Transcription
Very nice. But uh, so we have a variety of registrations. This is now practically the second webinar that we are doing from Side on the topic of artificial intelligence. Uh, generally without focusing on time. Uh, now specifically on the topic of ChatGPT. What can you actually do with it? What can you actually make with it? And, um, yes, the resonance has been extremely, uh, large, even the first time. Now the second, uh, significantly higher again, and, uh, therefore we are naturally pleased to, uh, yes, look together with you today at what ChatGPT can do in our industry. Uh, you have the opportunity, I'll start off a little bit, I would say, uh, you have the opportunity to ask questions during the webinar. Now we have a really very high number of, uh, participants. Uh, that means it will probably be difficult for me to see, uh, which questions are coming in. I will, however, keep this tab, uh, open on the side, and if it's suitable to answer something within a topic, then I will do so. Otherwise, I'll see if we have a little time at the end, and then, uh, yes, I hope I can answer everything for you. If not, then you are welcome to contact me anytime via mail, LinkedIn, however you prefer, uh, and ask me questions about it. I find the topic super exciting myself, and therefore, uh, I am always ready, uh, yes, to support accordingly and to give my opinion on it. Uh, the webinar will be recorded and uploaded, uh, like the last webinar on YouTube. Uh, that means you always have the opportunity to watch it again. That doesn't mean, uh, that you should be less attentive here, uh, but, uh, you can always look again if you want to briefly review certain things that we are discussing today. So, now we are already at 11:03. Uh, and I would say we start and, uh, tackle the topic of artificial intelligence, or rather, especially ChatGPT in the real estate world. Uh, very quickly about me. I am Nilas Möllenkamp. I have been part of the real estate industry for 7 years. Previously, I was a project manager for digital strategy at BNP Paribas Real Estate. So I looked at, uh, yes, uh, Proptech startups, looked at what we could use internally, what we could incorporate into our model, and since 2022, I have been, uh, with Side, the German AI startup of the year, and have gotten very deeply into this whole AI area, I would say. Uh, I am now a lecturer and speaker at the ADI CIA Bund deutscher Baumeister, uh, iWaps lecturer I am also, and in addition to my work, I provide consulting services on artificial intelligence, on digitalization in our industry. Uh, I am basically a digital person, uh, to put it that way, but, uh, I can't code, I can't develop, I can't program. Nevertheless, I can do great things with AI, I can automate many processes. Uh, and a part of that is indeed ChatGPT. For many of you, probably the start into artificial intelligence. For me, it honestly was too, and therefore we are looking at that today. So, this is a part of what I actually do daily, which I want to bring closer to you, because believe me, you can make a lot more efficient and, uh, yes, automate a lot by yourself with just ChatGPT. What will this webinar be about? Uh, I will show you all the functionalities that ChatGPT has, but always with a practical reference. It is very important to me, uh, not just to tell you, yes, the future of ChatGPT or the future of AI will look bleak. We could also use it in the real estate industry, and we could use it to, uh, do real estate valuations, to, uh, find out certain things. And these are the tools for it. No, uh, we will go directly into the, uh, application. You see it's already open here, uh, and, uh, we will look at all the corresponding tools. Uh, test them together. I have prepared corresponding prompts. Sometimes it goes faster, sometimes it goes slower, but I have also prepared the results, uh, if it is a bit, uh, a bit slower. Uh, and exactly that's what we'll do now. Exactly that's what we'll look at. Practice-oriented. That is always the important thing when we do these, uh, webinars from Side, where we present, uh, artificial intelligence in the, uh, broadest sense. Uh, at the very beginning, a very brief note. There is, uh, I've included a screenshot from Google, when I enter ChatGPT, uh, which many of you do, because of course nobody knows a domain, uh, directly and, and knows. If you enter ChatGPT, then, uh, of the first five results you see here, at least for me, uh, two are fake, maybe a bit the wrong word, but you definitely have two, uh, results that are actually not ChatGPT. And this, this first result, for example, Chat chatopenai.de. Someone has simply used the API, the interface of ChatGPT, and said, I'm building this on my own website. Here we have chatGPT jetgpt. Yes, that's right. Chch. And these are, in principle, these are representations of what you actually know from ChatGPT. If you want to use ChatGPT, that ChatGPT, then go to chatg.com. Uh, I know it's possible that you might have gotten confused recently. That's not bad. But, uh, you shouldn't do that anymore, because, uh, this third-party software, more or less, it accesses ChatGPT, but, uh, I don't think it's entirely clear what they actually do with the data you enter there. That means they could be intercepted and then used for further work. You don't want that. Perhaps for those of you, uh, who are considering whether to introduce ChatGPT in the company, uh, in the company. Uh, I hear very, very often, uh, everyone uses it anyway, even though it might not even be approved. Uh, if that's okay for you, then yes, fine, but actually ensure that the real ChatGPT is used and not these imitations of it. Therefore, I would rather recommend, ensure that you really use and integrate ChatGPT and not these, uh, I've called them questionable third-party software now. Yes, use it, but pay attention. Chatgbt.com, that is the correct domain. So. Uh, we'll go directly into ChatGPT. Uh, so this webinar is for beginners, for novices who perhaps don't know ChatGPT yet, haven't used it yet, but it will, I'd say, become more and more extensive and far-reaching in its functionalities. ChatGPT is in the registered version, this is it. Uh, the question also just came up, uh, what we are referring to today. Uh, I am referring to, or with the tools and, uh, uh, products from ChatGPT that we are showing here. Uh, I am referring to tools that are usable for free on the one hand. So, uh, simply asking a question, as it stands here, for example, is usable for free. Uh, but very many, uh, very many tools that I will show you shortly are included in the paid version, which costs €19 per month. Uh, then included. But these €19, I can promise you, are well invested. Chatt, this is the interface, and in principle, it works like this, I enter something here. Uh, for example, create an appealing description for a single-family house with four bedrooms, two bathrooms, and a large garden. Uh, and click on Let's go, uh, click on Enter, and then ChatGPT does that. Uh, it can create texts, it can create emails, but it can also research things, it can inform me about things that I want to know, be it historical, political, whatever, everything that you would normally find out via Google, you will also find there. But, uh, there is a difference. ChatGPT needs a question from you, so a kind of task to be solved, and which is solved. And, uh, if you enter this task here, then it is a different input than what you normally have on Google. If you search for a description text for a family house on Google, then you will find instructions on how to do it best. Here you can simply say exactly what you want. The whole thing should, we'll get to that shortly, uh, the prompt here, so this input that you make here, should actually be better than what I am now just as an example, uh, these details, uh, that you enter, the solution, ChatGPT gives it to you immediately, and it only takes a few seconds. You could, uh, alternatively also say, uh, research everything about the State Building Code NRW, I don't know. Uh, by the way, it doesn't matter at all whether you spell it correctly, uh, or whether you make mistakes. Uh, ChatGPT recognizes that. That's no problem at all. Uh, then, uh, there are other things you can research with it that you might otherwise enter on Google. Research the five most important and largest construction projects in Frankfurt am Main. You just saw briefly. Uh, ChatGPT researches that, uh, because it has, on the one hand, a knowledge base of a lot of knowledge that it has acquired from the internet over the past years, uh, and also from various other sources, and on the other hand, it can also search the internet live. And exactly the latter has just happened. GPT has become very fast at this and can now, uh, give you an overview, uh, of the properties that are relevant in this question as an answer, yes, how you enter things here and what you do with it, this so-called prompting, uh, on the one hand, it needs to be learned, that, uh, that, uh, I have to say. On the other hand, uh, too much emphasis has been placed on so-called prompting in the last, I would say, year and a half. Uh, on the one hand, you need to know how ChatGPT thinks and works. On the other hand, uh, you probably won't need to know that in the future, because ChatGPT will be developed so far that you only need to enter a certain, yes, a certain area or a certain question. ChatGPT will recognize the context of that and recognize everything around it, uh, and will practically build this prompt itself until it is completely 100 percent ready. It's already a bit like that now, uh, I would absolutely recommend that you, uh, yes, consider and master these tips for, uh, these tips for prompting. First, uh, make sure you formulate positively. If you say, uh, uh, don't do this and that, then, uh, it's a bit like with a small child. Uh, there is a higher probability that exactly what you don't want will be output, done, or, uh, what, what then is the solution. Uh, just because you said it. So, there are errors with ChatGPT, that might be the case in 2% of cases. So, try to formulate positively to fix that. Then, the so-called role prompting is important, simulating an experienced perspective. If you tell ChatGPT, uh, you are, uh, in the example we had earlier, it's probably not that relevant, but for the description of the single-family house, it is. Uh, you are an experienced real estate agent, uh, and, uh, create, uh, with a lot of passion and experience, descriptions for single-family houses. Uh, and now give me a description for this house. This role prompting helps ChatGPT very, very strongly to recognize its own context and to see what it is working with and what it is looking at. Then, I'll take this out for a moment. This is, uh, I'll attach this in a moment, but then it's always useful to break down tasks. Of course, you can say, uh, create the complete plan for the construction of a single-family house. What do I need to consider in every detail, etc. But, uh, that is a complex task that requires a lot of thinking. And there, I would recommend you to first, uh, break it down and create individual tasks, and say, I want to know what I need to consider, and then enter each individual point, one by one, into a prompt and into ChatGPT, because then better results will come from it. The coarser it is, the coarser the overview we get as a solution to the task. Write clearly and precisely, always provide context, provide all relevant information, and, if possible, provide examples. So, if you want data to be output in a certain way or in a certain format, uh, then say how that format should be structured. And, uh, that's always, uh, yes, I always like to put it at the end. Uh, build pressure on the AI. If you write at the end, my job depends on it, my life depends on it, or something else. Then the results are demonstrably better. There are studies on this. It sounds contradictory, it sounds really silly, uh, but it's true. So, if you write a good prompt, then at the end you write, my life depends on you solving this correctly. And then the results are better. If you consider all of this here, then you will achieve very good results with ChatGPT. ChatGPT is not Google. So, if you previously entered something on Google with one word, then you would now enter the whole question, perhaps even longer. This one minute, perhaps, that you spend on prompting, you probably would have searched for 10 minutes on Google and gone through various websites, had to read things that are not relevant to you at all. Don't use ChatGPT like Google. Uh, learn a bit, and then, uh, it is, uh, personally, I probably use ChatGPT four times as often a day as Google, rather more. Yes, and then you get exactly the information you need. And, uh, that is very important. ChatGPT and AI create the prompts for you. If you simply want to be very general and say, yes, I want an overview of certain properties in my city, then, uh, tell the tool, I want you to create the prompt for me. Then it creates the prompt, and you copy this prompt, enter it, and then you get the right solution. In fact, it is so, and it works that simply. So, prompting is very important for you at first. I also have an example of a promising prompt. Uh, you are an experienced real estate agent specializing in valuation in North Rhine-Westphalia. My reputation depends on me delivering a concise market value assessment for an investor within 60 minutes for a 58 m² existing property in Bonn. Research current comparable prices. Here I have omitted something. Uh, because researching current comparable prices is one topic, and I usually include other topics here. Yes, it could do all of that, but if you really want to be sure that what comes out is good, then do it like this first, and then give me a bulleted list of the five decisive influencing factors, check the information carefully, etc., and then you will get a very, very good result from it. That means it takes a bit of learning. Yes. Uh, now we'll go specifically into ChatGPT. You may have heard in the last, uh, weeks and months, yes, now ChatGPT 5 is out, before it was 4, the always very strange, very strange designation there. Until a few weeks ago, you could still switch which ChatGPT model you wanted to use. You can still do that here, you can also display an older model 4. Uh, I would recommend you to always leave it on Auto, uh. If you have it on Auto, then, uh, ChatGPT itself decides which, yes, which tool, which tool from ChatGPT it should use. There are different ones. One gives you very fast answers immediately, and the other thinks for a relatively long time, uh, to give you better answers. I can also show you that as an, uh, example. If I say, for example, uh, research very precisely and in depth, what are realistic square meter prices for a condominium in Berlin Charlottenburg in a renovated condition in August 25. Uh, if I actually want to sell such an apartment, provide several sources, show a table of different square meter prices, and justify your final answer. Also consider differences, uh, that the apartments can have regarding features, etc. Now you see here longer thinking for a better answer. I could skip it now, then the answer would be worse again. So, it needs some time here. And if I now, uh, shorten the prompt and, uh, honestly, also have to assume that the answers are not quite as good, I'll do it again and say, what does an apartment in Berlin Charlottenburg cost? And here it should, uh, well, it's still waiting here, uh, but here it should now give the correct, uh, solution and the correct answer very quickly. Uh, online search is running. So, here it doesn't say this "We are waiting" or "We are thinking again," but something will come out here, also very extensive. Yes. Uh, with the corresponding sources, you can always access them directly here to check again how good ChatGPT is with what it is outputting here. Uh, but we are not done with the other search yet. So, ChatGPT is still taking its time to find the corresponding information. I told you at the beginning, but I have also always prepared the solutions. I accidentally told ChatGPT that I would learn Arabic and asked it to include Arabic in it. It hasn't unlearned that to this day. Uh, I need to do that again. But you see, the answer to this question that I included here took 2 minutes and 36 seconds. The result was delivered in 10 seconds. Yes. Yes, good. Uh, and oh yes, result in 10 seconds means in this case that I have the overview in 10 seconds. Uh, but you also see the answer here is much, uh, much more extensive. Current price indicators with the corresponding sources we have, uh, with the corresponding square meter prices, uh, that have been found, from when is the data, from when is the information, justification and classification. Uh, that means ChatGPT is really working on this for a very, very long time and also gives very, very good results, even in Arabic, if you want. Yes. It was a bad idea to tell ChatGPT that I would like that. Uh, exactly, these are, in principle, two, uh, different or several different modes, it actually says several different modes. Uh, generally, we have also already researched information and obtained good information. Uh, it is important that you always consider, uh, or definitely read it, of course, and then think, uh, can this even be true? Does it fit? Uh, or does it perhaps not fit? Because, uh, unfortunately, ChatGPT still makes mistakes. And, uh, this is now a screenshot that I took a few months ago. In 2024, over 20% of large architectural firms integrated AI tools like ChatGPT for their internal processes. So, this is a screenshot from ChatGPT. Uh, that was an answer from ChatGPT. It said, give me a source, give me a statement, no, give me a source for the statement below. And the answer was, uh, yes, uh, it could not be substantiated by specific sources. Uh, then I think to myself, great, thank you. So, this is this hallucination. Sometimes, if ChatGPT doesn't know anything or, uh, for some reason cannot find the answer, ChatGPT might tell you something that has no basis whatsoever. It makes it up. You can curb this to some extent, actually curb it very strongly, by always saying, research, uh, and then give me the solution with sources. That works. Now, yesterday, uh, in preparation for this presentation, I wanted to know how long I have actually been part of the real estate industry. Is this 7 years that I showed you at the beginning correct? Uh, and there I had to argue with ChatGPT for a long time. I told it to research it itself. And in the end, it told me that I have been active in the field for over 14 years, since June 2011. That is not the case. Uh, I said, that's wrong. I told it, uh, where did you get that from? Among other things, it said from my Xing profile. However, the Xing profile states that this internship only lasted 6 months, not 7 years and 7 months. So, even if you point it out, look, you made a mistake or you might have made a mistake, please check it again. It is very possible that incorrect information will come out, and that is very important. You must know that. Read the information. If something seems strange, then, uh, ask about it, and if in doubt, yes, click on the source and look at the whole thing. That is very important. Yes. So, and now we come to a topic that I was asked about in advance of this, uh, of this webinar. Uh, what is always the first question after 10 minutes when talking about AI and ChatGPT. Uh, and honestly, not my favorite topic. Uh, there is a lot of, yes, a lot of thinking about what ChatGPT does with your data, whether it is GDPR compliant or not. Uh, most people say: "No, we absolutely cannot use it, etc." Uh, first, the first point is always, it is said, uh, yes, but if I upload a rental agreement to ChatGPT, then the AI will be trained with it, and then the rental agreement will eventually be presented to someone else or displayed. Uh, firstly, the AI only trains with your input if you do not disable "Improve the model for everyone" in the settings under data control. Then that does not happen. Fundamentally, it does not. On the other hand, uh, if you upload a rental agreement or upload anything that, uh, contains essential data. Uh, so, if I think about what kind of nonsense I sometimes enter into ChatGPT, if ChatGPT were to train with it, then after half a year, the company would have a significantly, significantly worse tool. The information you enter is not necessarily what ChatGPT is trained with. It is more the requests, how you send something, how you, uh, how you communicate something. ChatGPT is trained with that, if you don't disable it. Settings, data control, improve the model for everyone. So, that's the first point. The second point, uh, then there is the question, where is the data stored? And yes, everything that I enter here, uh, or sometimes upload, uh, and the corresponding results, they have to be, they are displayed on the left side in my chats, they have to be stored somewhere, and that, uh, actually takes place on servers in the USA, at least if we are in the free version or in the Pro version, then this storage of data takes place on servers in the USA for 30 days. That would not be data protection compliant or not GDPR compliant, as far as I am correctly informed. But you can apply for an enterprise license, uh, for, of course, several people in your company. The servers are then located, or you can then choose processing in European data centers. You can then choose that the servers should be in the EU, so that you have this problem less. Of course, you still have to inform your customers that, uh, that you are using ChatGPT and that data could go there. Uh, but you can also, of course, refrain from simply adding personal data. Yes, fundamentally, many companies also work with the fact that data is immediately pulled into their own systems via the interface, via the so-called API. That is also possible. Uh, that is also data protection compliant, as far as I understand it. I am not giving legal advice here. Uh, but that is also data protection compliant and works. Uh, and what else, or what else is there and what else is possible, there are always alternatives. The problem you have, that you say, I cannot upload data there because it is not data protection compliant, many, many others have that too. And if many others have this problem, then there will be companies that take care of solving this problem. And these companies want to make money. These include tools like misti.ai, I'll show you that here, uh, which, in principle, ensure that the data you enter does not even reach ChatGPT in the USA, but is stored in a GDPR-compliant manner. This possibility exists. Other possibilities. Uh, we also have the possibility to use tools, for example, via DAC AI, uh, yes, there you upload your data, they are anonymized once at ChatGPT, which has then also been built to be completely compliant with data protection law, uploaded to ChatGPT, uh, and what you get back is then de-anonymized and displayed to you. That means personal data does not go in there. There are many solutions for this problem, and therefore here I have an appeal. There are solutions for everything. The real risk is not using ChatGPT and AI. And that is absolutely serious, because if you say, I cannot use this for data protection reasons, we cannot integrate it, your competitors will do it. And if all of us in Germany say, we cannot use it for data protection reasons or we cannot use it correctly, then competitors from abroad will do it, who will overtake us to the left and right. And I can promise you that, because it is such a relief of work, I have never seen anything like it in my life that is so quickly learnable and, uh, brings so much and helps so much. Therefore, the risk is to say, we cannot use it, we are not allowed to use it. Uh, there are solutions for everything, be sure of that. And if your data protection officer says, it doesn't exist, then ask the AI properly, inform yourself briefly, then tell him how it could be done. So, uh, let's move on to the next topic. Now, one more small setting. I'll quickly go through it. So, let's go in here. If you click on your name here, you can individually configure ChatGPT. How should I address you? What is your profession? What personality should ChatGPT have? You can say, for example, it's more of a thoughtful and helpful listener, is it the informative and enthusiastic part, or the cynic? What characteristics should ChatGPT have? Uh, here you can practically set up your counterpart, namely the AI, as you would like it. Uh, and then the information will be played out in the end as you would like it. Perhaps there will be a joke in it or something, uh, depending on what you need, yes, or short, concise answers, you enter this once, and then you get exactly what you want. And now we'll go into the really strong functionalities, uh, and see what, uh, ChatGPT can do. I have created a fictional tenant list. Uh, I did that with another AI, uh, where tenant names are included. I have included a few more pieces of information: last rent increase, last renovation dates, total rent, etc. Uh, and this data, it could be a tenant list from anywhere, actually. And I'll bring this into ChatGPT now and want to have it visually represented and visualized. And for that, I still need to find the file, so not search. So, Tenant List Munich, I'll drag it in here. And now I'll tell ChatGPT, for example, create all diagrams for the temporal development. Uh, as described before, uh, timeline, right? I think I copied something wrong there, but okay. Timeline of move-in dates. Show how long, uh, the tenants have been living in the house. Temporal representation of the last rent increases, and, uh, I'll go directly to start. Now you see here the tenant list has been directly adopted accordingly, and, uh, now the whole thing is being analyzed, and I will soon get a graphical, uh, or graphical evaluation of this data with a diagram. Of course, this doesn't work if we, well, not that it doesn't work, but currently it doesn't work if we have 10,000 data sets or more. That might be a bit too much, but with data sets up to, I don't know, 100, 200 numbers, depending on how many columns you have in our table, uh, it works very, very well that we can see directly, uh, yes, how the whole thing is actually, uh, graphically represented. And because I said, uh, it sometimes takes a while for it to, uh, work. Ah, okay. Here we already see a timeline of move-in dates. That has already come out. Uh, I have also prepared the whole thing. Exactly with the same, uh, yes, almost the same prompt. I forgot something. Uh, what, what is above. First, the results were not really that exciting, or rather, this result here is honestly not absolutely exciting. Uh, and then I said, look, I now also want a landing page from you with interactive diagrams. Uh, please add further meaningful diagrams that I might need for the assessment of the property. Then ChatGPT thought for 28 seconds and wrote this code for me. As I said, I have no idea about code. I'm glad I don't have to write it, but I'm also glad I don't have to execute it, because I wouldn't be able to do that either. But I can now simply say preview. And what is happening now, uh, is that this interactive analysis of the tenant list, this landing page is being loaded. It's still taking a few seconds. It's somehow, uh, stuck right now. I have the feeling. I'll reload it again. Uh, I can now load it, view it, and so on. And, uh, yes, now it's going through. And I
So, with two prompts, with the first one you saw, which I entered, and then with the statement, "Can you please build me a landing page with an interactive diagram for this?", I got this landing page. I could download the code for it and also make it publicly available somewhere, for example. But for now, the focus is on adding this tenant list here, because as soon as I add the tenant list, these diagrams will also be filled with life accordingly. Rent duration of current tenants, yes, maybe not the most beautiful representation, but normally you need seven or eight prompts for something like this to make it really perfect, just as you would like it. Distribution of rental durations is shown to me. Timeline of move-in years is really particularly exciting. But well. Last rent increase by year, you can also see rental durations, rental duration versus current rent is displayed, and the potential list since last rental duration, since last increase, current rent, total rent, yes, such things have now been automatically included. I don't want to show you that we have the perfect overview and the perfect diagrams and the perfect data research here, that's not what I want to show you, but what I want to show you is how easily you can build this, exactly as you would like it. Yes, you just have to enter textually what you want with this file. And a landing page will be created. Diagrams and overviews will be created, and these are partly interactive. I could also say, "Only from two rooms." And you have seen that things are changing directly here, yes, or only from four rooms. So, I have an interactive page with which I can do all the analyses, and it's not just for data in this form, i.e., structured datasets, but also for all other forms of data that you can put in there. This is just one example of many. We are practice-oriented here and therefore build many examples. Yes. Okay, let's move on to the next topic. I'll go back into the presentation here and there. Just so we always see what the next topic is that we're going to deal with. And this is one of my favorite topics. It's been around for some months now, I think even for a year or a year and a half, so-called Custom GPTs. Why is this my favorite topic? You can train ChatGPT to do exactly what you want it to do, or what you would normally do yourself, but then ChatGPT does it. That means you enter recurring tasks once, build a so-called Custom GPT, and then, when you just enter information there or say "Start" or whatever, you get exactly what you actually want. This might sound a bit vague now. I'll show you what the whole thing looks like in concrete terms. Some of you may have already written emails with me in the "site" manner. I have to admit that my emails are not, or in rare cases, really formulated by hand, but they are always semi-automated. And I know that with Copilot, with other tools, there are now possibilities to write emails completely or whatever. Personally, I'm a fan of having semi-automated email creation, and that's why I've built a GPT for it. And if I now, for example, click on my microphone on the laptop and say verbally what I want here, then I just say, "Write an email to Mr. Müller. Hello Mr. Müller, thank you for the friendly conversation. Here you will find the presentation on the page, here is the sample analysis or sample expert analysis you requested. Let's have another phone call next week for the next appointment or just look directly at my appointment booking link. Appointment booking link, appointment booking link, best regards." So, that's fine. It already recognizes what it's supposed to do. And now I just press Enter. And then the following happens. The email is output to me. I took 15 seconds to speak it in. I could also just give keywords, then it would be even faster, but I still have those 15 seconds, let's say. And what comes out now is the subject line, exactly as I always write my emails, namely "site" and then this dash. The email is written professionally and friendly. And not because I said so here, but also because it's stored in the GPT. And what is also stored in the GPT are these links, namely to our presentation and to the individual expert analysis and to my appointment booking. This means I can now take this email, copy-paste, it would also work differently, yes, but that's how I do it right now. Copy-paste into my email tool. And then I send it. This means that if I had written this email individually, looking for the links and without a template, it would take 4 minutes and 30 seconds, and then it would be sent. I can skim through it. In most cases, it's correct. These GPTs, this is a way for you to automate recurring tasks and how to create these GPTs. I'll show you this with another example. It's not about emails, but about what do we have here? about a specialized assistant for renting office properties. You click on GPTs, click on Create at the top, and then you see the overview for creating the GPT on the left side, and the preview on the right side. You can always test it while you're creating it. And now I'll say "Office Assistant" is what it's called, description, compiles exposé documents. As I said, if you misspell anything, it doesn't matter at all. I have already had a very large prompt generated by ChatGPT beforehand. I'll paste it in here. You see now, I'll make it bigger. You see "you are a specialized," it's actually first, but it's actually first, "you are a specialized assistant for renting office properties," and basically it says exactly what it should do here. The goal here is, I upload all the information about the office property here and get all the information structured exactly as I need it for my exposé creation, for example, I get it out. That means, first, please give me the basic data, then give me the location description, the areas and layout, equipment, technical and legal information, special notes, additional rules. And here's an example behind it. I never wrote this prompt myself. I simply told ChatGPT, "Please create a prompt for creating a Custom GPT." I recommend you do the same. It's simply the easiest way. You have the option to upload further knowledge in this Custom GPT. Not too much. If you upload too many files from which the GPT should draw information, it becomes inaccurate, I wouldn't do that. One or two tables that are not too large are okay. And now you could also say, "Yes, please use the model GPT-4," that's enough for that. Okay, done. Alternatively, you create this GPT by clicking on "Create" and simply entering textually what you actually want, what the GPT should do for creating emails. I created it by simply saying, "You are now my email creator, and here are ten emails that I wrote myself. Look at how I write an email, how I write emails, and learn from it, and incorporate that into your description so that you know how to write my emails in the future." And then I also uploaded a file with the links to the knowledge or to the things I incorporated. And that's why it worked very, very well. The first time it took half an hour to build. It saved me an hour of work on the first day. You can't get a bigger ROI than that. Just say, "Yes, I'll leave it like this now," and say "Create" without testing it here. I'm now creating this GPT. Make sure you always say "Only me." Because if you click on "Everyone with the link," then anyone who has the link can actually use your GPT. Recently, there was a problem at Google, let's say, where you could actually search for public GPTs. I just entered "rental contract" and quickly found the IBAN of someone who uses it to create and write their rental contracts, but who made the assistant public. It was harder to find his email address to tell him something than his email IBAN. You have to know which prompt to enter to get the information behind it. But well. So, and then let's take the example property. I can also show you very briefly, right? Wait. This is what the files I upload look like. So, I basically have a commercial property, and I had fictional data generated for it. When are property viewings possible? What are the transport connections, and many different files. I won't show you all of them now. That would go too far here. Where was I? These are the files that you, for example, received from the customer. And for whom you are now supposed to market this office, and now you have to press Enter. Now it searches its knowledge and gives you this structured overview for the property Minzer Business Center in Mainz exactly as you intended and specified beforehand. Exactly in the structured form. Of course, there are ways to transfer this directly into an exposé via various other tools, etc. That's possible, but it's not possible with ChatGPT alone, and therefore not part of what we're looking at today. But what I want to say with this is that you create this assistant, upload all the files you have for it once, and get the files out exactly as you need them, for example, for an exposé. This is just one example of many, because in acquisitions or as a property manager, it might be necessary to simply ask a question about it or to be able to read out all the files at once. And that's exactly what you can do with this, and that's exactly why you create a GPT via this "GPTs" area. For me, this is one of the most powerful tools. You see, create LinkedIn posts. I built this very recently, after I had created many LinkedIn posts, I told it to create LinkedIn posts like I do, and it actually works very well. Yes, we'll move on to the next point. ChatGPT can perform tasks for you. But "tasks" is not what you might directly understand by it, but rather tasks are something that happens regularly. So there's a regularity behind it. If I now say, "Inform me once a week about the latest and only the latest new construction projects in the city of Düsseldorf in the office, commercial real estate sector, not just building permits, but also publications and planning, demolitions, etc." It needs to think for a moment, we already know that. and sets up automation. And I will receive the latest information about it once a week, we'll see which day and time it chose, because it will scrape the internet, find the information, and make it available to me. Now it says, "I will inform you every month on Monday at 9 AM about the latest new construction projects." This is actually very, very good and very practical. Do this, for example, with new subsidies for renovations, etc. There are many possibilities to use it in exactly this way. Yes, with an eye on the time, let's move on directly to the next tool, the next product that ChatGPT offers you, and that is called Projects. You find Projects here, and you create a new project here. And now let's take the Logistics Portfolio, for example. We create the project, and what we then do in this project is add files. I have to find those again. Logistics Portfolio. I also had files created for it. I'll show you one or two files. I had exposés created for seven logistics properties. Just as an example, and tenant lists from them too, or I had them created fictitiously as well, and I'll upload them all here now. It needs a short moment to load everything. And what this Projects actually is, is an intelligent AI data room. I recommend creating a data room for every project you work on, putting all the files in there, maybe all the emails too. You can also set instructions. I'm not doing that now, how ChatGPT should respond in the project, but you now have a very quickly automated plan with which you can ask questions about this portfolio, for example. If I now say, "Which tenant pays the most per square meter?" It searches all the files I uploaded in that second and gives me the corresponding answer. You can use these projects with multiple people. You can use them with your colleagues and simply upload all the information about properties you are currently reviewing, working on, and then you no longer have to talk to colleagues to ask what can be done. Visualization could not be displayed. Okay, but it's still analyzing. Strange. You no longer have to talk to colleagues, but you get an immediate answer and get many great objects or documents automatically searched. Now I see, yes, it's not working here. I also prepared this. I had this Logistics Portfolio. So, but that's already uploaded. Which tenant pays how much per square meter or pays the most per square meter? You see, there are also situations where the correct results come out. And the result, the highest price per square meter was achieved in Frankfurt, for example. Yes, you can do this for anything. Set up a project in ChatGPT, upload everything, ask questions, or get summaries. It's super easy and you'll never have to search for anything again. Trust me, you won't. Yes. We'll move on to another area. namely the area of photos and object recognition. We'll also go into the area of visualization shortly. I have selected a four-room apartment in Bergheim and have some photos, I think the ones you see here are the ones downloaded in the end, to show you how well where was I? to show you how well ChatGPT, or not well, depending on it, is at recognizing photos and deriving information from them. I can now simply upload these photos and say, for example, "Describe the condition of the apartment to me. You are a real estate agent and want to sell it, but don't sugarcoat it. Describe above all what you see visually. Focus on the condition, style, and modernity, the equipment. The furniture doesn't matter. Look at the substance of the apartment and floors, laminate in dark wood look. Looks clean, but not particularly modern. No visible damage. Walls and ceilings painted white, overall tidy, but with a textured ceiling that looks dated." You might see it here. I hope it's transmitted well. Yes, correct. In my opinion, the assessment that it looks dated is also correct. Absolutely correct. ChatGPT recognizes this. Bathroom hallway tiles, light beige tiles in retro style. Sanitary objects are older, but functional, not modern. ChatGPT recognizes all of this. If you need a description for a property, upload the photos. If you have hundreds of photos of the property you are reviewing, upload the photos and get a summary. You can trust it well, I'll tell you honestly. Yes, and in that, ChatGPT is very good. And now we come to what you might say, yes, maybe. These are then object visualizations, and I've chosen a property with a Google review, a detached house with a gravel front garden. That's not everywhere anymore. And I'm now saying, "In this photo of the detached house, replace only the gravel front garden with a green, well-maintained garden." And I want to show you this with the example photo I have here and then say, "Let's go." And now let's see what comes out. The results here for visualization are not perfect. There are other tools for that, where they are already perfect. It's very likely that they will be almost perfect in about half a year, because things are changing so quickly. If it's only interesting to you in half a year, you don't need to look for another tool for it. Since this is taking too long for me, let's also look at what I prepared yesterday. And you see this result here. Yes, the building doesn't look photorealistic, does it? But the building is basically exactly what we saw before, just with a different garden. That's fine. So, you can also do that with it. And, I think we just had that. And now we come to two topics, especially those that will likely bring you a strong "wow" factor, and for me, honestly, too. There are further functionalities in ChatGPT, namely a Deep Research functionality. That means you select ChatGPT here, the Plus Deep Research, and then start. That means ChatGPT is as fast as you normally know it in the overview of what it can do. Here, however, it takes 7, 10, 15, sometimes 20 minutes to pull all information about something you request from the internet. First, you will be asked questions here. I'll answer the questions first and then tell you what I actually asked. Unfortunately, you can't prepare for the questions because they are always a bit different. Planned investment period. Now, I don't know, financing status, closings, everything. Fourth, planned target group, doesn't matter, whatever gives the highest subsidy, doesn't matter, energy efficiency measures, doesn't matter, depending on the subsidy, I don't know. So, what did I actually ask here? I want to transform an existing office property in Cologne with 3000 sqm GFA into residential and build micro-apartments. I expect 80 micro-apartments. I also want to energetically renovate the building at the same time. I am explicitly looking for all subsidies that I can receive for the implementation of the project. Check if they are combinable with each other. Give me information on which combination makes the most sense, how I should calculate it, and how subsidies would be provided. Give me how many subsidies would be provided. Also give me the relevant links so I can check them. Before you answer, check your answer three more times. Now it's creating the research, and it will take time. That's why it's particularly good that I prepared everything. What comes out here, you can already see when I scroll up, this is the answer. What comes out here is incredibly much. It's a current representation of current subsidies, because ChatGPT is on the individual pages. It might be too much if you really have to read everything, but you can also have it summarized by specifying. Okay, it works, of course, but believe me, the answer here, it took 7 minutes for that, searched 8 sources and made 59 searches. This answer here, I read it yesterday, is extremely valuable, and basically you get everything you need to implement this project. We get the subsidy amounts, the individual KfW stages, which bonus subsidies we get, which cash options we might also have in addition. State subsidies, housing subsidies. I don't even want to go through it all because there's a lot here, but you can be sure that you've received very good research here, which a working student wouldn't have done better. And if you are a working student, just use it and submit it as your own, but read it through again, of course. It's very, very good, and it takes, now I have to find what I had before, sorry. And it takes time, but during that time you can go get coffee or lunch or whatever, you get the research for anything you can imagine, anything you've ever dreamed of. Yes. And then there's ChatGPT, Deep Research is one of the biggest recommendations here. And now comes the second big recommendation. It's only been available for a very short time. That's the agent mode. The agent mode does something similar, only that it not only researches text and information and PDFs from the internet, but uses a browser. And by using a browser, this agent mode, which also takes a similar amount of time as ChatGPT Deep Research, has the ability to click on things for you. Yesterday I spoke to someone who said, "Yes, the agent mode booked my hairdresser appointment." Yesterday I requested the invoice from Enterprise for booking a rental car, also via the agent mode, for which you can use it, for example, right? Automatic retrieval of documents from authorities. I would say, just try it and say, so we won't go through it to the end, and I would actually need more information. Request a printout from the contaminated sites register of the city of Düsseldorf. I am not the owner, I'll mention that beforehand. Agent mode, let's go. This will take a very long time. We won't manage to look at the result at the end, but where it asks again. It's thinking. Very good. And now you see this window here. And you can also see what the tool is searching for, what it's currently finding, and where it's currently clicking. Sometimes you'll be told to take over briefly because there's a CAPTCHA that's not a computer. You have to select something. And then it continues. Yes, it has now detected a gray screen, but it will find its way through somehow. In the end, it will automatically fill out this form for you if you create a GPT beforehand where you have all your details about yourself and what needs to be retrieved, and then just say, "Please retrieve the documents." Then it automatically goes to the page, finds the page, retrieves the documents, you might have to look again to see if you need to click on anything else, but it works, and it also works with other things. For example, I also did the test and told the agent, "I'm interested in buying a plot of land and want to build a multi-family house with five units there, which is barrier-free and age-appropriate." Now, please do a location analysis, a micro-location analysis, and connect my planning with the concept. And this here is the ImmoScout listing for the plot of land. And this one, yes, let's go to the next. And it took, I think, a very long time, we can look at it once. And it took a very, very long time to give me this overall assessment and the micro-macro assessment. That is also very, very good. I would use the agent for that, because the agent can click on things. For other things, I would use ChatGPT Deep Research. 16 minutes, 23 searches, 174 sources. The result is good. Once it didn't answer me, I had to briefly ask it, I had to briefly ask it how it's going. Ah, I didn't enter the exact full link of the plot. Never mind, here's the result we have. Another idea is to search for and find the development plan. That also works. It doesn't always work 100%, I've found, but it works. Due to time constraints, we won't go into that anymore. Yes, we have 5 minutes left. Therefore, I'll move on. All these tools together, as you've seen them now, are basically a complete assistant for a lot of your daily business. If you create a Custom GPT here, then you can use it, or you can create 20 and use them to handle recurring tasks. The tool can research perfectly with the agent mode and with ChatGPT Deep Research. The tool can visualize and display data and graphics for you. Also no problem at all. It can do a lot of what you do in your daily work, create emails, and so on. And now there are possibilities of use via a so-called API. That's an interface. An interface basically means we take over ChatGPT or we apply ChatGPT, for example, to your CRM system, and both work together. I can then directly say from the CRM system or press a button and say, "Find all information you can find about this customer. Why is this customer interesting for us?" And show me that here in the system. That means you don't need the detour via the ChatGPT website, and that provides the possibility for complete automation. You transfer data between two systems without having to copy anything. Completely automated. This is now step 3, let's say, for those of you who have just started with ChatGPT, with AI, with automation. This is step 3. But you can build a small automation using tools like Make.com or Zapier. For example, if an email arrives, have it summarized by ChatGPT, and send me an email from ChatGPT with the summarized email in two sentences. That would be something very simple to try at the start. What you can also build, I wanted to show you how it all works. I built my own website, by the way, with another tool, an AI tool, with three prompts in 15 minutes. It's very, very fast. And here I have stored the ChatGPT API key, and with that, this tool now works with ChatGPT. I can upload an acquisition profile, it is automatically extracted by ChatGPT, and then stored in a virtual CRM system that I built here, linked to customers. Now it's analyzing, because it just sent it to ChatGPT, and now it's outputting my acquisition profile here. Somehow in Wiesbaden I'm looking for properties that have these returns, etc. And, well, I was able to build the entire website very quickly with AI. Maybe you can't really see here what an API can or can be. Yes. But on this website, ChatGPT is working. That's the simple explanation. With regard to time, so APIs, if you really want to automate something core, use APIs and interfaces, and if you don't know how, ask the AI, ask ChatGPT, it will tell you. Prices and costs. I just wanted to briefly touch on the Plus version. I said before, 19 euros, 23 euros per month. I would recommend that. You'll get your money back on the first day, so to speak, because you have so many super good accesses to tools. I don't have the Pro version myself, I don't need it. ChatGPT-4 unlimited, Pro models unlimited, I don't need all of that in everyday life. And this Team version, that's the one where you can work very well from a data protection perspective, let's say in a company. Team and company accounts are the ones where you have to worry less, or where you can solve all problems. There we are at 29 euros per, not us, I don't sell it, I don't work there, but at 29 euros per user, and you have admin functionalities for companies. That's not a lot of money for what you can do with it. For larger companies, there are enterprise versions. So, I'm one minute before the end, just managed to get through it by skipping one example I wanted to give. This one hour should give you an insight into what's possible with different examples, with different tools in ChatGPT. The recording, as I said, you'll all get. It will be sent to you, we'll upload it. If you have any questions about any AI topics or about Side, we didn't talk about that today. This is the QR code. If you enter your email address there, you'll automatically receive an email back with my contact details. Then you can ask me questions, you can write to me, I'd be very happy to. I really enjoy it. You can also write to me on LinkedIn. Follow Side and me on LinkedIn. Nilas Möllenkamp is the name, easily confused with Niklas or Nils. That happens often. This is the correct QR code for that. And now, with an eye on the future, registration for the next webinar. I should have made this smaller. Damn. For the next webinar on September 23rd, we'll be talking about automated acquisition and property analysis thanks to AI. It will certainly involve ChatGPT to some extent, but also many other AIs. And we'll break down the acquisition process and see what we can do with AI. Also a free webinar, you can secure it in your calendar immediately and register immediately via this QR code.