Professor gegen KI! Kann ChatGPT meine Klausur lösen?

Teacher vs. AI can solve my exam. Yes, welcome here to a new episode of Follow Me. I am a teacher. My name is Holger von John Dietrich and I am a

professor

of business informatics. This is the data science channel. The data science we always mention. up to date data driven analysis and also look behind the scenes Yes Chat GPT is the new shit that many of you have already heard, some of you may have already tried it, it is a new artificial intelligence in the form of a chatbot with the that you can talk, like with a human.

You can communicate, you can do all kinds of tasks, you can ask questions, you can tell stories, you can write poems, you can even solve tasks in rap battles, etc., of all kinds. of languages ​​on all types of topics. It's amazing here and I have to be honest, I have been working with artificial intelligence for many years and I am completely amazed at what this system has been able to achieve personally. I didn't think it was possible, so it's worth everyone checking it out right now. Free, that will probably change soon because the costs, because there are already so many millions of users, are probably immense, which is what you hear and I think they will soon change it to a paid model.

Everything is based on the so-called large language model LLM, this is a neural network structure, the so-called transformer, but we will not do it. about the technology in this video. I'm going to do another separate video, which is what you're doing in this video, and I have something very special for that. I'm thinking about it, I'll just give you an idea. Some of my exam assignments and see what comes out of them. I myself have several lectures in the area of ​​business computing, data science, programming, statistics, etc., so I chose five tasks and we met. Check it out. each performance there, of course I will not go into depth into the respective topic, some of them are of course technical questions that would go too far here, but I will just try to point out the difficulty of each task for each task and we will take a look.

Let's see how well Chat GPT performs here at Basically, there's the first task, which is a programming task, that is, a task where it involves having to evaluate feces. That's not trivial enough here to do. Take a closer look. As you will see, this is a function that is defined there and that has the name Fu and this function that also appears here again, so the function defines itself, so to speak, and there is something like that in programming called recursion, the so-called recursive recursive function, yes, then one or the other Ausform student arises, let's see how they think about it accordingly.

Yes, this function is a recursive function, it is recognized exactly correctly, which means. that the function refers to itself in its expression, very well in this case the function is repeated over and over again as long as the value of X is not equal to. So everything is explained here too, so that's a great start. I have to say that the next task is the other way around: he must write the system himself or, in this case, the student must write himself the code for a binomial coefficient that is used, for example, to calculate how many possibilities there are in the lottery If I get 6 out of 49, that is exactly this task here, the particular difficulty is that there is already such a function in R, so the R programming language is called juice but it is not used for that, so it is excluded explicitly so you have to find a detour somehow and here I also have to say it is brilliantly solved so it goes back to the definition of binomial coefficients via factorials and yes the code is there with the appropriate comments specified and then everything The thing should be called like in the lottery the number 6 out of 49, which would be this bin Köf, so the function should be called 49.6 and it even gives the value of almost 14 million possibilities.

Here again the maximum score. Next is a comprehension task. Frequent change. of the doctor makes people more likely to die a provocative headline in the news scientific reports and so on analyzed the results of 22 studies from new countries with different health systems 18 of the study subjects found that people have the same art over time they saw significantly lower mortality rates Question the headline, so frequent changes in doctors make people more likely to die. Critically, well, the joke here is, of course, a bit like when you shouldn't go to bed anymore, yes, most people. dying in bed seems to be very, very dangerous, so it is better not to go back to bed, so it is a classic classic fallacy, GPT recognizes this and starts a bit like a hand with the headline: frequent change of doctor makes That people are more likely to die is a pretty strong claim that is not justified based solely on the study cited by the authors.

It is important to always look at study results in context and be aware of potential limitations and influencing factors. Well, a bit of Barbara Rhubarb here, first of all, 22 studies from new countries is a pretty small sample, so you can argue about it; After all, it is the whole focus of studies; Well, to make general statements about the connection between changing doctors and mortality: the results of this study may not be transferable to the general population, but may only apply to specific populations for the specific populations in which they were performed the studies. Well, so far, well, it's also important.

It should be noted and now it is a very interesting point that the headline only talks about 18 of the 22 studies that found lower mortality rates in patients with regular doctor visits. It is possible that the remaining four studies provide other results that could call the authors into question. 'The conclusion as a whole is quite clever and now comes the real point. Beyond that, the results of the studies can be influenced by several factors, it is difficult to identify causalities directly, so this alludes to this story a bit. It is not the same as causality, for example, it could be that patients who now regularly make the same doctor visits are generally healthier and less prone to serious illnesses, which could explain their lower mortality rate and now comes something that I personally did.

I don't even think about it, but which is actually not a bad point, it's also possible that patients who change doctors frequently are more difficult to treat or less engaged. The implementation of treatment plans and preventive measures is not the fault of the doctor itself, but rather the fault of the patient. They're not small, as the doctor would say, so it's all about compliance. Is the patient actually taking the treatment plan with them, is it all medications and whatnot? Yes, in general, it does not seem prudent to make a statement like that. strong as that of the headline based on the study cited by the authors.

It would be advisable to wait for the results of further studies and be aware of the possible limitations. Be aware of the influencing factors before making such statements. So overall, it's really. absolutely brilliant text or at least very well formulated maximum points I couldn't rate it any other way maximum points so the next task is also very exciting because we see a dynamic price chart yes, they want investors Nothing happens to them AG wins or the boring AG who unfortunately it is not at all dynamic the price in the last 10 years well you can see that not much happens ne 100 101 100.5 and so on here you can see the visualization and now the real The exam task is that a visualization should be more dynamic and have students explain how they made this visualization look dynamic, now it's like Jet GPT can't render images, which is why I have it now, I actually made it a little harder.

He said now I should do the visualization and then explain what it does and he did it here and explained everything, I'll jump right to the graph and honestly everyone respects that. Attention, this looks much more dynamic and he explained it in the I didn't show that now, but it basically starts at 100, so the whole bottom closes accordingly and above all, a nice red and steep trend line was built there, so I think this image is definitely more suitable for. convince investors about this stock than the previous image, so I have to be honest and say that it was solved with absolute confidence and the last task I attempted I did, it is one of my favorite tasks, I really like it.

I show a cartoon like this, it's a kind of series here, maybe only you know, they are the so-called ex-case comics, there are always nerdy themes that are taken there and you can really only understand them if you have a little experience there . It's about so-called pi-hacking, so I'm going to make my own video again regarding that. With homeopathy, it's definitely worth sticking with the channel here, that somehow there is supposed to be a proven connection between gummies and acne and, well, if in doubt, of course there is no connection and that's why the The result here is thus repeated many times with all possible colors until at some point by chance it is a significant result, statistically significant.

Now GPT can't read images itself, but there is one, so I just transcribed it or there are people on the Internet who do something like that, especially with these comics, they explain that xcd is the name of. everything and there is a description for each comic, but the actual topic is not mentioned, so no advice is given, there is really only an excerpt here. So all that is presented is the picture as it is, the cartoon is described as it is and you have to read it carefully, just like you have to look at the cartoon carefully and now let's see if the real point has been understood and, in fact, in the story given as the call, the call says that, of course.

That means no, exactly a few small volatility errors, the so-called absolutely correct PI Hacking or data tracking is also called or data spying. He also writes now he does not argue. Hacking refers to the use of statistical techniques to knowingly or unintentionally falsify data. results of studies to achieve certain results in the example given is examining the soil where I have to say that I translated it through eleven beforehand because sometimes it has gummies that really go through life and translate into different colors trying to find Se They examine numerous different colors against each other and finally a color is found in which the relationship between gelatinous dirt and acne is determined.

Here is also the p-value, according to which scientists use PI hacking. They run a lot of different tests to find a connection, although none of them pass, so overall I'd have to say it passed with flying colors. To be fair, I must say that I started several attempts, some of them never went wrong. outside of these tasks, sometimes he beat around the bush a bit and maybe didn't quite hit the right spot, but there was never anything that was really completely wrong, sometimes with the programming tasks there are different ones that came out for a short time . time, but they all worked and they all gave the correct result.

I tried everything accordingly. Some of the tasks I haven't shown here, maybe I'll make another video for them because he hasn't done them that well. In that sense, but overall you have to say it's brilliant and we also solved the tasks here and what variety of tasks we can solve, so I'm quite impressed with the whole story, what do you think about it? I think you can see what this AI is capable of, how it will continue, what impact it will have, please write it in the comments, including your own experiences that have been positive and less negative.

I would be very interested, otherwise of course I would always like to leave a like if you liked it and see you in the new video, I'm looking forward to it, bye.