application where exactly this model. including gpt3 in the background, it also takes place here, what's next, an announcement from the tenants, Facebook made me think about adding Suggest Edit and providing an explanation of its action. I would say it is a very, very powerful writing robot and writing partners. with his own actions, so I imagine it's a little scary if that's the case, but he's not credited like he would be, but I don't know, but I found it to be a development step that made me think a little, so I felt like I always moved a little bit away from my desk and continued to the back side where I say I have to make more and more space for the virtual comrade or the virtual comrade next to me, so then another announcement, I don't know anyway It is especially about the conclusion. which is also very exciting in a scientific context a Google announcement with 540 billion parameters as I said at the moment 175 in gptres and then already 540 with um flentiv five also open source 60 languages ​​1800 tasks so I also think I have Mine is very, very different, so with a view to time, here is a suggestion from me and on the problem of AI supported text production when we have students, I think we need a new form of declaration of independence and labeling and I have a ticket.

Here we also discuss it up and down in the competence center and I, that is, if we allow or even encourage the use of these tools, then they have to be tested or identified and from that Google search that is certainly not what I tried . show here with whitelist There are certainly tools that are AI-friendly, but they are so widespread and obvious that they don't need to be addressed, but anything that goes beyond that and certainly means sending a chat right now, which you would then have to specify here and of course that should not result in you being penalized, but we have to incorporate it into our evaluation practice.

It also makes sense to integrate it accordingly Yes recommendation for universities This is an old publication, which means it's all two years ago, but we have already made some recommendations to avoid primary descriptive synthesis work based on a bibliographic source. I have it here in red, but I actually already mentioned it. So it is better to do transfer tasks, rather than project, problem-based questions. , types of exams, including exam culture, I think we have to rethink many things, we have to focus more on these social skills and social interaction, certainly a separate field, the ethical aspects of copyright are much more in the foreground Conrad Art Code Code of Conduct, I think it goes hand in hand with that and basically I think there are all these teaching obligation regulations where we then have to do hours after semester weeks, so this Remuneration no longer fits the time, so they have to change roles and then they have to do it too.

I think that the general conditions are in agreement, and the presentation of our virtual school center compels and we found it now on September 1st. The background was the immediate trigger was that we were rejected. with a ki contribution from our group at a future conference for teaching, then we went to a meeting, licked the wounds and then someone said that our group would start, then soon we will hold conferences ourselves, then we will establish We got up and then we said, well , we will do it too, we will give visibility to the issue and that fits well, I think we have achieved this visibility in many places.

We want to promote research and knowledge transfer, qualify. advice and services, organize schools and carry out many workshops and we want to continue developing this and what I said at the beginning two weeks ago is important, I think, is that we network much more closely with each other because the solutions are not on the table now but it will really be a challenge here with these changes that we see that go fast enough as a heart and fast enough to adapt and they are empty and determined and you learn to find maybe a clue at the end and that's a little scary on the part by Dave Eggers of every fellow killer.

He told me last year that it is absolutely necessary that you read this because the Fuckfix algorithm is exactly his topic and these are more of a dystopian development in which we then see the power of language and the power of language is like a censorship, no, so There It is an intervention, the language is standardized, normalized, there are no longer certain identifiers and maybe you notice it a little in Pell, when a kind of standard language develops, I think I can also do many things, a linguistic culture. Be lost if we have such a high dependency on such tools there so that's it from me and I hope I haven't done it for too long and haven't talked too much and I'm looking forward to it now.

I'll ask when. with you to be great, thank you very much, thank you very much, what a pity that you cannot hear the applause that now rises here everywhere, what a pity, thank you very much for this complete blow to Chat GPT