[Music] Yes. hello. My name is Doljong , a civil engineer who codes . Now, today we will find out how well artificial intelligence recognizes formulas. We will first analyze just two, Chatzipt and Gemini . The things we use a lot in engineering work are ultimately formulas. In a way, unlike general work, what else is there after formulas? Should I call it a diagram or something like that? A diagram that explains something, like how a load acts or something like that . Then, there are a lot of tables and things like this that explain each other or, for example, diagrams. What I 'm testing now in this video is a general single-span view. So, when a load P is applied to a two-span beam, a reaction force of R will be generated. RA RB RC Then how much momentum will there be here? like this. This diagram and here is how much RA is. This and this are related to each other . Now, these things are different from general fields, such as text only or formulas only. That's right, that's why I'm going to test these things today to see how well artificial intelligence recognizes them . Now, since this may take some time, I prepared this formula in advance. The reason why I chose this is, first of all, the short span is the knowledge that the artificial intelligence itself possesses. He learned a lot of documents from the Internet. So, I thought there would be a lot of cantilevers for single-span pipes , so I deliberately chose a two-span continuous beam. So I scanned this at the official house and made an image like this. So, first upload an image and use chatGPT , so this is only possible in the paid version. Uploading images and things like that. So, I uploaded the image and I said, ‘Tell me about this image.’ And he said, this image is a diagram showing the loads and reactions of a two-span beam, a common thing in structural engineering . They say he made this diagram in English, translated it back into Korean, and so on. It's a reaction force, but it's not a reaction force. In fact, it seems to be used to calculate reaction force, shear force, bending moment, and deflection. The beams are supported at three points, each labeled ABC, with A and C seen as fixed points and B as a movable point. That kind of thing can't be expressed. Now, if it's fixed or pinned, we may have pins, rollers, etc., but it's not actually a movable point ... It doesn't matter. The length of the guard is divided into three equal parts, and the length of the three-part beam... is actually divided into two parts. That’s right, one span, two span. Since the length of each part is L , the total length of the beam is 3L. So, I think he's a bit like that right now. The fact that it is artificial intelligence means that it has the property of trying to answer certain questions as best as possible. So to speak, I'm just pretending to be better off? So, the length of each part of this is L, and since the three parts were wrong from here, 3 * L is now 3L. In the picture here, this is... Oh, right, but I think it may have been because of this. There was something called 1 here. So I think you may have been confused because of that, but anyway, this is now wrong . P represents the external load, and this load is directly above point D. Oh, he may have recognized this as a point. That's right. Actually , it's not the point, it's the center of this first span. The load acts on the center . It's acting just above D. D is located between points AB and the distance from D to A is 1/2. This is correct again. That's right. The distance from D to B is also 1/2. Right. half. It's a reaction force. The reaction force represents the vertical reaction force at each support point and is denoted as RA RB RC. The values are 13/32 respectively. This is what we now want to calculate with a formula. 13/32, 11/16, -3/32. Right. RA RB RC This is now correct. He eventually recognized this, this image expressed as a formula, and found the formula. That's right. This is an explanation of the diagram because he thought, 'This is what it's about,' and the formula is explained from here. These values are calculated using structural equilibrium conditions. Wow, I even know these difficult things. Additionally, the shear force Qbd and moment MB MD are given, and their values are 19/32, respectively. This is it. I think I'm recognizing this accurately. 13/16 Like this. MB It will now be a moment here. If it goes like this, a submention will appear here. I think it might be like this here. It will look like this, right. So the moment here, the positive moment here is the maximum. Here, parent comments are the maximum. Since there is no load here, it might come out like this . Is this going to happen? ruler. There is no load here. It's only here. Then this place explodes. Perhaps the negative negative reaction here means negative reaction force. I don't know what the absolute value of minus is like this. Anyway, it will come out like this. So , so. Well, finally, the deflection at point D is given as 23 PL to the power of 1536 EA minutes. Here, E is the elastic modulus and I is the second moment of inertia. Here, there is no word on E being the modulus of elasticity and there is no word on what I is. That's right. But this is just a picture. Now, I don't know if you recognized this sign in this diagram or not, but maybe that's how you figured it out because R is usually used as a reaction force. Anyway, he understands this accurately. 23 PL3 wins out of 1536EI . We know that it is the second moment of inertia. So structural engineers use this information to evaluate the safety and performance of structures. So, I calculated this. Calculate RA when p is 10. Of course you can get it. When p is 10, calculate RA. It's easy to calculate. Well, just divide 130 by 32. I calculated it and it was ten times more. That's right. like this. Now, chatGPT is a code interpreter function. When he does a calculation, he does not do the calculation himself, but codes it first . I code it and run it. It is quite accurate. That is, if you code properly, you will get accurate results. So, I think you can think of it this way. Oh, sometimes he doesn't calculate like this anymore and just calculates himself. You can think of it as doing mental calculations. I do mental calculations using my knowledge in my head . There are cases where it is done that way, and calculating with coding like this is similar to using an electronic calculator. So , it means that there is a difference like the concept of using a mental calculation and an electronic calculator, and chatGPT always codes it himself when he has to do calculations like this, even if it is very simple . If you code it and calculate it, RA will come out accurately. Now, I asked him to calculate the MB. What was MB? It's the moment here. The parent moment. That's right. MB parent comments will appear. MB will make a parenting comment. here When I asked him to calculate the negative moment and calculate the moment at the midpoint of a certain point, what he did was that according to the information given, the MB value is -3PL/32 . But I did it like this on purpose. I haven't tried giving l. How does he do it? Since the value of length L is not known , the calculation here assumes l to be a unit length . In this case, when P is 10, the MB value is calculated like this. Well, it would be 30 divided by 32. That's 30 divided by 32. Even if you do it in your head, you can multiply it by 10 times. Then 0.9 would be correct. This is also a simple calculation, but l is assumed to be 1. This is how he does it himself. Oh, I just assumed this to be 1, so if I want to calculate it properly later, I have to set it to a different value. The actual length needs to be entered, and things like that are now being coded. Now, let's do the MD calculation. Here is a positive moment. Now, when a load acts like this, what I'm most curious about is the moment and reaction force. So, when I asked him to calculate MD, he said that he would assume that the unit length is ah, and when p is 10, he said earlier that p was 10. So he remembers that p is 10. So if P is 10, we substitute 13/64 p 10 and it works out to 2.0315. I think it's pretty great to do it this way . Now, in the process of explaining, I did make a few mistakes. Well, it's 3L. Maybe it's because it has ABCD, so it seems like it only recognized something like that, and there's no information about Jigan here. Is it 2 spans or 3 spans? That's right. But I don't know if it was recognized in the picture or here, and it's a little unfortunate that I mistook it for a 3-kan, but other than that, it was recognized almost accurately. Now, Gemini. I 'll give Gemini a try . I watched this fun movie now, and there were a lot of parts that left me a little disappointed. There have been so many. this. When something doesn't work out, I usually try again and again. Now what I meant was that the first thing I tried was to give the same image and now this image. Now, explain this image. I wrote the exact same prompt. And then. Oh, this is suddenly a diagram of an electric circuit. He knew it was a diagram . Think of it as a diagram, and it shows the electrical components and how they are connected. oh. Okay, so now there's R here, D, M, etc. But since he knows that, he just says that C is the capacity, and it 's probably a formula in the biography. Used in electrical engineering. I is the current and P is. Perhaps he recognized the picture here , or what he was doing with these letters . Now, the original LLM is about finding something with a certain probability using what information I gave you . But now he has gone this way. So, the resistor R is a reactor. It is true, but it is not that resistance. It's called reaction force. It's a force, but it's resisting now. Well, anyway, resistor. In electrical engineering, R would probably be expressed as a resistor. Anyway, I think I found the correct value here . Was it 13/32 or 11/16? Yes. So, that formula was recognized correctly, but it seems like he chose the wrong field, so he always searches the Internet like this now . So, the advantage of Google is that it has its own search engine , so I think it does something by combining the two. However, if there is an image I gave you, I have to make a decision based on that image , but it seems like he is doing his own search and has found reference materials like this . So, if you go into this reference material, there are six recommended circuit diagram drawing programs, such as EDraw, etc. , but there is no mention of R or anything here either . But I don't know why I found this, but I think I decided that this image here was similar to the image I uploaded. In fact, Google also has an image search function. Well, we don't think it's very similar, but I guess he thought it was similar, so I gave it to him as a reference, so I told him to give me the formula to find RA, but now the problem is. Because here, RA, RB, and RC appear like this in the front , and this is correct now. Now, when I asked what the formula for finding RA was, he gave me the wrong RA. It seems like V is probably a voltage or something. That's right. I think it's an electric formula. That's right. You learned physics in high school. Right, like this. That's right. Something weird, Ohm's law or something like that. That's right, now we've thrown it in the wrong direction, which represents the relationship between electrical resistance and things like that. This is what it is now, so in the end, this could be seen as a kind of hallucination . In the process of finding something of my own and trying to answer, I went astray. So I say this, Google is Google, but Google shouldn’t do that. That's right. So I tried this again. Now, please explain this image . I did it, but it was the same. It was the same. So I 'm not at all an electric circuit this time . Let's review it again. I did this, but I misunderstood. I checked the image again. It is an RLC parallel circuit, not an RLC series circuit. Steady State Analysis Steady State Well, it was a circuit diagram showing steady state analysis. So, I'm talking nonsense again. So here, RA, RB, and RC are correct. But now, strange things are appearing, angular velocity, current, and angular velocity are irrelevant. This is kinematics, but it's weird... Here again, it shows some kind of reference, so I told it outright. This image is about the formula for calculating the reaction force and member force of a two-span continuous beam. I checked again and found that it shows the formula for calculating the power and member force of a two-span continuous beam. And then the halfway point came out like this. Oh, has he come to his senses now? I did it like this, and now the point reaction forces are RA, RB, and RC. But suddenly there are only two, and this is also hallucination. When I asked about the treason and absence of continuous beams, he said, "Oh , it's something from structural mechanics ," and instead of using the image I just gave him, he answered because he had to use his imagination and answer first. So, I just asked outright what is the formula for finding RA in this image, and I got a wrong answer. So, I don't know what this is, but it doesn't seem like it's a report from the Tang Dynasty, and it's a strange formula. Once he goes down the wrong path, he doesn't come back. So I started chatting again and it was like this. In the end, I couldn't come and it just ended. So, I'm still Google, so this can't be happening. So this time, I misrecognized the diagram next to this and took a side road. Out of curiosity, I just gave it the formula like this. So, during the image search process, I focused on the diagram. Fun age. So, looking at the diagram, it looks like a circuit to me. And now I'm starting to go down the wrong path, so it's good that I create an image with just the formula. Tell me about the formula in this image . This time, I will just focus on the formula and tell you about the formula . Then, I think it came out pretty much the same. I don't know what the intersection torsion theory is, but this also seems like hallucination . hmm. So RA, RB, and RC are the reaction forces of each beam in their own way. It is not the reaction force of each beam , but the reaction force of each point. This was also wrong, but I roughly knew that Q was shear force and M was moment. P is not a point load acting on each beam. It's a point load acting at location D, and I wasn't even here. Oh, there is I. Here below. Because EI is needed when finding deflection. I is the length of the beam and L is the length of each beam. It's just one span. The term "each beam" is a bit strange now, but E is the elastic modulus of the beam and I is the secondary moment of section of the beam. Delta D is not the displacement of the end of the beam. In fact, Delta D here. D. That's right. Since there is no diagram here, I don't know where it is, and it is expressed as an end, but this is probably because delta is usually obtained from a cantilever or something like that. Yes, cantilever, this is also a type of hallucination . It's like this, but now you usually give P and get delta, right? Delta probably has a lot of things like this. When learning on the internet, this is probably the most common way to find deflection at the end of a cantilever. Probabilistically. I think maybe I was mistaken because of that . So, I am giving a detailed explanation like this . If you read this, there may be some mistakes, but I don't think it will be a big problem, because the important thing is that you recognize the formula correctly . So, I did the calculation. Just like before. Calculate MD when P is 10 . Now, you mentioned earlier that MD requires knowing l. So, I deliberately didn't tell them L and just calculated P only when it was 10. As mentioned earlier, chatGPT requires knowledge of L. But you didn't give me L. But then, I will calculate L per unit length. I told him I would calculate it as 1 , but what did he do now ? ruler. Using the formula in the image, we calculate MD when P is 10 : So, it ended up being 13/64. Is that right? MD, MD 13/64, right? I said it was 13/64PL, and then substituted 10 for p. We substituted 10 into P, and now L remains the same, so the problem here is that if you suddenly take 13/64 and multiply it by 10 , then 130 divided by 64 should remain the same. That's right. This is the part. Since we multiplied by 10, 130 divided by 64 should remain the same. That's right. Here, L should be like this. Right, so if you do 130 divided by 64, you get the answer. That's right. So, this one is MD, this one is MB, and 2. How much will it cost? That's right. If you divide 130 by 64, you should get 2. That's how much. But it's now 0.81. You calculated it wrong here . That's right. I don't know why it came out as 16. The wrong value came out now. Why did 16 come out? No matter what I calculate, I don't get 16. Well, even if I miscalculated this, the relationship 13 between 64 and 10 has not changed. Can 16 come from the relationship between 64 and 13? I don't think I'll get 16 for some reason . Well, anyway, I miscalculated the wrong answer. So, since L is no longer given to him , his L is 0.81 times greater. So just calculate this value and it is several times L. I calculated it like this, but it's 2. It's wrong, and if the length of the beam is 1, since L is not given right now, I'll calculate it with L as a variable. So, the calculation is that if you substitute 1 for L , this is this. But it's not a meter. I didn't bother to include the unit. Just because he is 1m doesn’t mean he is m, right? Now, this thing doesn't have a unit, this calculation formula doesn't have a unit, so there's no way he could add a unit and have an odd moment of less than a meter. That's right. I made a mistake answering like that . So, calculate MD when L is 1. Just because he had L, I thought he might have miscalculated by giving L as a variable, so I just gave him L as 1. Clearly. It's the same. If you calculate 1 from the beginning, this is not L, so L disappears here, but the calculation is the same. I don't know why I got 16, so I got it wrong again. So, I thought the calculation process was wrong, so I tried to calculate it again, and they said there was a mistake in the previous calculation process, but now 64 is left. as it is. Oh, but where did the 10 go? This is it. So where did I lose the 10 ? I only calculated 64. 14/64 only. 0.2 So, if you multiply it ten times, it is correct. That's right. 2.031 If you multiply it ten times, it is correct. So I just blew it anyway and gave it a 10. As he said earlier, mistakes like this are bound to happen. It must be a type of hallucination. In the process of calculating, just like a person does mental calculations, this is virtually the same as doing mental calculations, right? That's why chatGPT is so powerful now because it's a code interpreter. Code interpreter. Even when calculating something as simple as this, he uses his own coding to calculate it. That function is called a code interpreter, and the code interpreter function is actually a very powerful function in chatGPT. Now, gemini is a bit unreliable, and it may show different performance every time you do something. But in this test, gemini didn't perform well at all . So, this test was to give a formula containing a diagram and calculate the formula , but since gemini gave an odd answer in a very one-sided way, of course, depending on what kind of picture we upload, he may do well. I think I might not be able to . So, I deliberately uploaded something that is not likely to be found on the Internet a lot, and was about finding a certain moment of a two-span continuous beam. So, it seems that the gemini has the tendency to try to play games with the knowledge it has , and this chatGPT is trying to solve it as much as possible within the content we uploaded. So it appears that there is some difference in performance. About the image I posted recently . Now, the difference between chatGPT and gemini is that it is calculated using a code interpreter, which I told you about earlier, and then look at this. Let's change the formula to the style shown in the textbooks we use . But gemini is not like that, if you look at the solution process earlier. If it is 13/64 like this, the denominator is now at the bottom. The molecule goes up and this is how you can see the formula clearly. He's like this now. So this is just a coding style logo. When you run chatGPT, you can see the process of unpacking it like this, unpacking it like this, writing it, and changing it like this. In its own way, it's not a huge feature, but it was decided that it would be easier for users to recognize the formulas as if they were in a textbook, so they applied it to the UI, so to speak. There is such a difference. So, today, I gave an image containing a diagram and a formula to chatGPT and gemini to calculate it . The conclusion was that chatGPT was good at recognizing and calculating accurately, but gemini was strange at recognizing it and only gave it a formula and made the calculation. I ordered it, but the calculation details were also wrong. I think the difference and calculation difference is due to a difference in the code interpreter. Next, the difference in recognition seems to be a difference in their own functions, and isn't the reason why Google is not good at picture recognition or anything like that because they clumsily combined their strengths, such as search and such? Well, that's what I think. Now, let ’s end today’s video here. thank you
