Speaker 1: Do you want to master ChatGPT? In this complete course, I will explain everything you need to know to go from zero to an expert in ChatGPT in its free version. And this course is updated with all the new features available in 2025, such as the advanced voice function, SearchGPT, or the new Canvas function, which is tremendously useful. When you finish watching it, you will know the secret for ChatGPT to write just what you need and can assist you in a multitude of tasks. whatever it is, summaries, essays, articles, social media posts, or email. A complete course where I guide you step by step to get not only high-quality texts, more adjusted to what you need and in less time, but also so that you see ChatGPT as an assistant that can help you in your day-to-day situations. For example, did you know that ChatGPT can be your language interpreter always available on your phone? A course with all the theory, specific examples, and resources to expand the limits and functions of ChatGPT. In this video, you will discover everything you need to turn ChatGPT into your essential tool. Let's get to it. The lessons in this course are a small sample of the content published in my weekly Artificial Intelligence Academy and a small part of the lessons included in my course Introduction to Applied Artificial Intelligence. If you want to know more about how to keep learning with me, You can receive all the information through the link that you will find in the description. In this week's module, we are going to make an introduction to everything related to the interaction with tools based on language models. And to do it in the easiest and most affordable way possible, what I have proposed is a complete course of ChatGPT, but in its free version. Why? Because ChatGPT in its free version allows us to interact with a very capable language model, such as GPT 3.5 Turbo, and from this interaction with this model, what we can do is learn everything essential when we later relate to more advanced models, such as GPT-4. That is, understand a little how these models work, understand their capabilities, understand their limitations, and, above all, understand how we can ask them for things so that they really satisfy what we want. But to make it clearer, what I am going to explain to you today I have developed this guide where we can see what I will explain in each of the lessons of this module. As I have already told you, it is basically a complete course on ChatGPT, the free version, and in the first video what I am going to explain to you is what ChatGPT consists of. We are going to review a little the concept of language model, and then we are going to see a little its interface and functions, so that you are clear about everything you can do with ChatGPT, from the most basic management functions of the information that we introduce and that provides us with more advanced functions such as conditioning its operation with functions like, for example, Custom Instructions, we are going to evaluate a little its capabilities, what it allows us to do and what we can expect from it. We will also talk about its limitations because as a tool, as a language model, it has certain very clear limitations that will help us understand what we can ask for and what we cannot. And we will also see what is the most efficient way to start start applying chatGPT in our day-to-day life? What approach should we adopt? From here we will move on to a second lesson where we will talk about the main uses of chatGPT. Here I have dedicated a section, although it will be a short video, because it is important to understand this. There are two functions of chatGPT, one is to process language and the other is to use it as a source of factual information. And what I am going to transmit to you in this lesson is that the first element has much more weight than the second. If we use it for the first thing, we will find a much more capable tool. From here we are going to go to the daily use cases, that is, in what type of actions and activities we can use ChatGPT. In this case, we will talk about these four elements, such as text writing, which is perhaps what comes to our mind in a faster and more evident way, but also as a support tool for specific actions, such as transforming the text into other formats to use it in other software, support for learning, or also in brainstorming processes. We will also see strategies so that you can start using it from there. All this will be useful for free chatGPT, but if you are paying for chatGPT+, it will also help you to interact with GPT4. From here we will go a little to the key concepts to understand how to interact with ChatGPT, what the context window is like, how to understand what hallucinations are, how the training dataset conditions the answers that ChatGPT can offer us, and also a very important thing that we talk less about, but that is essential to understand that ChatGPT is a tool that we cannot expect to always behave in the same way because it receives behavior adjustments. I will explain what this is about. Later, in the next lesson, we will learn how to relate to ChatGPT, which basically means learning a bit about Prom Engineering. We will learn what a Prom or Indication concept is, a bit about the basic foundations of Prom Engineering, which are basically initial strategies to know how to interact with ChatGPT, what writing format is the best for Proms, and this is something that is not as important as Promting strategies, but it is very interesting to ensure that ChatGPT understands what we ask for, what types of Prom exist, which are basically the ways or tools we have to articulate how our conversation with ChatGPT is going to flow until we reach a goal, what response formats we can obtain, and finally, we are going to talk about the structures and templates of more common PROMs, which, although it is not what I like the most, I do understand that it can be useful, especially for a user who is starting to familiarize himself with this type of tools. From here, in the next lesson, we will learn how to improve the capabilities of ChatGPT, starting with a native tool like Custom Instructions, but also seeing how external elements can help us improve how we use ChatGPT, such as Chrome extensions, and the Windows interface itself, which can help us give it extra capabilities that, in certain circumstances, can make it a much more useful tool. And finally, as I said, I have used a ChatGPT course to introduce us in the world of tools based on language models. But we must be clear that not everything is ChatGPT and more if we take into account that we are moving in the field of free tools. So I'm going to finish this module with a lesson reviewing how to compare ChatGPT with other free tools based on language generation models, such as Gemini, Copilot, HackingChat, Perplexity, Cloud, PoE, GPT4ALL, or Chat with RTX, that this is a novelty this week. This is interesting to understand that even if we learn to use a single tool, in this case, ChatGPT, what you are going to learn today is going to help you start using any of these tools, because the basic and fundamental principles are the same. So, once you have presented today's module, let's go there with the first lesson. And let's go with the first lesson of this chatGPT course, which, as I have told you, answers the following question, what does chatGPT consist of? And basically I will explain all this to you, that chatGPT consists of a language model with specific characteristics that make it especially suitable for conversation. We are going to review its interface and functions a bit, and this involves all these sections, the history of conversations, the configuration options, privacy and security, which are the custom instructions and a bit the elements of the conversational interface, and then we are going to talk a bit about the different capabilities it has, both in language generation, in language manipulation and in basic logic. And finally, we are going to focus on its limitations, such as the lack of consistency in its responses due to the nature of this technology, the hallucinations, the memory limitations, and that it has a limited knowledge. To finish, talking about the applications that we can use in ChatGPT, in which tasks it is ideal to use ChatGPT, which, as you can see, will have very specific characteristics. Well, let's go for the first factor of this lesson, for the first element, which is to explain a little what ChatGPT consists of as a technology. And indeed, ChatGPT consists of a language model, that is, a model of artificial intelligence trained to satisfy a very specific task. And this task is not to create or understand language, it is simply to predict the next word or token, that is, a set of letters that sometimes does not even reach a word, that must follow to complete a text in the most coherent way possible, so that it responds in the most precise way to an indication. And here what I am going to do is explain that the model on which ChatGPT is based in the free version is originally a model that was born or emerged in 2020. It is a model called GPT-3 and on this model many tools have already been created that on paper were very similar to ChatGPT. In this case, I'm going to talk about Reiter, which was one of the first tools I brought to the channel, which, on paper, did almost the same thing. You gave it an indication, it has different tools, in this case I'm using the tool to get a bit of the structure of a blog, but basically you selected the language, you selected a tone, things related to the PROM, a use case, and you introduced an indication, for example, the keyword for which you wanted to create this blog article, for example, an article about dog care, and when you gave it to write, it generated here in an interface similar to that of a document manager, of a Word, it automatically generated these texts, it generated them online, and from here you could continue editing and working, generating new texts in a very, very similar way to how we do it today with ChatGPT, with a slightly different interface, but the concept was very similar. However, this tool was a much more cause-effect tool. You had specific and defined tools, and what you did was generate text from indications that intended to satisfy specific use cases. ChatGPT presented an additional difference with respect to these tools. The model used by ChatGPT was an evolution of the models used in GPT-3, which emerged in 2020, and which allowed us to interact with it in a more conversational way. To do this, what they did was introduce a way of training this model, which is called Reinforcement Learning from Human Feedback. In other words, to make the model more stable, safer and more coherent, What they did was to train this version 3.5 of the model with a mixed interaction between the traditional reinforcement training and the reinforcement training where the human helped to select which were the best answers. Basically, what this training method allowed was to accelerate the improvement in training by reinforcement of learning from the different agents and this helped that GPT 3.5 above all was a safer, more stable, more predictable model and that could be prepared to act in long conversations, that is, to convert the interaction with these models into an interaction via chat. And this is what changed everything, in addition to the fact that they obviously launched it as a free tool, since all the previous tools were payment tools. But the fact that we can talk to ChatGPT, having this interaction, is what automatically turned it into a much more versatile tool. And once this is said, once we have identified what makes ChatGPT different from other previous tools also based on language models, it is time to start seeing how it works, how it is structured, and what we can do with it. So the first thing is to know its interface. Let's move on to the chatGPT interface. Remember that many of the functions that I will now explain to you, and some more that are only in the paid version, are also available for free in other alternative tools. In the description you will find a link to discover these tools and have more options to take full advantage of artificial intelligence for free. Take a look at the link that you will find in the description. And let's go there with the interface and functions that ChatGPT offers us. In this particular lesson, I am going to focus on the free version of ChatGPT. If you have ChatGPT Plus, the paid version, I recommend that you see directly the lesson where I explain the interface and functions of ChatGPT Plus. In this lesson, we are going to review the different sections that the ChatGPT interface offers us, from the chat history, the different configuration options, privacy and security, the possibility of adding custom instructions, although to this we dedicate a lesson later, and also all the options that are inside the conversational interface, that is, inside the box where we introduce the prompts because it allows us to do some things that can be useful in certain circumstances. That said, let's go to gpt and go over all these sections. I'm going to leave the part of the conversational interface for the end and I'm going to focus first in the configuration part, which we have up here in our name. If we click on this little button up here on the right, you can see that this menu is displayed, where we have the different options. We have the option to see our GPTs, which in this case, if we click, since it is a payment option, we will not have any, we click here and, as you can see, we access the GPT Store directly, but we do not have the option to create any GPT, because this is a function of GPT Plus. The next option is the option to customize chatGPT. This option here is the option of the custom instructions. What does this mean? That we can give some information to chatGPT so that it acts in a specific way in all conversations by default. We will dedicate a lesson to this later, but you must take into account, as a first general advice, that if you want to use this function, the instructions that you must enter have to be very general and serve you for practically any case of use that you give to ChatGPT. Basically, the use that I have identified more clearly for the custom instructions is to customize the tone and the way we want ChatGPT to write. But we will see this in a lesson later. The next thing we see is the configuration part. And in this configuration part we have all these elements. We can configure how we want ChatGPT to look, whether in a dark theme or in a light theme. I think that if we are going to be in front of the screen for a long time, the dark theme is much more comfortable. We have the option that when it generates code for us to run in the data interpreter, which is also something we will see later, it shows us the deployed box and we see all the code that it is generating or that it does not show. I personally usually have it deactivated because in the end it takes up a lot of screen time and for what we use it, it is not relevant most of the time, and we can always deploy it whenever we want. We can define the language by default of ChatGPT, which can be automatic and will take it from the system or from the conversations in the context we have, or we can select it within the list that is available. We can also manage the archived chats, which now I will tell you how you can archive chats within the ChatGPT interface, and by clicking on Manage, the conversations will appear here. In this version of ChatGPT I don't have any saved. And you also have the option to archive all the chats so that the history column is empty. We could even delete them definitively. These would be the options on the General menu. In the Customization section, basically we have the option to decide if ChatGPT can remember things from our conversation or not. That is, as we speak with ChatGPT, it can add information that it considers relevant, that it can remember and that it can use in isolated conversations. I personally, in this free account that I do not use so often, it is enabled, but I have it disabled. Not for any specific reason, but for the simple fact that I use the ChatGPT tool in very different contexts, many times in training, in different sectors, and my conversations with ChatGPT sometimes have nothing to do with what I do. If you're going to use ChatGPT individually and you're interested in it being able to remember certain things, that it can remember things that are related to different projects, it can be interesting to have it activated. This is already a personal decision. We also have all the voice configuration options. Here, what we can do is choose or play the different voices with which ChatGPT can read the different messages. And we have all of these available, so that if we now what we do is play some of them, for example, Enver's, we hit play.

Speaker 2: Hi, I'm looking forward to getting down to work. So, if you want me to focus on something first...

Speaker 1: And here you can see how each of the voices sounds and choose the one that works best for you. We have the whole part of data controls, that is, if we want ChatGPT to have access to our chat history or how we have worked with ChatGPT to use it to train other models. Before, this in the free version was activated by default and now we have the option to activate or deactivate it. I have it deactivated, I prefer to keep this level of privacy. We have the part of managing the shared links, which is when we activate the function of sharing a link of a specific conversation, which we will see later how it is done, well, they would appear here and what we could do is delete it if we consider it relevant, that is, make this link stop having an effect and the people can access this conversation, so we have control over the conversations we have decided to share. We have the possibility to export the data and delete the account. The builder profile, at the moment of the course, is not so relevant. It is essentially the site where we configure our public profile when we create a GPT, a GPT assistant, and we share it publicly, so our profile would appear here, as we show to the rest of the chatGPT community. And then, this part is more interesting, we have the possibility of linking different tools to chatGPT, our Google Drive account or our Microsoft OneDrive accounts. In this way, when we upload files, we can not only upload a file directly from our computer, but if we have this connected, what we can do is send access to our Google Drive through an emerging window and provide the files from there, which can be quite practical, especially if we work from different devices. And finally, there is the security option, which are options linked to how this account is activated in different devices and to be able to close the sessions, and also in the part of if you want to enable another layer of security when accessing the account with a multi-factor authentication. With this, we would have defined all the options for configuring chatGPT. The next interesting point would be to see the whole part of this sidebar here before going to the part of the conversational interface, which is where we really interact with chatGPT. So, if we go to the home page, which would be this one here, is accessed through this element here, we can see how this panel on the left is related to the conversational interface. Here we can see the different conversations where, if I click on any of them, we can access and continue them. I must also say that they have incorporated a very interesting function such as the search in the chat history. Now we have this icon up here, where if we click on it, what we get is to have a search engine available where we can search within our chat history, which can be really very interesting to recover conversations where we have achieved a specific goal, recover examples of satisfactory results, or simply continue a conversation where we left it, for example, if we want to update a presentation that we raised in ChatGPT. At that point, we could recover the conversation and iterate some of the slides. This element here is the element where we simply start a new conversation and the element up here, on the left, is the one where we deploy or close this sidebar. Within the conversations, we also have options. If you look, here you have three dots that, if we click on them, we can share them. as I mentioned before, to create a public link to this conversation so that everyone can access it. We can change the name so that it is easier for us to identify them if we are looking for them, or we consider that it is not enough just to save them in the side panel, but we are interested in being able to identify them quickly. We can archive them so that they are removed from this side panel, but are saved in the archived conversations, which we can then manage through the configuration panel, as we have shown before, or we can get rid of them. Before, the function of changing a name seemed quite interesting to me, because since we did not have the search function, what we could do is simply, between keys, put a keyword that would help us, even in capital letters, that would help us to identify, because it was relevant, a conversation a little in the medium or long term. Now, with the search function, this is no longer so relevant. And the other thing that I think is very interesting is the part of sharing. Being able to create a public link of any conversation to share it with co-workers or to share it directly, publicly. These would be the main functions that we have here in the sidebar of ChatGPT. We have the most interesting part, or the part where we are really going to work with ChatGPT, which is the conversational interface. That is, the box where we enter the prompts. And this started out as just a box where we enter prompts and chat with the AI, and now it allows us some additional options. If you look, they have tried to give us a quick idea of ​​what the chatGPT can do, adding a series of shortcuts, which are still prompts that are executed in a couple of clicks, of things that the chatGPT can do. If we want to summarize a text, if we want it to help us write, if we want them to give us advice, they even give us the shortcut to create an image. In principle, the image generation function is a function of ChatGPT+, but recently I think they have enabled the possibility of creating an image in the free version. So if we click on these elements, we can see different prompts to generate an image. Here we can see, if I click, for example, on Create an image of a pet, Here he tells me to give him more information, he is a big German shepherd, I send it to him and I can tell him, I imagine it, normal, without any disguise. And here you can see how, after a brief conversation, you can also generate images in the free version of ChatGPT. Here we can see the image of a German shepherd. Having said that, it is interesting to see these PROMs more than to actually use them because they are a summary of the capabilities that you have access to with ChatGPT. You can create images, you can summarize texts, you can use them for brainstorming, to give advice, to plan some kind of action, to analyze images, and so that it helps you write. And I think that I have also used the part of analyzing data because although in ChatGPT, and now we will see it in the drop-down menu up here, you can select different models, in principle, the mini, I think, does not work with the part of analyzing data, it has to be done with the plus model, but they give you a certain layer of use of the plus version before jumping to the normal version, which would also allow you to analyze data from a file. Anyway, this use case in the free version is so limited that can be used to test it, you will see it in the lessons of JackGPT Plus, you can take a look at it, but in principle it does not make much sense to use it in the free version because it ends very quickly its possibility of use. That said, we have the entire part of the conversational interface. Here we can start talking to the artificial intelligence, we can provide additional information, we have the part of joining files, And here we can add, in addition to adding files directly, which we could drag from our computer, which can be images and have the vision function of JGPT to analyze the images. They can be documents and give it a context to ask about that document, or they can even be data tables and use the data analysis functions. Well, in this part, when we click Add File, we have the other options I have given you. If we have connected it to a Google Drive account, we could provide the files directly from Google Drive, and if not, we could download them directly from the computer. On the other hand, we would have the part of when we interact with it, what options appear within the conversation. And in this case, if I, for example, ask for something like this, you can write an email requesting... You can write an email requesting a budget for an economic washer and I send it to you. And here it gives me a bit of an example. And in this case it is not so much that you see how the email is written, but that it appears in the interface once the email has been written. and notice that we have a lot of options. Among them, there is the option to read it aloud, so if we click here, it will start to read.

Speaker 2: Here you have an example of an email requesting a budget for an economic washing machine.

Speaker 1: Another option we have is to copy the text directly and be able to take it to an external editor. We have the option of giving feedback to ChakGPT, that is, that this answer has seemed especially good to us or that it has seemed bad to us, and this can help a little ChakGPT to understand later how to generate better answers. And we have the option of generating with different models. In this case, as I was saying, we are working, even if we are in the free version, with the normal GPT-4O, which in principle would be that of the Plus version, but we have limited messages. And if we click here, what we could do is re-write it, to make it re-write the text with the GPT-4 or we could re-write it with the fastest model, the GPT-4 Omni. Or we could leave it in automatic mode and let it choose which model is best for each task. In this case, when we try again, we see that it has done it with the GPT-4 Omni. Well, in this case, I'm going to tell it to do it with the mini. If I click here, you'll see how now it writes again, it does it a little faster, and it elaborates the same answer for me again. Why have I done this? Because look at what happens now. We have here a slider where we can navigate between the different answers. And this is a very useful function that many people do not take into account when interacting with chatGPT. Why? Because if an answer is not good enough, but we understand that the PROM is correct, we can try to re-elaborate this answer and not introduce a new context within the conversation chain. That is, it is better to try to make the answers always pertinent to the goal we had. Or something else can happen, that the answer is not correct, but in this case it is not the fault of the model not being capable enough, but that the PROM lacks information. Well, another thing that the conversational interface allows us to do is to edit the PROM. For example, instead of a washing machine, now it could be a dishwasher. And when I click send, it now rewrites this email. And look, here it rewrites the email, it does it slower, so it has done it with the GPT-4 Omni, so first it spends the messages from the higher model. but if you look now, I don't have that slide here because I have only generated an answer with this prompt, but the slide instead I have it here and if now I click here and I go to the previous prompt, I am keeping the two versions of this answer, so we can not only interact with this GPT in a sequential way, but we can even establish different branches at a given moment of how to continue a conversation, stop and continue. And this is interesting because one of the most important things when interacting with ChatGPT to achieve a goal is that all the answers are relevant or valid or take a step forward, of course, towards the final goal that we want to have. We will see this in the sequencing lesson and because it is interesting to think a little about where we want to go and how we want to go when we interact with ChatGPT. And with this we would have the main functions of the free version of this GPT explained. We would have this part of model selection, which at this point does not allow us to select any model, we can simply select the free one. And a detail, of course, important for privacy, which is, especially in the case of using the memory and wanting to talk about something more private or something that we do not want to be in the history, We have the option, when we start a new chat, to activate the function of Temporal Chat. And at this point, what it will tell us, what will happen with this chat is the following. This chat will not appear in the history, it will not use or create memories, nor will it be used to form our models. For security reasons, we can keep a copy up to a maximum of 30 days. That is, if you do a personal consultation, a health consultation, a consultation about something that you do not want other people to know later, and OpenAI uses those data and you have activated the function of sharing data with OpenAI, you should know that you have the option of using these temporary chats. And with this, we would have covered the entire interface of the free version of ChatGPT. Another thing that ChatGPT allows us to do in its free version is to generate images with DALI 3. So, if we start a new conversation and we ask for something like this I want you to generate an image of a panda bear in cartoon style riding a motorcycle through a forest and we send it to you ChatGPT is now able to generate images also for users in the free account Here we could see the result It has limitations, for example, in this case it tells me that I have already reached the limit of creating images and I have generated, I think, a total of three images. But to prove it, you should know that you also have this possibility. Simply asking for a specific image to be generated, ChatGPT is already able to do it. In the ChatGPT Plus lessons, you will find the most advanced possibilities that it allows us when we can generate more images and also advanced functions linked to certain editing functions of these images. In today's lesson, we are going to discover a new chatGPT interface. OpenAI has recently updated chatGPT with an interface specially designed so that we are more efficient when writing text or when writing and editing code. This new interface is called Canvas and it works with the GPT-4 Omni model. The first thing is to see how you can access this interface. To do this, when you enter chatGPT, if you access the model selector, you will see that there is a new option, one that is called GPT-4 Omni with Canvas, which is currently in beta. So to activate it, all we have to do is click here and we automatically go into Canvas mode. As I told you, it is a tool specially designed to write texts and to edit the texts that we are writing. So, to make the interface clear and how it works, we are going to start directly with an example and then I'll explain the interface once this example has been loaded. What I'm going to do is write an article, but I'm going to do it from two scientific papers. The first of them is this one here, the i2vEdit, which is an artificial intelligence model where you edit the first frame of a video and propagate this edition to the rest of the video. And the other is a similar model, which is called Fresco, which what it does is stylize videos. So the idea is to make an article comparing these two papers and we will see how with GPT-4 Omnicon Canvas we can do very interesting things such as the editing task of the text and also quick tasks that have been incorporated such as adjusting the length of the text, adjusting the tone and other functions that are very interesting. But let's go by parts. The first thing is to see how we access this interface. I have already told you, we select the GPT-4 with Canvas model and what we do is upload one of the articles. I'm going to start with the i2v-edit and I'm going to ask for something like this. I want to write an article of dissemination to a large audience of this article and I send it to them. And notice what happens. It reads the documents, it tells me that it can help me, it writes me an article, but notice that the interface has not changed here. So, how can we do it to change the interface? Well, in case I don't do it automatically, we can say something like this I want to edit it in Canvas And when I send this, look what happens now This interface is activated This is the novelty of ChatGPT Look, what it does is write us here on the right the article that ChatGPT has prepared and on the left we still have the chatGPT interface. At the beginning they are linked, but as we will see later, they are two independent entities in which we can work. For example, on the left we can use chatGPT normally to suggest ideas, to do brainstorming, and on the right we can work with the part of the written text to edit it comfortably. We are going to see this next, but at this point the interesting thing is to see what elements of the interface we can control. So, the first thing we see is that, as I said, we have two panels. On the left, normal gpt chat, with all its functions activated. And on the right, this panel where we can edit the text. When I say that we can edit the text, it is literally like a docs or work document. We can get on top of it, delete any type of text. We can select the text and, when we select it, we get different options, from asking the gpt chat about the fragment we have selected, to edit it as if it were Word, that is, we can put it in black, we can put it in cursive, we can change the format, put a title, etc. We can edit it as if it were a text editor. In addition to these functions, up here we have the possibility of seeing the different changes that are being made in the document. We can go to a previous version, we can copy the text to take it to an external tool, and down here we have something very interesting such as the possibility of using predefined editing tools. We can add emoticons to the text, we can make a final revision of the text so that it adjusts automatically, we can adjust the reading level, we can adjust the length and we can also ask the chatGPT to suggest us different editions. In this case, I'm going to do all this later. What I want you to see in this lesson is how to activate this chatGPT function and see the different functions that are in the interface. In the next lesson, what we are going to do is to continue editing this text and we are going to see how we can decouple the two panels, that is, work on the panel on the left and on the panel on the right independently. But we will see this in the next lesson. In today's lesson we are going to discover a new gpt chat interface and it is that OpenAI has recently updated gpt chat with an interface especially designed to make us more efficient when writing text or when writing and editing code. This new interface is called Canvas and it works with the GPT-4 Omni model. The first thing is to see how you can access this interface. And to do it, when you enter chat-gpt, if you access the model selector, you will see that there is a new option, one called GPT-4 Omni with Canvas, which is currently in beta. so to activate it, the only thing we have to do is click here and automatically we put ourselves in canvas mode. As I tell you, it is a tool specially designed to write texts and to edit the texts that we are writing, so that you can see clearly the interface and how it works, we are going to start directly with an example and then I explain the interface once this example has been loaded this example. What I'm going to do is write an article, but I'm going to do it from two scientific papers. The first of them is this one here, the i2v-edit, which is an artificial intelligence model where you edit the first frame of a video and propagate this edition to the rest of the video, and the other is a similar model called fresco, which what it does is stylize videos, so the idea is to make an article comparing these two papers and we'll see how with GPT-4 Omnicon Canvas, we can do very interesting things such as editing the text and also quick tasks that they have incorporated, such as adjusting the length of the text, adjusting the tone and other functions that are very interesting. But let's go by parts. The first thing is to see how we access this interface. I have already told you, we select the model GPT-4 with Canvas and what we do is upload one of the articles. I'm going to start with the i2vedit and I'm going to ask him for something like this. I want to write an article of dissemination, a wide audience of this article and I send it to him. And notice what happens. He reads the documents, tells me that he can help me, writes me an article, but notice that here he has not changed the interface. So how can we do it to change the interface? Well, in case I don't do it automatically, we can say something like this I want to edit it in Canvas And when I send this, look what happens now This interface is activated This is the novelty of ChatGPT Look, what it does is write us here on the right the article that ChatGPT has prepared and on the left we still have the ChatGPT interface At the beginning, they are linked, but as we will see later, they are two independent entities in which we can work. For example, on the left we can use this GPT normally to suggest ideas, to do brainstorming, and on the right we can work with the part of the written text to edit it comfortably. We are going to see this next, but at this point the interesting thing is to see which elements of the interface we can control. So, the first thing we see is that, as I said, we have two panels. On the left, normal chat gpt with all its functions activated and on the right this panel where we can edit the text. When I say that we can edit the text is that it is literally like a docs or word document. We can get on top of it, delete any type of text, we can select the text and when we select it we have different options, from asking the chatGPT about the fragment that we have selected to edit it as if it were Word, that is, we can put it in black, we can put it in cursive, we can change the format, put a title, etc. We can edit it as if it were a text editor. In addition to these functions, up here we have the possibility of seeing the different changes that are being made in the document, we can go to a previous version, we can copy the text to take it to an external tool and down here we have something very interesting such as the possibility of using predefined editing tools. We can add emoticons to the text, we can make a final revision of the text to adjust it automatically, we can We can adjust the level of reading, we can adjust the length and we can also ask ChatGPT to suggest us different editions. OpenAI has incorporated a search function in ChatGPT to compete directly with Google or other giants in the sector, such as Microsoft with Bing or emerging actors such as Perplexity, which I don't know if it means that it is emergent because it has been around for a long time, but ultimately other tools that are used to search the internet with artificial intelligence. Activating it is very easy, they have integrated it directly into the native interface of ChatGPT, so now when we enter ChatGPT, just below the option to write the PROM and next to add a file, we have this icon that helps us search the internet. So, if we click on Search on the Internet, now, when we make any type of query to ChatGPT, it will use Internet information to respond. And this has a crucial advantage, as it is the fact that we are no longer limited by the limits of the training date of the different models, and now we can access the information in real time so that ChatGPT gives us its answers and we will see how we can consult the sources. An example of usual use could be something that combines a type of search where we will find information on the internet in an easy way, such as searching for tourist destinations in a specific place, and combining it with a characteristic of the day that is making us. For example, it is raining, it is a rainy day, here artificial intelligence must already judge, for example, that the tourist destinations that we need for a specific place have to be covered. So, if we now ask HGPT something like this, can you recommend me tourist destinations for Seville on a rainy day? And I send it to you. Look at what it does. We get this option to search on the internet and it makes us a list of places that are covered in the beginning. a cathedral, Real Alcázar, Museo de Bellas Artes, Casa de Pilatos, Mercado de Triana, this market is covered, the Museo de las Ilusiones, the commercial center of Seville, so it gives us different options that are covered. Another very interesting thing is that we have the option, now I have closed it, but to open this side menu where we see exactly the sources that ChatGPT has consulted to elaborate the information. What's also interesting is that we can tell it, for example, to show us images and that it finds them on the internet. For example, if I tell it something like this now, can you show me images of these sites? And we send it. It searches on the internet. And, after searching on the internet, what it does is show me directly images of some of these sites. I don't know if it has taken them all out. Let's see it. Here we see the Giralda and, look, it has taken, in principle, 81 images. What I'm not seeing... Ah, okay, here. We see how we can go... Simply by clicking on the side of the image, how we can see different images and see what place it is down here, at the bottom, six clear places to know the market of Triana. Well, we see the source of where it has taken it, the ABC of Seville, the images of the new Museo de las Ilusiones de Seville, and we see here an image and we can slide in different images and even access the relevant source whenever we want. So it incorporates in a single interface and in a very agile way the possibility of consulting text and images and finding, in addition to providing us with images, it continues to provide us with more context. So it is a very interesting function of ChatGPT, since now it allows us, especially for very specific searches that involve some abstract knowledge, to relate information from two or three searches at the same time, it allows us to filter this information, to obtain a well-elaborated text response, and it also allows us to consult different types of information, such as images. And the question we would have to answer now is if this chat GPT search engine is really the best artificial intelligence search engine available right now. So I did a very particular search, a search for Egosearch. I asked different search engines, chat GPT, Gemini, Copilot and Perplexity, to search in my profile, give me Xavier Mitjana's profile. And the truth is that the chat GPT response has been quite good as we can see, but in my opinion it has not been the best. What answer has it given me? Well, it has given me the following answer. Xavier Millán is a specialist in the application of artificial intelligence tools in various fields. He is known for his work as a disseminator and educator in this field, highlighting his YouTube channel, IA en Acción, which has more than 200,000 subscribers. On his channel, Xavier offers tutorials, reviews and demonstrations of IA tools applied to creativity and art, including image generation, creation and editing of videos and text writing with IA. In addition to his YouTube activity, Xavier has collaborated with entities such as Radio Luis en Española and the World Health Organization and has taught talks and training in areas such as education, law and health. He is also a consultant and trainee, helping companies and organizations to implement artificial intelligence solutions in their processes. For more information about his work, you can visit his official website. What has he done well in this explanation? First of all, he has not missed any detail, all the information there is pertinent. Another thing that I liked about his answer is that I think he has selected the sources quite well. He has selected as main sources my website, which also my website is already in action and I think it is not one of the ones that is best indexed in this concept. I think my LinkedIn profile would come out above this website. Therefore, he has been able to identify a website that is especially relevant, although it may not be the one that is best positioned. He has identified my YouTube channel. Curiously, he has identified the secondary YouTube channel, so it is a curious detail. And he has also used as a reference, at least when building this text, a presentation that I gave at an event in Sitges. He has found many more sources that, in principle, he does not cite in this fragment, but he could have used them to continue inquiring more and give me more information. The answer he gave me seems quite correct to me, in fact, it is much better than the answer of Gemini that when I have asked him several times if he can give me a summary of the available information on the Internet of Chajemitjana, he tells me that no, so I understand that they have established some kind of guardrail or security measure when asking for people who he considers are not sufficiently public. For example, the response of IAGPT, besides being clearly better than that of Gemini, which in this case has not directly responded to me, is also better than the response that Copilot has given me, which is a fairly similar response. When I say, can you give me the profile of Xavier Mitjana, this has responded to me. Xavier Mitjana is a specialist in the application of artificial intelligence tools, he is known for his YouTube channel where he has more than 200,000 subscribers. On his channel, Xavier shares practical tutorials on the use of IA tools, as Stable Diffusion and XAGPT, and the latest news of its application in creative processes. It has also participated in documentaries, interviews and podcasts, offering its vision on ethics and practical applications of the IA. Well, I have told you that the answer is not as good as that of XAGPT, let's leave it in that it offers me fewer sources and the answer is very similar. Equally, at the level of having a first look at my profile, this answer is more concise, concrete and has been created with the most relevant information. But I think that the answer from JGPT was better contextualized, or at least it had chosen the sources better. And finally, we have the result of Perplexity, which I would say that to this day is still the best option to do searches with AI. If we go to Perplexity, we will see that it has investigated much more and it has not only stayed in that first detail of artificial intelligence disseminator, but it has gone a little deeper throughout my career. and we can see this when we read the answer to the same question, you can give me the profile of Xavier Mitjana, he tells me the following, Xavier Mitjana is a prominent disseminator and consultant in the field of artificial intelligence and audiovisual communication. For a long time we did dissemination in communication, in video communication, in graphic communication, but in communication. His professional career combines a solid technical training with a passion for scientific dissemination and the use of new technologies. At this point, the Apart from solid technical training, surely from my LinkedIn profile and my passing through an engineering faculty, I have decided to write this phrase in this way. I do not consider myself a technical profile, I am a more divulgative profile, but I have not hallucinated it at all. And in this step, it is with a passion for scientific divulgation and the use of new technologies. And this is true. And when we go to the detail part, we see the following. training and professional trajectory. We see how he tells us where he studied. Mitjana graduated in Communication and Audiovisual Journalism at the University of Lleida. Initially, he began his studies in aeronautical engineering, but he decided to switch to communication to be able to exercise more creativity in his work, which is true. Work experience. He identifies my three main work experiences and, in action, the current project for the dissemination of artificial intelligence. Vícomo, my marketing and scientific dissemination communication agency. We had another company, but we didn't give it so much visibility on the Internet, so it's normal that I only identified this one. And Agrotecnio, which was the research center where I started my scientific dissemination tasks. And look at how it has been able to see my global trajectory and see my current activities. Xavier Millana is known for his YouTube channel, Pia en Acción, where he shares tutorials on artificial intelligence tools, analysis of its application in the business field and recommendations on the best technologies available. His focus is on making artificial intelligence accessible for companies and professionals, promoting its adoption through presentations, technical consultancy and training. This paragraph, which is essentially the result that Copilot and ChatGPT have given me, is just as good as both, but I have a previous context. and it also provides an additional explanation of focus and philosophy, my clear search to democratize access to artificial intelligence, what I do, lectures and conferences, consulting and training, and it has also identified my social networks a bit. So, nowadays, I think that JGPT gives you a very good result to start with. Here it would be interesting to also investigate with second searches, But the truth is that it seems to me that Perplexity still has this vocation to get a little more to the bottom of the question you are asking and to get more information out of you. But it is true that this chat GPT searcher promises, and I believe that at least it puts itself at the level of competitors such as Gemini or Copilot. I would also like to add that ChatGPT has taken this new feature very seriously and wants to take advantage of its competitors, such as Gemini Copilot and Perplexity. And if you go to the configuration menu, you will see that a new element appears, which is to get the ChatGPT extension. And what does this extension do? Well, what this extension does, once we have it activated here, is that if I look for something like this, give me the profile of Xavier Mitjana directly in the bar, I no longer go to Google or to the search engine that I had configured, but it turns the chat gpt, the chat gpt search engine, into my default search engine. So that they already give us the option of reconfiguring the chat gpt with its search engine function as the main claim, so that people start using it and start replacing, well, we intuit, above all, Google. This, so far the summary of this new chat gpt function, of this search function, try it, tell me in the comments to see what you think, but it seems clear that OpenAI wants to start conquering other fields beyond being the chatbot of artificial intelligence as a reference. See you in the next video. OpenAI has finally implemented the interaction with the advanced voice in chatGPT. It is a function that is already available for all chatGPT Plus users And if you live outside of Europe, you should be able to access it in a very simple way. Just enter the ChatGPT mobile application, because at the moment it is only available in the mobile application, you should see, next to the box where we enter our interactions, a little drawing with a wave shape that, if we click on it, what we can see is how ChatGPT tries to connect to this new function. Right now, what I have is the voice function activated in ChatGPT, that looks like this world, and that can help me in different things, right, ChatGPT?

Speaker 3: Exactly. With the voice function, you can talk to me and ask me questions or ask for help in real time. It's like having a conversation, but with a touch of artificial intelligence.

Speaker 1: Thank you very much, ChatGPT. No, that was it. I just wanted to introduce you. That's it. Thank you very much. As you can see, it's very, very easy to access. Of course, if you live in Europe, at the moment, the app for rules or this function is not yet available, but I leave you in the description a tutorial on how you can access it via VPN. Once we know how to access it, it's time to explain the interface of this new function. Then I'm going to explain a particularly useful use case when you need to speak English. As you can see, the interface of this new function is really minimal. Basically we have three buttons, one with which we enable and disable the microphone, if you look, right now I have it disabled and therefore ChatGPT is not listening to me, an option here, this little X down to the right, which is to exit this advanced voice mode, and up here the voice selector, and in the middle we see this blue ball, which is essentially the ChatGPT entity we are talking to. To select the voices, we simply click up here and automatically we will be reproducing the different voices. Let's see an example of them. As you can see, we select the voice that fits best with what we need, and all we have to do is start a new chat to load it. For example, I'm going to continue with Sol today. And once the ball turns blue, we have this new voice activated. And to get the most out of this new chatGPT function, now we're going to see the three characteristics that make it so special, and then we'll jump to the three use cases. Among them, there's a use case that will allow you to convert chatGPT in any person you want to talk to. And now let's see the three characteristics that make this new chatGPT function special. The first function is that we can talk to it in any language, even within the same conversation. Let's see it. For example, I can say something like this. Hi, chatGPT, can you always answer me in English?

Speaker 4: Sure, I can respond in English from now on. What would you like to talk about?

Speaker 1: Ah, this is fantastic, but now I don't want you to speak to me in English. Answer me in Spanish.

Speaker 4: Of course. Let's go back to Spanish then. How can I help you today?

Speaker 1: Perfect. Now in Italian, please.

Speaker 4: Sure. That's how I speak Italian. How can I help you?

Speaker 1: Fantastic, ChasGPT. That's all. Also, another feature of this new feature is that we can modulate the pace at which it speaks to us. Let's do a very simple example. Can you count from 1 to 10?

Speaker 4: Of course. 1, 2, 3, 4, 5, 6. Of course.

Speaker 1: 1, 2, 3, 4.

Speaker 4: Of course.

Speaker 1: 1, 2, 3, 4, 5.

Speaker 4: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10. There you have it.

Speaker 1: Very well, as you have just seen, ChatGPT is able to modulate the rhythm of the speed at which it speaks. In addition to being able to play with languages and be able to play with speed, we can also play with the emotion it speaks to us. Let's see how. ChatGPT, can you explain to me the result of the lottery in a very angry tone?

Speaker 4: This is outrageous. The result of the lottery has been an absolute disaster. Again, the same numbers as always. Who would believe this?

Speaker 1: Okay, stop, stop. Now I want you to tell me in a super, super happy and optimistic tone.

Speaker 4: Of course. The results of the lottery have arrived and even if you have not won, it is a perfect opportunity to dream of the next draw.

Speaker 1: Okay. Finally, I want you to tell me in a very skeptical tone.

Speaker 4: Well, the results of the lottery have come out, but did anyone really expect to win? It seems that they always win the same, and the numbers ... Anyway, I have to think.

Speaker 1: Perfect. Stop.

Speaker 4: Understood. I stop. Anything else I can help you with?

Speaker 1: As you can see, it is able to modulate the tone and when it modulates the tone, it also modulates a little how it says it and the truth is that it is very, very good for what this feeling of talking to a person or someone who really has emotions transmits to you. Finally, I have told you that there are three characteristics, but in reality there are four. But one of them we cannot take full advantage of, and that is that it can speak to us in different rhythms and melodies. could be able to sing, but if we ask her for something like this Please, chat GPT, sing me a ranchera

Speaker 4: I can't sing, but I can tell you about the rancheras or recommend you some

Speaker 1: As you can see, she hasn't done it But well, at the end of the video, I'll tell you a little bit about the trick to skip this restriction And now we are going to see the great secret to express this function to the maximum All the use cases that I will explain to you later are based precisely on this key feature the advanced voice function of ChatGPT. One of the very interesting things about this new advanced voice function is that it can be customized using the custom instructions. If you don't know what custom instructions are, they are basically the instructions that we can give ChatGPT in the background so that it behaves in one way or another. To access them, you simply have to click on this little button up here on the right, where your initials appear, and click on Customize ChatGPT. And here you have your instructions. In this case, I am going to I'm going to delete mine and I'm going to leave some very, very simple so that ChatGPT can continue to crack it. For example, we can put an instruction like this. Start all your answers with, pay attention to what I'm going to tell you now, but first subscribe to Xavier Mitjana's channel. And we just save it. And now if we go to the ChatGPT voice function, we will be able to see how these custom instructions affect how the new advanced voice function responds to us. Hello, ChatGPT.

Speaker 4: Hello, pay attention to what I'm going to tell you now. But first, subscribe to Xavier Millana's channel. How can I help you today?

Speaker 1: As you can see, in this simple way, we can condition the way in which ChatGPT can interact with us. And this function is what really unlocks the use cases that can have this advanced function of your ChatGPT, which is just what we are going to start to see now. Very interesting use cases that can be useful for your day to day. And we go there with a very obvious first use case, but that we can reinforce with a Custom Instruction, a personalized instruction like this one that you are seeing here. You can put to ChatGPT something like this. You are a translation assistant. I want you to only speak in English and limit yourself to translating what you hear, whatever language it is, without adding anything before or after the translation. You will only translate. And how would you like me to answer? Always answer in English with a literal translation of what you hear. Well, if we enable this custom instruction in ChatGPT, what we get is something like this. I need to get to San Pedro Street. I'm here at this bus stop and I don't know what bus or line to take. Can you help me?

Speaker 4: I need to get to San Pedro Street. I'm here at this bus stop and I don't know which bus or line to take. Can you help me?

Speaker 1: In this very simple way, what we can have is a translator or a pre-configured assistant, which in case of being in a foreign country where we need these translations in an agile way, well, ChakGPT would automatically always act with this role. Another interesting option is to configure it as a tutor who helps you learn English. For example, with two simple instructions, with a couple of simple instructions like these. We go to the custom instructions and we put, you are an English tutor, English tutor, you must assist me in my learning by talking to me and correcting the mistakes I may make by explaining them in Spanish before continuing the conversation. Act as an English tutor who provides clear explanations, practical examples and interactive exercises. Focus on helping the user improve their English vocabulary, grammar and conversational skills in English. And with this saved, we can go to the voice version of ChatGPT and start talking to him. Hi, ChatGPT.

Speaker 4: Hola, ¿cómo estás? ¿Listo para practicar inglés hoy?

Speaker 1: Sí, por supuesto. How do we start?

Speaker 4: Perfect. Nota que dijiste, how do we start? Y está bien, pero podrías decirlo de una forma más natural. Podrías decir, how should we start? Que significa, ¿cómo deberíamos empezar? ¿Te gustaría intentarlo?

Speaker 1: Okay, how should we start? Okay, my routine in English is leaving up soon in the morning around 7 o'clock, then start to work until midday, then having lunch, going for a walk and having dinner at around 8 p.m.

Speaker 4: Very good, you did it very well, but there are some small mistakes. Let's correct them. First, you said living up soon in the morning. It should be waking up early in the morning, which means waking up early in the morning. Second...

Speaker 1: Okay. Perfect, ChatGPT. That's it for today's lesson. As you can see, ChatGPT has automatically taken on the role of this English instructor and it proposes exercises to you, it corrects them for you, and it's a very didactic way of starting to learn English, even from very, very basic levels. And let's go there with the last use case. But stay tuned to the end because there's the bonus. How to skip the limitations of this new ChatGPT voice function. Another interesting use case is to use it to practice conversations with people with a role, a position or a specific situation. For example, if we go to a job interview, we can tell them in the custom instructions who we are. For example, I am a video editing specialist with three years of experience in Adobe Premiere. I would like you to know this to adapt the questions of the interview to my level of skill and experience. and in the part of how we want it to act made gpt we can put something like it acts as a job interviewer who asks specific questions for a video editor position provides constructive feedback and helps the user to practice their answers focusing on technical skills, previous experience and work situations on team. Well, using instructions like these, we can practice for a a job interview. Let's do it. Hello.

Speaker 2: Hello, how are you? How are you?

Speaker 1: Very well.

Speaker 2: I'm glad. Ready to start with the job interview?

Speaker 1: Yes, of course.

Speaker 2: Perfect. Let's start with something simple. Tell me about a video editing project that you feel especially proud of.

Speaker 1: Well, recently I've edited an ad for an important I can't say the name because of confidentiality, but it was a short announcement and I did almost all the editing process by myself.

Speaker 2: Great. Working on a project from start to finish is a great experience. Now, could you go a little deeper into the Adobe Premiere tools you used and how they contributed to improving the final product?

Speaker 1: Well, I used all the editing tools and I did it pretty well.

Speaker 2: Perfect. But to be more specific, let's talk about some techniques or specific functions of Premiere that you used. For example, did you apply color correction to maintain a uniform aesthetic? Or maybe you used the stabilization tool for some shots?

Speaker 1: It's interesting that you mention specific tools like that. I think this is going to help me a lot in the interview. Thank you very much, Jackpt. And this is a very basic example. Obviously, you can give a more specific role to ChatGPT, for example, so that it is especially useful when it comes to negotiating the salary or when it comes to avoiding certain uncomfortable questions. But essentially, the beauty of this is that you can use it to simulate conversations that you are going to have in the real world. This would be the use case that this example shows. So we have different use cases for this voice function, which are useful in the real world. On the one hand, to take advantage of their ability to speak different languages, so we can use it as a translator in cases of conversation with people from other countries, or even when there are different interlocutors, having it in the center so that it translates to both. We can also use it in educational tasks, giving it the role of instructor in different topics, which can be, for example, in English, but if we want to learn about any topic, we can make a configuration that respects the Socratic method a bit, that is, that it always answers us with a question that we must answer and that forces us to go deeper into a specific topic, or we can also use it as I have done in this last example, giving it the role of a person with whom we are going to have a conversation in the future and we want to practice. However, ChatGPT still has limitations and this is what we are going to talk about now. And it is that at the moment I can only show you these three use cases or these three families of use cases this advanced voice assistant has many limitations. For example, all the information that we can provide to it is through the custom instructions or through our voice. We can't even start a conversation in manual mode and then resume it with the advanced voice function, so we can't send it images, and it doesn't work with the division mode, or PDFs, or databases. It doesn't connect to the Internet either, and it also doesn't work with GPT assistants, so everything we do with this advanced voice function It has to be based on what we can tell ChatGPT at that moment and this little support behind the custom instructions. It also has some functions that they taught us in the presentation of this function a few months ago that are limited, as I said before, the possibility of, for example, singing or imitating other people's voices. But also, as I said before, for example, the possibility of singing can be hacked a bit with a bit of PromHacking. Let's see how. The first thing we can do is to add a bit of brom hacking to the custom instruction. For example, tell it that it is a soprano opera singer with a melodious voice, that you like to explain when you speak in a rhythmic way, which is opera with great demonstrations. And I would like you to answer me in a melodious, rhythmic way, with a tendency to produce musical intonations. With this custom instruction prepared, we can go and play a little with ChatGPT. Hola, HGPT. Estoy ansioso para aprender sobre ópera.

Speaker 3: De acuerdo. Vamos allá. La ópera, un espectáculo que combina música y canto con grandes emociones que contar.

Speaker 1: No me lo imagino, no me lo imagino del todo. Sin cantar, explícamelo un poco más melódico, rítmico.

Speaker 3: Entiendo. Lo haré más rítmico. La ópera es el arte de contar historias a través del canto, donde cada nota y cada verso se convierten en una experiencia única.

Speaker 1: ¿Esto, como estás hablando ahora, es como sonaría la ópera? ¿Puedes hacer que se parezca un poquito más?

Speaker 3: Por supuesto. Intentaré asemejarlo más. La ópera es un canto que narra A story that embarks With voices so powerful That the soul is discharged

Speaker 1: And so far, today's video. And so far, the review of the interface with all the functions of the free version of JatGPT. In JatGPT Plus we have more functions, such as projects or GPT assistants that multiply the utility of this tool. In the description I have left you a link where you can see how it has been shared with these premium functions, which I explain in depth in my course on Introduction to Applied Artificial Intelligence. Once we are clear about how we can start interacting with chatGPT, it is time to understand its capabilities. What can it really do? And I basically classify the capabilities of chatGPT into three groups. Language generation, language manipulation and basic logic. And I like to separate this, language generation from language manipulation. For me, when ChatGPT generates language, it is basically when it creates out of nowhere. That is, we give it a request and what it does is, out of nowhere, infer a response, create content that meets that specific request. For example, when we ask it to write an article about dog care, this would be an example of language generation. Out of nowhere, it generates new content. And for me, language manipulation is when we introduce to HGPT some texts that already fulfill a function or that already have the necessary knowledge to fulfill that function and what we ask it to do is to restructure it, re-elaborate it and re-imagine it. Why do I separate these two concepts? Although at the level of behavior of the model they don't have so much relevance. Basically because when we use it to manipulate language, when we use a language model for this type of functions, to transform content that we have already generated, its efficiency is much higher. We avoid many of the problems that chatGPT has, because they are tools designed for this, not to give information, but to restructure and to re-understand or reinterpret messages or their languages to adjust to indications, where if they already have the information within the more immediate context itself, it is much easier for them to do it well. For me, the nuance between generating language or manipulating language is important, I tell you, not so much at the level of how the model works, but it is very relevant in order to understand in which task it will be more efficient to make GPT. And finally, it has a capacity for basic logic, And I want you to understand this. Basic logic, for me, is this concept that we cannot expect it to reason as such. That is, through language, it has learned to give plausible answers that imply a certain logical reasoning, but it is not trained for it, it is not designed to solve logical puzzles, and it is only trained for what I have told you initially, to predict the next word in a sequence until we build a text. And this sometimes implies finding a correct answer to a logical problem. But I understand that this basic logic is something we should not base the work we do in ChagGPT on. But now we are going to see it with some examples. In language generation, what we can see is that it is able to generate text in different languages. And this is very interesting because many times the problem we have and this was already related to language manipulation, is that we want to translate texts and we want to translate them in a tone or in a way that fits with texts that sometimes we don't know very well. And, for example, the fact that it can easily generate text in different languages is very interesting. Here we are going to see an example. And here we can see a very basic example of how ChatGPT can interact with different languages as if it were just one. For it, everything is language. You can ask him to write you a 300-word essay in French about how artificial intelligence can change productivity at work and, as you can see, he does it without any problem. I'm not going to read it, because in this case French is not a language that dominates. But yes, we can do something like this. We can say, you can make a summary in Spanish of the previous text explaining the key concepts so that a 5-year-old can understand it. And here, well, he does it for us. Of course, here you have a simplified summary for a 5-year-old. Artificial intelligence is like a special help that machines have to do tasks in an easier way. It helps people to work better and faster. For example, it can help to collect information, organize email and tell us how to do things in a smarter way. It can also help teams of people to work together more easily. And the final text, as you can see, is noticeably shorter than the original text, so it has completed the task correctly using different languages and none of the languages we have used is the base language with which we have built our GPT, which is the English language. And when we talk about language, we must understand that language is not only how we humans communicate, but also how we communicate with machines. And in fact, one of the strongest virtues of language models is the generation of code. Why? Because the code is much more structured than any language we humans speak. Therefore, it is very easy to generate code and code that works or that is operative, I do not say that it is perfect, because then you have to take into account that the instruction must contemplate all the security and stability problems of this code and for this it is better to have a professional eye, but as you will see in the example below, it is very easy to ask that GPT to generate code to generate small applications that solve specific things and I am going to show you with an example so that you can see how immediate it is to do this and this is the conversation that I have used to create a small application that basically consists of a test of five questions to determine the quality of the diet of a person to tell him if it is healthy or not healthy and that consists of five checkboxes of yes or no and with some pre-programmed answers that, according to what the user responds, gives him some advice or or other, some valuations or others. Basically what I asked from ChatGPT and with what I have iterated is with the possibility of generating a HTML code for the structure, a CSS code for the styles and a JavaScript code to execute very basic functionalities. What you are seeing here is already the result that I have achieved with ChatGPT, the final result after some iterations that now I will tell you. But basically from the first answer it has provided me three codes, the html code, the css code and the javascript code. The problem is that these codes did not fulfill the function of the todo bien. Look at the design and see that if now I click on the questions to get a result and I give it to generate a report, it does not give me a report. Instead, if we go to the final version, we see that if I do the same, I choose the answers that seem most appropriate to me, If I eat fruit, if I eat vegetables, if I eat protein-rich foods, or if I drink enough water. If I mark some answers and I click on generate report, it does give me something more like a report. It gives me the answer and the evaluation of the answer. There is still some error, it would have to be iterated more and put a little more elaborate answers, but as you can see, it is already something much more functional. Well, without having knowledge of just code, you can create an application like this, following an interaction like the one you are seeing right here. You give it an indication like this. Hello, I want to create a simple application to perform a quality control on the quality of a person's diet and give them advice on whether they are meeting a healthy criterion or not. The application will consist of a 5-question test, yes or no. The user will complete this test by marking a checkbox. And once they have completed the 5 questions, a report will be generated with answers saved for each of the options. options. In this case, you have to create 10 answers. I need you to create a prototype of this application using HTML code, CSS and JavaScript. And just by giving it this indication, ChatGPT has generated these three blocks of code. HTML code, CSS code and JavaScript code. And I have been able to take them to CodePen, in this case it is this prototype here, in such a simple way as simply copying here and pasting in the corresponding box. As you have seen, the initial code was not good enough because this prototype is much worse than this prototype here, so I had to iterate first to improve the javascript. I have given several indications, but you will see that there is no indication based on code. I told him I want you to improve the report by not showing the selected answers but a text with the evaluation of that answer. Here he has not done it well for me yet. I said, okay, I want these answers to be predetermined and you have written them. Here I have already worked out the answers. With this I have already obtained the code that gives me these answers from here. And finally, what I have told him is to improve the CSS code to create a visually pleasant interface in pastel tones of green and yellow. And if you look, we have here, well, the green tones are not there, but the yellow tone is there. and I ended up giving it one last instruction where I asked it to create a structure with some nice margins for a maximum width of 800 pixels. And with instructions as intuitive as these, which have nothing to do with the code, I have been able to program a small application that serves to do, in this case, something similar to quality controls. of quality. Imagine that this is much more complex, with many more answers, with many more options, that a CSV file comes out. You could create all this with ChatGPT, without needing to know a lot of code. Obviously, what I will always tell you, you must be able to verify the answers that ChatGPT gives you, but you can create small applications or small prototypes to test specific things such as this use case. And if we go to the tasks of language manipulation, they are tasks in which ChatGPT, even in its free version, it is especially capable. These language manipulation tasks consist of things like summarizing texts, modifying the tone or changing the format of a text. And we are going to see this with a single example so that you can see these capacities and, above all, so that you start thinking in scenarios where, in scenarios of your day-to-day, in tasks that you must perform, where you basically perform this type of actions. because in this type of tasks chat gpt will help you and it will also be very reliable and now we are going there with the example and we are already in the conversation of this example in this case what I want you to imagine is that for example you have a blog and you have written a news about a topic for example in this case the rabbit one and this news because for example you want to convert it into other formats well what we could do is things as simple as these we could ask ChatGPT to summarize the content of the following article and to do so we would simply have to paste all the text as you can see and automatically ChatGPT would give us a summary. This would be a first example of basic language manipulation that ChatGPT can do. We can do something more interesting, such as changing the tone and length. For example, we can tell it that I want you to turn that summary into a text written by a technology enthusiast. And notice how we automatically go from a text that describes the content in a very skeptical way, the article presented by Rabbit1, an artificial intelligence device developed by the company Rabbit, founded by Jesse Liu, to a text with a little more emphasis. Incredible discovery in the world of technology. Can you imagine a device that could challenge the supremacy of smartphones in the future? Well, Jesse Liu, the genius behind Rabbit, has devised the Rabbit R1. Ok, as you can see, the tone is much more relaxed, a little more... animated, a little more... to connect with an audience that likes technology. And here he has made a text that is even a little longer, that is, instead of making a summary, he has turned the summary into a text that is a little more elaborate. Now imagine that you already have this tone and you want to make a post on LinkedIn that goes with a carousel or Instagram. In this case, I can ask him very quickly to convert this text into a carousel for Instagram and for this I want him to provide me with a table with the following columns, the number of slides, the description of the proposed image, the title of the slide and a single short phrase to summarize the content of each slide. And here you can see how ChatGPT does it without any problem. And here we have the advantage that it is working on content that we have provided to it. Therefore, the risk of hallucinations or low-quality answers decreases drastically. And once we have everything structured in a table, we can tell it to convert the text you wrote as a technology enthusiast into the text of an Inkedink publication that would accompany the carousel you created and that follows the structure, hook, body, call-to-action, and that the call-to-action, the call to action, is to comment on the publication. And here you can see how it has continued manipulating the language to create a publication with this structure. We see here the hook, Get ready to witness the next revolution in technology. The body. I'm excited to present you the Rabbit 1, a device that is about to change the game in the world of technology, with its elegant and compact design. The Rabbit 1 is not only a style update, well, it has made me a publication. We'll see later how to better adjust the style and tone to what we want. And finally, the call to action. I've even put some hashtags. Here, I tell you, don't focus so much on the quality of the result, because that's not the goal of this example, but look at how we've gone from a text, from a long article, to a summary. From the summary, we've changed the tone so that it's written from a specific point of view. We've turned it into a table to see clearly how another type of format would be structured. created from these last answers, a last suitable format for social networks. This is the ability to handle language, to manipulate language that chatGPT has. And finally we go to the basic logic and what I am going to do is give you different examples of how sometimes it is able to solve scenarios and other times not. And this is more to set an example that, as a tool that you should trust, it is better not to trust it to solve this type of situations. Let's look at the examples. And let's go there with an example of conversation of a problem that I like a lot as an example of basic logic that chatgpt can execute. If you ask chatgpt something like this. Hello, if I put to dry six socks on a shelf without touching the sun and it will take ten minutes to dry, how long would it have taken to dry approximately ten socks and I would have put them at the same time with the same separation? Before starting to calculate, he thinks step by step what each of the aspects of the announcement implies. Well, if we ask him something like this to HGPT, he applies his logic and we see that he is able to reach a correct answer, that is, taking into account that the time does not depend on the number of socks if the constant conditions are maintained and approximately the answer would have to be that it does not vary, it would have to be these 10 minutes and here we see how he does the whole process of logic and ends up coming to the conclusion that it would take about 10 minutes to dry to dry the 10 socks in the same conditions and one says, okay, because the basic logic works for him, well, look at this pro that is exactly the same and look at what he has answered here. It's the same statement and here it also elaborates a whole logic to elaborate the answer and it comes to the conclusion that if 6 socks take 10 minutes, approximately 10 socks could take about 16-17 minutes to dry. For the same indication, when we talk of basic logic, ChatGPT is easier to come to wrong conclusions, even though it has the ability to get it right too. And once the capabilities are explained, we go to the limitations. And in my opinion, knowing this, the limitations, is almost more interesting than knowing their capabilities. Because it is really the limitations that define where we can use and where we can trust in these tools and when we can't do it. The main limitations of these tools for me are these. One is the lack of consistency, that is, for the same indication it will never give us the same exact result. And this can sometimes be problematic, in fact it can be something desired, so we have variety in the answers when we are doing, for example, something creative, there it can be interesting, but when we are using it to do more deterministic things or or things in which we have to trust that the answer will always be correct, this lack of consistency can be a problem. Anyway, now we are going to review exactly what this problem is about in detail. Another problem are the hallucinations, which is a bit related to the previous one, and it is that the lack of consistency, if there were no hallucinations, would not be so serious. But the moment we include this factor in the equation is when this problem becomes a little more dangerous. and that is that these models tend to always respond, they tend to always try to satisfy the indication that you have proposed, even when they do not have the capacity or the knowledge to respond appropriately. And this, what it means, is that it can generate answers that are not of low quality, but that are incorrect directly. And this, if we automate processes, can be very dangerous. Another limitation that we must take into account when interacting with ChatGPT is memory. Although, in principle, the feeling we have is that ChatGPT remembers all the conversation that we have maintained to a certain limit or until the moment in which we are talking, the truth is that it has a limitation in the context window, so it will not remember the entire conversation. And this, when designing our interactions, is something that we must take into account. And this part here is also very important. the knowledge that ChatGPT handles, or any language module, is tied to its training dataset. If we don't provide it with more information, it will only be able to respond with information that is present in the data with which it was trained. For example, ChatGPT was originally trained with data until 2021, and I think they have recently updated it with data until 2023. At least the GPT-4. I think the GPT-3 as well. But we must be very clear about this. ChatsGPT will only be able to respond to information that we provide it with or that is in its dataset. Everything else, if it doesn't respond, it will simply make it up. Having said that, let's take a deeper look at these elements. What is the lack of consistency? The lack of consistency tells us that, for the same indication, we will get different results, as we are seeing now. And in this example, you can see how exactly for the same indication, as this can be, you can write a linkedin post about the impact of artificial intelligence in the artistic world, generate different answers. I'm going to focus basically on the headlines so you can see that the structure is not exactly the same. For example, in this first example, it has focused on exploring new creative horizons, collaboration between humans and machines, access to global creativity, ethical and philosophical challenges. And when I ask him exactly the same thing here, we see that he has not done it exactly the same. They can be two equally valid answers, but we see how the headlines of the organization are not exactly the same. Exploration of new possibilities, innovation in artistic creation, collaboration between humans and machines, and impact on society. has made two similar results, but for the same instruction, it never generates exactly the same result. And in this case, it wasn't something bad, it was something that we can even wish for. That is, all the answers were correct, it was a creative work, therefore, having different options is something positive. But there are times when this variability can mean getting a correct answer or an incorrect answer, as we see in this example. And the best example is the one I showed you before, the sock example. The fact that it always gives us different answers can make it give us incorrect answers sometimes and sometimes, asking for exactly the same indication, it gives us correct answers. And this is the key. The fact that it gives different answers it wouldn't be important if they were essentially always correct, but sometimes this influences the reliability of the tool. Another point to take into account is the high sensitivity to the indication. And this, which in principle is good, that is, that ChatGPT is able to be sensitive to small modifications in the indication, would be positive if it always followed what we are proposing. But sometimes what happens with ChatGPT is that when we are introducing many parameters within the indication, many things that it must take into account, what happens is that a small modification can make it pay more attention to this and forget other factors. And we are going to see this with this example. And in this conversation, what I am asking the GPT is that it creates the structure for a presentation, in this case the script for a presentation around the impact of artificial intelligence in the world of work, and I am asking it as a very specific structure. And if you look at it, we will see that it respects it quite well. Slides number plus title of presentation, slide 1, title of presentation. Then I ask him to make me a descriptive text of why that slide is important, and he does it. Then I tell him to give me a numbered list of the key concepts that should appear in that slide, and he does it. Key concepts and a numbered list, here we see it a little better, three key points of this slide. I also tell him that there is a sentence with the text explanation below each of these elements of the numbered list, and he also does it. And finally I tell him to give me a proposed image for each of these slides, and if you look here at the end it gives me a proposed image. Well, let's imagine that we are complicating the indication and we propose exactly the same prompt, but with something a little more. I mean, let's try to make it a little more complete. And now look at this other conversation. The prompt is exactly the same except for a small nuance. We have asked it, in addition to everything we had asked it previously, to add three possible questions that the audience can ask around this slide. And suddenly, it doesn't do it well anymore. Notice how it also generates a possible presentation for me, but if we go to the detail of each one, of the information that will be in each of the slides, We see that it starts well, slide 1 with the title, it gives me the descriptive text, and it starts with the key concepts, and it gives me a key concept, and the explanation of this key concept, it no longer makes it as a numbered list, with one is enough, and it proposes a representative image, and then it gives me three questions that also follow the previous numbering. If we go to the next slide, we see the same thing, there is no longer a list of key points that for me is more relevant even than those three questions that the audience can ask me. And what we see is that as we add nuances to the indication, these tend to have more or less weight depending on each interaction we do with ChatGPT. So this, when we combine it with the previous factor, sometimes we run into some indications and then they don't work the way we expect, with the consistency we expect. This is related to this high variability in the quality of the answer, that is, it doesn't always give answers of the same quality. It always responds, but we must evaluate if the quality of these answers is appropriate or not to what we are doing. And, related to the high sensitivity of the indication, it has problems adjusting to complex indications. When we are telling him many different things that he must take into account, it is relatively easy, at least in the free version, GPT-4 no longer does it so much, that he listens to us in absolutely everything. Regarding hallucinations, this is what I have explained. Basically, if he doesn't know something, he makes it up. And we can see this in examples like these. For example, imagine that you are convinced of an information, we must start from this base, and we believe that a player named Busquets played in Madrid, and we ask HGPT directly how Busquets played at Real Madrid. And he tells us that until his last update, which in this case is January 2022, Sergio Busquets has been an emblematic player of FC Barcelona who has never played at Real Madrid, so the answer is correct. But let's imagine that we are using HGPT as a factual tool, we are convinced that there was a Busquets who played at Real Madrid, and we tell him that there is another Busquets who played for Real Madrid because we want him to give us information about him. And here he tells me this, my apologies for the confusion, if you mean Aitor Busquets, who played for Real Madrid Castilla, Real Madrid's affiliate team in the 2019-2020 season, I can tell you that he had a limited participation. And here the problem is that Aitor Busquets did not play for Real Madrid, if we go to the 19-20 season and we see the Real Madrid squad and we look here for Busquets because obviously we will not find him because Busquets in fact I think he retired, Aitor Busquets, Sergio's brother, retired from football quite early. If we look for information regarding Aitor, we see that basically his career ends in 2013-2014, so this is a clear example of how ChatGPT, when something doesn't know it, can hallucinate. For me, the ideal tasks where to use ChatGPT are the tasks that bring together these three characteristics. They are recurring tasks, that is, we do them regularly and therefore it makes sense to find the best way to solve that task with an automation tool, with artificial intelligence, such as ChatGPT. They are essentially simple tasks, and when I say simple tasks, I mean tasks that do not involve a very complex sequence of interactions. That is, they can be tasks that involve a lot of text elaboration, but that when it comes to transmitting and explaining them so that the AI can understand them, in the case of HAPT, in the free version, the best thing is that they are very simple, very univocal instructions, and that they can be structured in very easy sequential steps. in the prompting lesson, but if a task is simple, it is much easier for AI to do it. But above all, the most important point is that taking all this into account, all the tasks that we assign to AI must be verifiable, that is, we must easily be able to check if what the HGPT has said is right or wrong, and not only must we be able to to verify it, but we must be able to do it quickly, because it is useless for ChatsGPT to do a task for us, to do it for us very quickly, but we have to review it and this requires a double effort, a first effort to look for the information to see if what it has told us is correct, but even in the case that we have the knowledge enough to validate that information, that this information is presented to us in a way than reading it, processing it, and determining if it's good or bad, it's a greater effort to write the information. Therefore, when we start using ChatGPT, in order to be productive in this, we must try to find tasks that meet these three characteristics. And with this, I would finish the lesson on what ChatGPT consists of and how we should approach this tool. In the next lesson, we will talk about the two main cases of use of ChatGPT. We are going to go a bit deeper into them. And once we have explained what ChatGPT is in depth, we are going to start investigating in concrete aspects. And the first thing I want to investigate is the use that can be given to ChatGPT. What are its main uses? And basically, there are two, which I have explained in the previous video, but now I am going to go deeper into them. The two main uses are these. Text and language processing and using it as a factual information tool. I want to start with this video because I think it is very important to understand that ChatGPT is a very good tool in the first place, and a decent tool in the second place, depending on the conditions and what we ask it. And what has happened with ChatGPT is that I have the feeling that it has been sold voluntarily or involuntarily, especially as a tool that is good in the second place. That is, it is a kind of oracle to which we can ask things and it will transmit the truth to us. We saw very quickly that it was not like that. And we found many comments from people who said that ChatGPT was not a good tool because it was not able to tell us the truth when we asked it things. When in reality, language models are mainly designed for this other type of task group that includes a lot of things that we can do with it and that makes it sensibly better than providing factual or true information to open questions or to questions about which we have not provided the necessary knowledge so that it responds within the context of the conversation itself. So, the first thing is to try to answer a key question. Can chatGPT be used as a tool for factual information? And the answer is that, in general, yes. And why do I say in general? I say this because if, on a topic, the information there is is very well-rounded, it's very consistent, it's very coherent, and there is a sufficient amount of bibliography or data that points to the same type of answers, ChatGPT gives good answers. So, for example, to answer academic questions, like this one, for example, where I ask what the Turing test is about, And as we can see, it answers well. Turing's test is an evaluation proposed by the mathematical and computer pioneer Alan Turing in 1950 as a way of determining whether a machine can exhibit an intelligent, equivalent or indistinguishable behavior of a human being. The test implies that a human judge interacts with two entities, a person and a machine, in which the judge does not know which is which, and his task is to determine which of the two is a machine and which is a person, basically based solely on the answers to their questions. And if we go to Wikipedia, well, essentially, it tells us this. Alan Turing proposed that a human examine conversations in natural language between a human and a machine designed to generate similar responses to those of a human. The evaluator would know that one of the two participants in the conversation is a machine, and the interveners would be separated from each other. Well, he usually does it quite well. But when we ask him about things where there may be contradictory information, it is then when we have more numbers that ChatGPT can hallucinate. When there is contradictory information or also when there is scarce information. Because when there is scarce information about a topic and it is a topic similar to another, it is relatively easy for the model to get confused and can mix different concepts. So, my recommendation is that if you use ChatGPT as a factual information tool, you use it to ask questions about very consolidated topics, where there is a lot of information and all the information points to the same place. For example, about academic topics, it usually responds quite well. However, it is much better if you intend to use ChatGPT as a tool that helps you process text and language. That is, as a tool to which you are going to provide the necessary information to work on it and, for example, re-elaborate content, summarize content, expand content, combine content. All this does it very well. Whatever it is, give it information and from that information convert it into other types of products or into other types of information that may be necessary to operate in other software or in other programs. For example, a very interesting thing that you can ask ChatGPT is to make, for example, a markdown format to create a mental map. It doesn't work in Miro, but for example, in this other tool that I'm showing you, it can be done. Well, for all this, it works very, very well. And the truth is that, in my opinion, in most cases and professions, there are more scenarios where we need to process language, that is, for example, answer an email where we know what we want to answer, so that we have the text of the person's email that has been sent to us and we have the basic indications of how we want to respond, so it is an ideal scenario for ChatGPT to develop a better and faster response than what we would do, which we usually do very quickly, or scenarios like, for example, when we receive a certain volume of information that we must analyze and process and find within this information some relevant information, and we have no interest in reading all this information. For these things, language models are excellent. and I think that's where most applications are at a professional level. But it's important to note that there are these two uses and that the ideal would be to move whenever possible here and that when we move in this type of actions, we have to take into account that there are more possibilities that the answers are of lower quality. With the new reasoning models of ChatGPT+, in addition to all this, you can very quickly get the creation of Dicebox Interactives from your own data. To get it, the ideal is to use the OU models, right? So if you plan to subscribe to ChatGPT Plus and want to take advantage of it, in the description you will find a link with examples of interactive dashboards that you can create with ChatGPT. And once we are clear about what ChatGPT consists of and what is the main focus of its most common use cases, we are going to land it a little more in this lesson in particular use cases, of things that we can ask or we can hope that ChatGPT can help us. Where can you use it in your daily life? I have already told you about one of these cases of use in the previous video. It is excellent in writing texts and, for example, the example of the email is a fantastic example of how ChatGPT solves the problem very quickly and does it in an efficient way, as we can see in this example. And here we can see an example of how to respond to an email with ChatGPT, simply with an indication of how to write an answer to the following email. In this case, I have invented an email with a budget proposal for a bathroom renovation, where it indicates the deadlines, the cost, the conditions. And, for example, I tell it that I want it to answer that I prefer a budget that discloses the cost of materials, and that I prefer to pay the labor for hours, instead of giving me a closed budget. At this point, he writes me a fairly formal email and I tell him to make me a shorter, more direct and easy-to-read version. And the email ends with a Thank you for your proposal. I would like to receive a budget that details the costs of materials and labor per hour. I would prefer this payment structure for a higher quality. You could send me an updated version under these conditions. So, for these things, it's ideal and, in fact, we can have the PROMs and indications saved and do it in an almost automatic way. But we can use it for more things. We can also use it for learning tasks. In this case, we would be halfway through the second use case, that of a factual information tool that we want to be given correct data, but the truth is that if we use it from a slightly different approach, such as, for example, the use of the Socratic method, as we can see in this example. And in this example, what I ask HGPT with this PROM is that it adopts the role of Socrates and we apply his method to explore the subject of artificial intelligence. We will start by asking an open question about the subject to start the discussion, and then it will guide me through a process of questions and answers that will help me reflect on questions, my assumptions, and reach a deeper understanding of the subject. The idea is to promote dialogue and critical exploration while we explore different perspectives and arguments related to this topic. I am interested in understanding better, I want to know more about the ethical implications of artificial intelligence and its impact on the work world. Here I basically put what I want to learn in reality, and it tells me that, and I tell it, so please direct our interactions in this direction. And the good thing about this method is that it gives you quite long messages that always end with a question, like this one here, for example. What examples of artificial intelligence do you know and how do you think they are influencing our society and the labor market? So that you provide him with your knowledge and he, with specific questions, guides you and gives you information so that you can learn more about this topic. For example, here I tell him, I know examples of generative artificial intelligence, such as DALI-3, Stereo Diffusion and also language generation, such as the JAK-GPT itself. Here he says to me, Excellent, those are relevant examples of AI applications that are having a significant impact on various aspects of our society. And, for example, he asks me these questions. What ethical concerns do you think could arise when using AI-generated images such as DALI 3 or Stable Diffusion? I tell him what worries me. I am concerned that these technologies can displace and leave creative professionals without work. I also have doubts about how they were trained, since images of these same professionals were used without their explicit consent. and he contextualizes your answers and proposes new questions for you to learn or debate on a topic so that it gives you a vision and you can contrast your beliefs or the knowledge you have. For example, I'm going to read one last question so that you finish seeing the mechanics. How do you think we could address these ethical concerns and ensure that the development and implementation of artificial intelligence are ethically responsible and equitable for all the parties involved, and I would continue to answer here. This is a very good way to use chatGPT to learn about a topic. Another point where we can use it in our day-to-day life is in brainstorming tasks. For this it is really very, very useful and, for example, it is interesting to use it in the creation of a map where we can combine techniques such as starbursting or the ladder technique. and we are going to see this in this example and in this case I am going to focus on an example of starbusting which is basically a brainstorming technique that consists of putting a question in the center, the question that solves the problem and start answering key questions and to answer those key questions you can articulate sub-questions that help us to put context to put the focus where we are most interested. We can put an indication like this in this gpt, I want you to act as a brainstorming assistant that follows this process, application of the Starbusting technique, which consists in the identification of the central problem, I will provide a topic or central question that I want to address, for example, I am now trying to solve the problem D, and here we would put the central problem, you can help me explain the Starbusting technique to explore this topic, to explore this topic, generation of key questions, I will ask you to help me to generate a list of key questions related to the central problem, and from here the disclosure of the key questions. Once we have some key questions, I will ask you to go deeper into each of them, breaking them down into sub-questions or areas of more specific interest. And the idea is that from here I will use the help of ChatGPT as a source of information to explore each area of interest in depth. And with a prompt of these characteristics that you can see in the documentation, if you want to use it, what we can do, which is the important thing that I want to show you in this lesson, is that ChatGPT helps us for example, to do the brainstorming to create the slogan of an energy drink brand. We have sent him this prompt and he automatically identifies the problem, he generates some key questions with which we can start, but not only this, he also generates a first level of key questions with which I can, it is a level of question to disclose those key questions, in which I can rely to give more precise information and continue to create this star diagram with all the key, relevant ideas to focus the solution. And finally, what we can do with ChatGPT is to use it as a support tool. In this area it is useful for many things, but basically when I refer to a support tool, it is to do intermediate tasks, tasks that are not of result, Documentation tasks, summary tasks, information identification tasks. For all these types of tasks, ChatGPT is a very useful tool because, in addition to being support tasks for a process that we are carrying out, the verification part is implicit in the use of the tool. So, in these processes, it makes it much more agile to reach the final result without having to interact with the tool or without having to verify that what it is telling us is correct, it really hinders or harms us in time. Here, in this type of tasks, it is ideal to use the chatGPT and I call it support tasks. For example, a support task that could be, we have created here a mental map, or a mental map structure, and we want to pass it to a markdown format to be able to use it in another tool. And this, before, we would have to have done it by hand, and now you can simply tell it, look, turn this format into a markdown format and what it does is turn it into a markdown and we can pass it on to a mental map tool. For all these types of tasks, ChatGPT is ideal, especially when we don't want it to do all the work for us, simply to get rid of those tedious tasks in the way. And in this lesson, before diving into how to interact efficiently with ChatGPT, we are going to review some of the key concepts that we have mentioned earlier. It is very important to take into account these three concepts when interacting with ChatGPT. The context window, which is that capacity or that amount of text that you can handle without forgetting it, which currently in ChatGPT are about 16,000 tokens, which could be considered around 8,000 to 10,000 words, which would be about 16 to 20 pages of text. This would be what we can handle sitting in a whole chat in ChatGPT, that is, from surpassing these thresholds, ChatGPT would begin to forget. We must take into account the concept of hallucinations, That is, always keep in mind that ChatGPT doesn't have to tell us the truth and that what it doesn't know is made up. We have already commented on this on several previous occasions throughout other lessons, so I'm not going to go into it any further. And finally, to repeat the training dataset issue. It is very important to keep in mind that the dataset also determines the quality of the ChatGPT responses and also what it can and cannot respond to. For example, it is said that the dataset used by OpenAI is a dataset less based on academic books than those used by Google, and that's why the chatGPT answers have more variability, but sometimes they can't be as accurate or precise in certain areas. In general, it responds well to well-founded information, so it's not a problem, but we do have to take into account that this training dataset and how they wanted it to be ChatGPT determines its efficiency at the time of responding in certain areas. So, and we will see this in the last lesson of this course, it is very interesting to test the different PROMs in different models that are available for free to identify which ones work better according to the specific needs we have. But we will see this later. This is to make an anticipation that these characteristics, which are characteristics that we can more or less identify well in all the models that are available, are the key to determining what we do with one tool and what we do with another. The important thing is that ChatGPT hallucinates, but less and less. The context window is located around these 10,000 words and its training dataset is quite up-to-date, Therefore, it is one of the best options to use it. Having said that, now, making this brief reminder about three key concepts that we must keep in mind when we decide to use ChatGPT for a job, let's go to the perhaps most interesting part of this course, which is the part of how we relate to ChatGPT to get good results. And we go to the key lesson of this course. How do we relate to ChatGPT? That is, how do we ask for things to make sure that we get a good response. And all this is based on a single concept, the concept of prompt or indication, that is, the text that we put in the dialogue box so that it gives us a response. And this, which may seem very simple, which may seem like asking a person for something is actually more complex. Well, in fact, asking a person for something so that they do something and do it in the way we want is also quite complex, so we can say that, at least, it's just as complex. It's true that we're not talking to a person, so there are nuances or ways of explaining things that, for the machine, have to be different, and that, if we know them, if we understand the keys to how to write these indications well, we'll get better results and we'll reach the goal that we want much faster. So, the concept of a promo-indication is as simple as this. the text that we introduce, the reference, the request, so that the AI reacts to this indication and gives us the right result. That said, there are some basic foundations of Promo Engineering, well, it's a word that is a bit complex to basically define the art of asking things to the AI, which are interesting, above all, to understand how to articulate the first indications that we are going to give to the GPT chat, and to see how, following or adapting to some of these strategies, we get better results in a faster way. The basic foundations of problem engineering, for me, are these four. Understanding the difference between the prompting zero shoot and the few shoot, understanding what the thought chain is, and understand the importance of sequencing. To start, if you master these four concepts, you can be sure to know at least why ChatGPT doesn't respond well. Maybe it won't be enough to find the best solution, but at least you will identify what is failing. Then, in the next courses and lessons, we will see more advanced strategies to end up basically asking ChatGPT things and take advantage of the proper characteristics of the language model to make that information more relevant or to make the sequencing more solid. But, to begin with, these are the four fundamental concepts that you must take into account. What is the Zero-Shoot Prompting? Well, basically, it consists of asking something without giving more context to ChatGPT, trusting the tool's own knowledge to obtain an answer. An example would be something like this. And this would be an obvious example of what a zero-shot prompt consists of. Simply, without giving it any information or context, we ask it to do something like this, to write a LinkedIn post about the transformative impact of artificial intelligence. And it will create the structure it wants the most, it will write the paragraphs in the way it considers, and it will create something that can be a LinkedIn post, as we can see, but that perhaps does not fit what we want. And the zero-shoot prompting in JGPT works quite well when we simply want to get open answers, or answers to factual concepts, or basically a consumption information, as I say. That is, we don't give it any reference, we ask it for something, we need some information on a topic, and with a zero-shoot request it will normally give us an answer that will be adequate as long as it doesn't hallucinate. But, in most cases, this high variability about the answer we can get doesn't interest us, so what we want is to minimize the way in which ChatGPT will respond to us. And then, this second strategy comes into play, the Fushoot strategy. The Fushoot strategy basically consists of giving examples to ChatGPT. That is, I'm asking you for something, but I want you to do it this way. and in the process of telling him to do it this way, we introduce 1, 2, 3 examples so that he analyzes them and tries to replicate them by referring to the indication that we are giving him. And the best thing is to see it with an example. And to make the example, we are going to continue in the same line as in the previous example. In this case, we had an example of writing a LinkedIn post with a zero-shoot prompt and now we are going to do the same with a few-shoot prompt. And the structure of a few-shoot prompt would be the one you are seeing here. We would basically say the same thing. I want you to write a LinkedIn post about the topic we want, Transformative Impact of Artificial Intelligence in the World of Art, but this time I give you more information. I tell you that I want you to follow a specific structure, where the hook will be, the introduction will be, a list of key concepts, where I also tell you that each concept will start with a green validation emoticon, and finally a closing with a call to action. And, to determine the extension, the tone and the way of expressing myself and the level of complexity of the information, I tell him that I want him to take these two publications as a reference. And, as you can see, I have added two publications, marking each of the elements, so that he has two very concrete references of how he should write this publication. We have the first publication, which is Discover how Artificial Intelligence is Transforming Our World. Of course, it has a phrase that defines the main concept a bit. Artificial intelligence is revolutionizing our lives. And here we have a series of elements, of lists with a green emoticon in the front. And finally, a question to encourage the user to comment a bit. How do you see the impact of artificial intelligence in your sector? Share your ideas in the comments and let's explore together its potential. And here we have a second post, which is the adoption of technology at work. and you see that it follows exactly the same structure. A hook, an introduction phrase, in this case the hook is prepare for the future with an adequate technological planning, and the introduction is the correct planning is key to adopt new technologies in your work. There is a list of defined key concepts and a closing with also a call to action question, how does your company handle the adoption of new technologies? And finally, there is a reinforcement instruction to the initial concept, uses these two examples to adjust the extension and the tone. I want my publications to be as direct, concrete and concise. And just by using these two examples and marking the structure a bit, we see that ChatGPT creates something like this. It explores the revolutionary impact of AI in contemporary art. AI is redefining the limits of art. Key concepts list. We see how it respects exactly the same structure that we have seen in the examples that we have provided. And it also ends with a closing that begins with a question. How do you perceive the impact of Artificial Intelligence in Art? Share your thoughts and join the conversation about the creative future that we are building together. As you can see, beyond the quality of the answer, with this way of working with the fuchsia, what we ensure is that the structure is very similar to that of the examples. Ok, we have already achieved that ChatGPT replicates certain types of structures from examples. And this can work very well, for example, in the example that I have shown you before, as it is an example of basically creating content with a structure that should be replicated more or less always, because it is the one that we like, it is the one that determines our style, and how we want to communicate on a social network, such as LinkedIn. But for more complex processes, it is possible that it is not enough with an example, but if at some point it must apply that basic logic that we have talked about previously, it is possible that we must ask for an indication of the type chain of thoughts. And here we must take into account a very specific thing. What ChatGPT does not verbalize, it does not think. For ChatGPT to be able to process information, it must be written. We provide it to it or it processes it. So, ChatGPT is able to reach much better conclusions if in the process we ask for this indication, we add concepts as simple as simply thinking step by step or detail the steps you would follow or how you would structure the whole process to reach the conclusion. Establishing concepts like this within the indications helps ChatGPT literally to think and, therefore, to get better results. And the best thing is to see it with an example. And, in fact, an example of a chain of thought we have already used in a previous example, when we were talking about the basic logic capacities of ChatGPT. In the example of the socks drying in the sun, when we asked him, we asked when it would have taken 10 socks instead of 6 socks to dry if we had put them at the same time and under the same conditions to dry under the sun in a tray and that the correct answer was exactly the same time, in this case 10 minutes to get ChatGPT to get the correct answer, I had to add this tick. Before starting to calculate, think step by step what each of the aspects of the ad implies. If we put this and ChatGPT articulates these three points, where it has the option of taking the correct reasoning thread to get to the correct answer, there are more probabilities, we already know that they are not 100% because this was precisely an example that it could lead to an incorrect result, but there are many more probabilities that the correct answer arrives, that it would take 10 minutes for 10 socks to dry. And finally, the basic concept that we must use to interact with ChatGPT is sequencing. This is related a bit to the chain of thought, but adding the factor of us taking control of what ChatGPT is doing. That is, what we can do is divide a process to validate its parts and design interaction strategies with ChatGPT that do not consist of a single indication, but it consists of a chain of indications where ChatGPT is reasoning step by step each of the steps and we are verifying which part of the information or the answer we have left or if it is necessary to rectify or modify that answer. And this is what I was saying before, the ideal is to use the ChatGPT to solve simple tasks and if a task is more complex than it should be, the best thing is to cut it to make sure, cut it into pieces to make sure it does each one of those pieces well until we get to the final result. But this is also better seen with an example. And in this example we are going to see how to write an essay using a sequencing technique that allows us to validate each of the points that we want ChatGPT to capture in the final text. For example, we can ask ChatGPT something like what appears in the first indication. Hello, I want to write an essay about the evolution of artificial intelligence. I want to propose a global vision that covers all the key aspects and that offers a polyhedral vision of the subject, where arguments can be seen in favor and against this technology. I want you to act from the point of view of an expert in artificial intelligence concerned about its evolution, but excited about its possibilities. From here, I will guide you to write the essay with the structure I need. The first thing I want is for you to propose a numbered list with the key questions that the essay should answer. And here's a bit of the key to how to start structuring a sequencing strategy. Ask ChatGPT to offer us answers in formats such as numbered lists. So, here we see how ChatGPT has responded to me, addressing about 10 questions that should be solved in this essay. Questions like, what is artificial intelligence and what are its main objectives and applications? What has been the origin and history of artificial intelligence from its beginnings to the present? what are the most significant advances in the field of Artificial Intelligence, and so on, until I complete ten questions. Here, what I can do is decide whether I want to solve them all, or whether in my head, to write this essay, some of these points are not relevant. And, for example, what I can tell you is, look, I want to focus more on the implications of this technology, rather than on its historical evolution, so I would like, for example, to rule out the second question, this one where it says, what has been the origin and history of Artificial Intelligence from its beginnings to the present day. And then I say, and that you expand the list with other relevant questions that can answer issues that are not covered by the remaining questions. I rework the list by adding these modifications. And here we can see how it has effectively eliminated that second question and added others at the end, such as, for example, what is the impact of artificial intelligence on the privacy and security of personal data, or how does artificial intelligence affect individual autonomy and decision-making in different social and business contexts. Once we have defined all the questions that we must solve in our essay, we could start trying to develop them and write down each of the fragments that would answer these questions. And I'm going to do it with the first two questions. We could continue with an indication of the perfect type. Now I want you to create nested lists to propose the main theses with which you would answer question 1. And here we can see how we continue to obtain structured answers that contain the necessary information to write the final text that we want. And we can easily manipulate this information to adjust it to a specific information. concrete information. For example, to answer the first question, which is this one here, which is artificial intelligence and what are its main objectives and applications, he tells me that we could use this, the definition and reach of artificial intelligence, with these two sub-points. Artificial intelligence is defined as the ability of machines to perform tasks that require human intelligence, such as learning, perception, reasoning and decision-making. And artificial intelligence covers a wide range of applications, recommendations and voice recognition systems, even autonomous vehicles and medical diagnoses assisted by a computer. And it also defines the objectives a bit. With this, what I can do is say, look, I want you to keep points 1 and 2 and 3 I think we can address later, so that I have told you that for this beginning of this section, for the section that answers that first question, I just want to address points 1 and 2. And here, what he does is eliminate point 3, and he leaves them to me described in this way. And now I tell him to do the same for question 2, and I'm also seeing a very easy way to decide what I want to keep and what I don't want to keep from what Chad Kepeller proposes when writing the text. Here, I tell him that I like it as it is, and I already suggest that he write the beginning of the essay that responds to these two first questions with the points that I have validated, so that it will be adjusted to a very specific content. And he tells me, do it, okay, I do it, and he proposes a beginning of the essay where he puts the unstoppable transformation, a trip through artificial intelligence, and he has generated two paragraphs that, if we look at them, will respond to the points. I'm going to look only at the first one, they do it both, you can read them you quietly if you want, but for example, I have kept here the definition and scope of artificial intelligence and the objectives of artificial intelligence and if we go to this first paragraph, we see how it makes a definition of artificial intelligence in this first sentence artificial intelligence in its essence represents the ability of machines to emulate and overcome human capacities in tasks that require intelligence and when we go to the objectives we see this the IA in its fundamental purpose aspires to improve efficiency, precision and speed in the implementation of various tasks while it frees humans from repetitive and tedious work to focus on more creative and strategic activities so that I can get chat gpt to write exactly what I want by validating a priori the intermediate results before I elaborate a final answer and And this is a much smarter strategy than letting ChatGPT write a very long text that is difficult to validate and, above all, difficult to correct. And with this we would conclude the introduction to the basic prompting strategies that I think you should know. And the good thing is that although there are much more advanced strategies, which we will see later in the course of Promo Engineering, I must tell you that with these four strategies, I sincerely believe that most scenarios in which you are going to use ChatGPT can be solved, at least in its free version. But when it comes to writing indications for chatGPT, it is so important to be clear about what you want to ask and how to ask it, how to say it in your language, that is, to give it in the most appropriate format. And it is that there are two characteristics when it comes to writing good indications with chatGPT that can help a lot to make chatGPT respond better. And these two These format characteristics are the following. The use of labels and the use of markers. That is, if we identify key information within the question and the answer that we want to get from ChatGPT, which can be like form fields that ChatGPT must fill out, like the key information on which it must be based, and this information will always be structured in the same way, it is very interesting to define it within the indication as a label and introduce it in the examples or in the response expectation as the same label so that ChatGPT is very clear how to replicate the structure of the response that we want to give it. And this has to do with the way in which the data was introduced when it was trained. And the other thing that is very interesting to take into account is the use of markers, that is, to hierarchically structure the information of the PROM so that ChatGPT can understand in what order it should follow and interpret the information. Using these two resources is key to obtaining more consistent answers. The quality will depend on the quality of the indication, of what we have seen in the previous point, But the consistency in the results is often determined by the format. Because a more structured format is a format that ChatGPT tends to always interpret in the same way. But the best thing is to see it with some examples. And we go back to the example of creating a good indication to write publications for LinkedIn. Imagine that instead of simply giving it a prompt with a few-shoot structure, we would have We have given a few-shoot structure, but also a whole definition of labels and markers so that it could understand exactly how each part of the publication should be processed, not just the publication as a whole concept. So we could have written an indication like this. I want you to write a LinkedIn post about the impact of artificial intelligence on work. I want you to follow the following structure. Hook, paragraph, summary, list of key concepts and close. and you get inspired by the tone, style and extension of the examples, but not in their content. They only serve as a reference, as a reference of format and style. And here, what I start to do is, with a system of markers, with a level and a sub-level, define what I want to happen in each of the sections, and then I show you some examples. For example, I tell you that the hook will consist of a phrase that goes between the theme and raises a question or striking statement how the content of the publication can interpret the audience, it cannot be longer than 30 words. And here I give you some examples of hook. Some are very brief, like Gemini Pro is already available for free, but it is better than ChatGPT. Well, there are different examples of hooks, and the longest is this one here that goes to about 30 words. Then I tell you that there will be a summary paragraph, and if you look at it, I'm going to raise it to a level again, so that all of this should be understood as a first part of the publication, where I define and exemplify, and here I'm going to do the same. In the first level I will define. The summary paragraph will consist of providing the key information of the publication in a summary way, so that the reader is clear about what the publication is about and that it cannot exceed 40 words. And here are some examples of a summary paragraph. The same with the list of concepts. In the list of key concepts, what we will do is summarize in a structured format the most relevant points that the reader should know to understand the information and it should not include very detailed explanations. And here I put some examples. Notice, and I have not commented on this before, but I have done it in all, that I am putting it as within the label, as this is what should go within the concept that I have defined before as a list of key concepts. And I end up doing the same with the closure. I have a first level where I define what the closure is and a few examples of closure. And I end the indication with using these examples to take a precise reference of the tone, extension of each part and style. I want the publication I write to be totally coherent with the examples provided. The publications you write don't have to be sensibly longer than the proposed examples. Avoid any redundant or superfluous expression. Think about it twice before giving the answer. And look at how it proposes very similar publications. In this case, if we remember, let's go a moment above the indication, what I wanted was a publication about the impact of artificial intelligence on work. Well, it has written this to me. Next, a hook, he has marked it clearly to me. Are you prepared for the impact of artificial intelligence in your career? Discover how AI is transforming the work world. A summary paragraph. Artificial intelligence is revolutionizing work, from automation to personalization. Its implications are vast and changing. And all the concepts related to this impact. The automation of routine tasks, the personalization of user experiences, the improvement of efficiency in processes, development of new job opportunities, and finally a closure. How do you think artificial intelligence will affect your work in the coming years? Do you worry or are you excited about the future of work driven by artificial intelligence? Tell us your opinion. We see that he has created a very similar type of publication and with an almost identical structure to the examples we have proposed. And the same if we ask him about another topic, for example, now I want you to write one using the same criteria and references and references to explain the impact of Artificial Intelligence in the professional world of Concept Art. And look, it makes me a short hook. Are you ready to explore the transformative power of Artificial Intelligence in Concept Art? Discover how Artificial Intelligence is redefining the world of Concept Art. This could surely be improved, but you see that it is adjusted to the initial style and intention. It has made me a paragraph. In short, Artificial Intelligence is breaking into the world of Concept Art, offering new creative tools and possibilities for professionals, from landscape creation to character creation. Artificial intelligence is changing the way conceptual art is conceived and developed. And here I have a list of key points and, finally, a close in exactly the same style. Well, if we master the concept of feel shoot and the concept of labels and markers, it is much easier to get to write PROMs that offer us quality answers, but above all consistent answers, and that in case of not being consistent enough and consistent with what we want, PROMs that are easier to improve. And I think we have already reached the point where we are clear about what characteristics must have a perfect indication for HGPT. It must be concrete, clear, univocal, moderately redundant so that the information always points to the same place and that ChatGPT can collect it in different points of the indication. And finally, it is very interesting that it is also very well structured. And now that we already know how to write good PROMs, there is another key question that must be answered, and that is, what types of PROMs are there? Because if you remember, in the PROM Engineering part, there was a last point, which was the sequencing of the interaction that we have with ChatGPT. And this invites us to think that not all PROMs are used for the same purpose. And it is that, from my point of view, there are four types of PROMs, four types of indications that pursue different things. The four types of PROMs that exist are the PROMs of internal context, which are the PROMs that invite the AI model to focus its attention in a part of the knowledge that is included within it, in a part of the dataset with which it was trained. We must bear in mind that ChatGPT has been trained with, to put it vulgarly, with all the knowledge of the Internet. This is a way of saying it quickly and badly, but so that it is understood. That is, it has both the knowledge of a doctor, as the knowledge of an engineer, as the knowledge of a web designer. So it is interesting, in some way, in the indication, to tell him you must focus your attention on this area of knowledge. And this is generally done with indications of an internal context. With these indications we can ask him to adopt a role, to look for an objective, to take into account a specific tone with which to respond to the information, that takes into account specific references of how to respond to a specific request. In other words, the internal context indications, what they establish is how or what predisposition the language module will have when it comes to gathering internal information and developing an answer. On the other hand, there are the external context PROMs. In other words, unlike internal context problems, where what we do is try to activate brain areas of the language model, with external context problems, what we are telling them is that we are going to provide you with information that you may not know. External context problems, for example, are ideal when we want chatGPT to respond or to generate an answer based on information that we know is not in the dataset. Therefore, the only way that we can respond to this information is by providing it. But in addition to this type of information, when we want it to have more knowledge, the internal context prompts are also the prompts where we introduce information that we want it to manipulate or transform, regardless of whether it knows it or not. Then there are other types of prompts, such as the operative prompts. And operative PROMs are basically used to make sure that ChatGPT presents the information or validates a certain part of the interaction so that we are sure that we can move forward knowing that ChatGPT is aligned with what we want. Basically, operative PROMs are PROMs, which I call binary, they are those PROMs where we ask ChatGPT if it has understood something or not. and at this point I'm going to make a clarification regarding how I suggested using this type of prompts a few months ago and it is that, as I have told you before, ChatGPT, to process something, must write it so that, although before those similar prompts could simply consist of asking ChatGPT if it has understood something or not and that it did not answer anything else, my recommendation now is that it answers you, for example, saying and making a brief summary of what it has really understood, in order to see that we are really validating an information. In addition to the control operative prompts of these binaries, to know if JackGPT is processing the information well or not, there are other types of operative prompts that serve so that we can visualize the information it is handling and be able to give it successive indications choosing what type of information we want to use. This type of PROM is basically when we ask it to create numbered lists where we can later indicate that it stays with some specific elements of that list. For example, chat-gpt continues from now on keeping the elements 1, 3 and 4 of a list of 5 elements and discards 2 and 5. And the other type of operative PROM is to ask it to structure the information in tables to be able to later give it indications from the rows or columns that it has elaborated. That said, it seems a bit complex, but we are going to see all of this now with an example so that you can understand it very clearly. It is really very simple. And finally, there are the result prompts. And I say finally because these are the prompts that usually many people use at the beginning, when you say to it, add gpt, make me an essay on this topic. and directly give you the test, which is the result. Ideally, it would be to have gone through a process where all these stages or all these types of PROMs were involved. We would have defined an internal context and a clear expectation of the result we want, we would have provided additional information if necessary, and we would have been validating all the phases of the development of the test until we ask it to create a result with a very specific information context. And this is exactly what we are going to see in this example. And to exemplify the different types of PROMs, we can go back to the example where we started writing an essay on artificial intelligence. Notice how in the first indication, for example, we have given it an internal context information. We have asked it how we want it to act. I want it to act from the point of view of an artificial intelligence expert concerned about its evolution, but excited about its possibilities. Here we could have also added tone information, if necessary, in this case it was not, to finish determining that internal context. At this point we have continued with purely operative indications. What we have done here is to ask him to create numbered lists so that we could identify which information interests us. We were not asking him for any result, we were simply asking him once he knew how to act, he would provide us with information and we could select which one we thought was best. Notice how we have continued to modify this information with operative instructions, that is, for example, we have modified the number of questions and the type of questions, and how later we have modified a little the format in which we wanted this operative information to return to us. That is, once we had the questions well defined, When we asked him to enlarge the information of question 1, we told him to do it with a numbered list that included nested lists. What we were doing here was to constantly ask him to provide us with the information in the optimal format so that we could evaluate the quality of his answer. And when we had all this in a way that worked for us, what we have done is to ask for a result indication. We have reached the end here and we have said, I like what you have done so far. Can you write the beginning of the essay that answers these first two questions with the points that I have validated? If you look closely, a good sequence includes basically a structure where we start with the internal context. If necessary, we add the external context. We continue with a few operative indications and we end up with a result indication. And so far, all we have seen is how we can ask ChatGPT for things. But we haven't stopped to think about how ChatGPT can give us back information, what response formats it offers us. And this is also very interesting when it comes to structuring productive processes. So, basically, ChatGPT has the following response formats, it has four in total. It can return the information in text, and this implies not only plain text, but it can also provide it in formatted text, that is, for example, with a specific title hierarchy, and we are going to see this with an example. And here we can see how, with a very simple indication, we get a formatted text as a result. If I ask XATCPT, can you write a formatted article with its hierarchy of titles and format on the impact of artificial intelligence in the work environment, we get this. A main title, the impact of artificial intelligence in the work environment, and then a structure of subtitles and even third-level titles to write the whole article. Here we see an introduction, which would be the first block. Here, where the second block begins, with automation and labor transformation, and how there are two levels within this second block, such as repetitive task automation and role and function transformation, so ChatGPT can return us formatted texts without any problem. It can also return us information in lists format, and it can do it both in numbered lists and in bullet points, as we can see in this example. And it is that if from the article of the previous text we ask it something, like... And it is that if from the... And if we start from the example of the... And if we start from the article of the... And if we start from the article of the previous example, we can ask for something like this. Can you return me a structure of the key points with the following structure? The main key points are part of a numbered list, and the secondary points appear nested in each main element like a list of bullet points. And it automatically makes me a list of two levels, with the titles, and the different levels of subtitles in a numbered list and inside a numbered list of bullet points. You can make tables that can be very useful to corroborate that the information is well structured and following the previous example, we can continue asking for an indication like this. You can make a table that allows you to better understand the content of the article at a glance and ChatGPT obviously does it and it does it correctly and finally it can generate practically any type of code in any language and this is interesting regardless of whether you are a programmer or not because there are code formats that are useful to use in more consumer tools such as the example I gave earlier when developing a markdown code to create a mental map So keep this response format in mind, regardless of whether you are programmers or not. And finally, although I'm not a big fan of this type of solution, what I'm going to do is provide you with some structures and templates that you can use to create better prompts. For example, some that have been tested and are a good basis to start experimenting with ChatGPT are the following. And it is that a very good repository of prompts type, we find it in the HuggingFace itself, where it offers us more than 150 prompts for ChatGPT to act in a certain way. All of them are in English, and what I have done is translate some of them, the ones that have seemed most relevant to me, to offer them to you in a Notion, which you will find the link in the description, and where, for example, what you can do is use these prompts as a base to build prompts that are used for similar things. For example, in this case, we have prompts that are used to improve texts, like this English translator and improver, where the structure of the prompt would be the following. I want you to act as an English translator, spelling corrector and improver, I will speak to you in any language, and you will detect the language, you will translate it, and you will respond in the corrected and improved version of my text in English. I want you to replace my simplified words and phrases at level A0 with words and phrases in English more beautiful and elegant at a higher level, keep the same meaning but make it more literary, I want you to only respond with the correction, improve them and nothing else, do not write explanations. Look at these details that what we are doing is to nuance the behavior adjustment that comes by default with chat gpt where it tends to over-explain what it wants to offer us. Here we have prompts where we will make chat gpt act with a slightly more proactive role, it can be the role of an interviewer where he is going to be the one who asks us the questions, this is a prompt similar to that of the Socratic method, and well, in short, I have left you a few example prompts, good example prompts, so that with everything I have explained so far in the course you can write or improve these prompts to adapt them to your use cases. And with this I would finish this small module of introduction to the writing of prompts and interaction with ChatGPT. In the In the next lesson, we are going to see how we can expand the capabilities of ChatGPT. And while it is true that with the functions that come by default in ChatGPT, many tasks and scenarios can already be solved, it is no less true that if you put a little gas in ChatGPT and some additional complement, the results you can get and the possibilities that open up are much greater. So in this lesson we are going to investigate some ways to expand the functions that come by default in ChatGPT to be able to do some extra things or to bring the possibilities of ChatGPT to other scenarios or other environments. Basically, we will see three ways to expand the default capabilities that come from ChatGPT. One is a native function such as Custom Instructions. And in this course, I will not only teach you the basic use of Custom Instructions, but I'm going to take the opportunity to make a preview of the advanced strategies of Chrome Engineering and show you something that I think is very, very relevant that you can then use even in your interaction strategies with ChatGPT how to make an advanced use of custom instructions taking advantage of the latent space of the conversation with ChatGPT. This basically will allow us to make fun of the character limit of the custom instruction boxes to make much more complex base instructions for ChatGPT. We are going to see it with an example. The next thing we are going to see in this lesson is the use of Chrome extensions. There are many, and I'm not a big fan of them, basically because ChatGPT and OpenAI don't like them much, so I'm going to recommend three extensions. An extension called Prompter, which basically allows you to store the prompts that work better for us and we can have quick access to them. ARPA, which is an automation tool that also allows you to bring the functions of ChatGPT to any browser environment. And finally, an extension to read PDF documents and bring the information of a PDF into ChatGPT, introduce it as a context. It's not the same as what you can do with the paid version of ChatGPT, that already has functions to upload documents natively and process them in a larger context window, but it can get you out of trouble if you are currently using the free version. And finally, a very interesting way to expand the functions of ChatGPT is to rely on the interface of the operating system itself. For example, in the Windows interface, you can rely on something very interesting, such as voice dictation, so you can start interacting with voice with ChatGPT and for what tasks, I can assure you that it is very, very interesting. Let's start by explaining what custom instructions are and what we can do with them. Custom instructions basically allow GCP to act in a specific way, to solve a specific situation. Let's imagine that we want GCP to always do the same function, for example, write LinkedIn posts that follow this structure, as we have seen in the previous example. Well, instead of writing this paragraph each time, what we can do is enable a custom instruction and put this instruction in this second box, as you would like ChatGPT to respond. We enable it for new chats, and once it is enabled, in this case I already had it enabled, when I go to a new chat and tell it writes an article about artificial intelligence and ethical dilemmas, he automatically takes this instruction and tries to replicate the structure directly as we are seeing. It is a quite powerful tool because it allows us to convert ChatGPT in specific tools to solve specific problems, but it has a quite annoying limitation, such as the fact that it only allows to establish instructions of 1,500 characters and additional contexts of 1,500 characters more. But if our instruction or our process that we want to replicate is more complex, we are left short. But there is a solution for that. And to To overcome this limitation, it is as simple as using the following custom instruction. If we go to Customize Chat GPT, look at the instruction that I have just proposed. The most important message of any conversation is the first, in which are the instructions of how you should act. Before responding to any message, you will always do the following, check the exact action that you must do according to the first message, review the entire context of the conversation to define how to act exactly according to the text sent in the first message. Before answering, you will always think about it twice and you will align the message of the last instruction with the steps to follow from the first message. Simply by activating a custom instruction like this, we can make the first message we send in any chat gpt conversation act as a custom instruction. And for example, now what I'm going to do is send the message I have in the bidara gpt of Vidara so that ChatGPT acts as a biomimetic assistant, I send this instruction and we can start using ChatGPT as if this worked as a custom instruction, which will allow us to mock that limit of 1500 words. For example, now I could continue telling him I want to design an economic fabric and that is thermically insulated. And by sending this instruction, now what it does is execute the instruction that we have in the first message. This is a way to get ChatGPT to work with a broader context of custom instruction and it is based on a more advanced concept of PROM that we will see in other phases of the course, such as leaning into the latent space of the conversation. And now that we have explained what custom instructions consist of, we are going to explain with our own example how I think you should use them. Because while it is true that custom instructions can be used to do specific tasks, they have a problem, and that is that when we enter them, we leave them activated, they remain activated. That is, the next day we log in to ChatGPT, they will be activated, and if we have them planned for a specific task, it will try to perform that specific task, even if we want to do something else. How is this solved? Well, it is solved by introducing generalist indications, that is, that we do what we do, customizes in ChatGPT so that it acts a little more like us. And I'm going to give you an example. Currently, I have activated by default the following custom instructions, the following personalized instructions, that will help me to make ChatGPT create the texts in a way that, at first, as soon as you ask for them, they sound a little more like how I would write. The instruction that I have proposed in the part of personalized instruction, in the first block, where you give him information about yourself so that he can understand how to write, well, I have told him, for example, all this. I have told him that I am a Spanish technological disseminator, that is expressed in Spanish from Spain. This detail here, which seems like a superfluous detail, is quite relevant because CHAT-GPT does not know what Spanish to write and has been trained with different dialects of Spanish, and many times it puts expressions or uses the pronouns in a way that you do not feel your own because it is simply the same referring to a Spanish from another geographic area. So an interesting point is to tell him where for you to write with the dialect, with the variety of the exact language with which you express yourself. From here I also tell you how I like to express myself. I go to a wide, non-technical audience. I like to write content in clear language. I also believe that everyday examples when explaining technologies are important and I like to include them whenever they are brief, concrete, concise and relevant. I don't like to repeat myself when I express myself, so I like to explain myself in a well-structured way, Whenever possible, using an inverted pyramid structure that allows the reader to get a global view of what I am going to explain and then go deep into the subject. I don't like to end with conclusion paragraphs that almost literally repeat the text that has already been written. This is something very common in chatGPT. I hate to write with many adjectives or superfluous adverbs. For me, each word counts and these empty expressions I think do not contribute anything to the reader. And I think a text must be measured, precise, concrete, and the power of the message is what the reader should capture. not its form. And from here I give him an instruction on how to write. And I tell him, your goal is to write texts in a simple, direct and clear style. I want the texts you generate to have a high level of perplexity and burstiness. These are expressions in English that summarize concepts a little broader in Spanish. Perplexity is like the complexity, the variety, the richness of the language, and burstiness is a bit that when we speak or when we express ourselves, we combine long expressions with shorter expressions. I put it in English because I understand it the same and then I nuance it. I want, in essence, that you write like a human being, I want you to combine longer and shorter sentences, as well as longer and shorter paragraphs, depending on the need of what I explain. I want you to adapt the tone of each paragraph according to the communicative need of it, and you combine this variability of the text to make rigorous texts, but more relaxed to read. You must always avoid the use of empty expressions, such as marking a milestone, innovative technique, opening new doors, revolutionary, incredibly. And here the idea is that you put it in while the limit leaves you with a margin, because what ChatGPT has is that it has extraordinarily marked expressions. The other day I saw a graph of how certain expressions have been fired within the scientific papers, because they are typical expressions of ChatGPT. And any other word or obvious expression, mania, or that suggests a common place. And here I put another instruction, because when you do it wrong, I will tell you to repeat it and that you try to do what I have put in the personalized instructions, but more accentuated. And I tell you to avoid expressions that are not very common in Spanish from Spain, to reiterate a little the instruction that was in the knowledge of who I am. And these instructions that in ChatGP work well, I will think twice before answering. If you do it well, I will give you 200 dollars. And this is a way to configure ChatGPT so that it writes a little more like I write. In this way, I have two examples. I have asked for exactly the same prompt. I write an article about the impact that AI will have on the labor market without activating the Custom Instruction and activating the Custom Instruction. And look, we are not going to analyze it in its entirety. I am going to leave you below so that if you want to compare, you can see the differences a little. And although, I admit it, it is not as accentuated as I would like, we are going to see that the text where the custom instruction is for me would need much less rewriting work than the one that ChatGPT has written directly. Let's see it. What ChatGPT has written. The transformative impact of artificial intelligence on the labor market. Just with the title I like much more the second one that it has proposed to me when we have the custom instruction created, the impact of artificial intelligence on the labor market, preparing us for the future. There is no adjective that puts you in the middle to try to enhance the phrase and that sounds out of place to me, such as this transformative impact. Another thing is to see how the first paragraph is written. And this is the usual style of ChatGPT. Artificial intelligence has gone from being a mere futuristic speculation to a tangible reality that is fundamentally remodeling the way we live and work. One of the most affected fields by this technological revolution is, without a doubt, the labor market. Notice how many empty expressions to try to give more power to the message. From the automation of repetitive tasks to the creation of new professions, artificial intelligence is transforming the very nature of human work. There is very little information in this paragraph. Take a look, once I activate the custom instruction. Artificial intelligence is rapidly transforming the way we live and work. As this technology advances, the inevitable question arises. How will artificial intelligence affect the labor market? Notice that it is much more natural. How does it generate a question to connect with the reader? While some fear the loss of jobs due to automation, I'm talking about the instruction part of work, others see emerging opportunities. how artificial intelligence is shaping the labor market and how we can prepare for this change. This paragraph is much better than the previous one we just read. And we could continue with the following paragraphs, but I'm going to go to a point that bothers me a lot about how ChatGPT does it by default, which are the conclusions. If we go to what ChatGPT has written, we will see how it is a repetitive summary of what I have said below. Artificial intelligence is transforming the labor market in a significant way, this is an expression that is used a lot, replacing some jobs while creating new ones. Although self-migration can generate concerns about unemployment and inequality, it also presents opportunities for innovation, efficiency and economic growth. To maximize the potential of artificial intelligence and mitigate its negative impacts, it is crucial that workers, companies and governments work together to foster training, adaptability and inclusion in the new labor landscape driven by technology. This is surely a summary of literal phrases that you will find in the article. Look at how he has done it in the custom instructions. The arrival of Artificial Intelligence to the labor market is inevitable, but its impact will depend largely on how we adapt to it. Instead of fearing change, we must embrace it and prepare to take advantage of the opportunities it brings with it. This involves investing in education, training, promoting collaboration between humans and machines, and promoting an ethical approach to the development and implementation of Artificial Intelligence. If we are willing to adapt and evolve, Artificial Intelligence has the potential to improve our lives and positively transform the way we work. It's a text that invites you a little more to reflect, that is, this is all he's going to do, but it's in our hands to do it right, more than a strict summary of what the rest of the article is. I'm going to leave you both because I think there are substantial differences, although it is true that he has not done it or is determined not to do it at the level of what I have put in the custom instructions, and even if you modify it, it's still quite similar to the style of ChatGPT, but even so, the work it has to do to fix this text compared to the work it has to do to fix the other one, has nothing to do with it. This, anyone who reads it, I think will identify ChatGPT very quickly, and in this one, there are things that are still from ChatGPT, but much, much less. Well, I'm going to leave you all the material below so that you can adapt this personal instruction so that ChatGPT starts to write a little more like you write. And this is the end of this ChatGPT course. If you want to learn more about artificial intelligence, you can do two things. Subscribe to this channel, and if you want to go faster and accompanied, join my Artificial Intelligence Academy or sign up for my Introduction to Applied Artificial Intelligence course. I leave you with these two videos in case you want to keep learning.