IA en 20 minutos: lo esencial y 2 trampas que debes evitar

In the next few minutes, I'm going to give you the exact mind map so that you can finally understand what artificial intelligence is, but without technical jargon and in such a clear way that it will put you ahead of 90% of people who repeat its acronyms without even knowing what they mean. Between you and me, if you're here, this probably sounds familiar to you. You open a news story, listen to your partner or a friend, one watch a YouTube video and hear LLM or neural networks, and you feel that pang of disconnection. It sounds completely like gibberish to you, and you have no idea what they're talking about. The truth is that sometimes you feel like you're a technological Alphabet. You feel like you're one step behind and that the world is moving forward and you don't know how to get on board. Three years ago, I was lost in the same labyrinth. Since then, and throughout all that time, I've studied and read a lot of unreadable documents that have given me a headache. I've taken courses, watched tons of videos, trained myself immensely. And today, I've condensed all that information into just 15 minutes to give you an accurate map that will help you go from darkness to clarity, from being completely lost to knowing exactly what someone is talking about every time you hear one of those things that up until now have sounded like total nonsense. With this map, we're going to make that feeling of overwhelm disappear completely. And it's exactly the same thing I teach my students and all the companies that I can help today so that they don't go crazy in this jungle of artificial intelligence. So I won't go on any longer. And let's get started.

Okay? As.

As I was saying, I'm going to share my exact map of AI let's start with what AI is. The two pitfalls that everyone has when using any type of current artificial intelligence model. We're going to see how it thinks and how it learns, and we're also going to see why it has changed and the best way we can use it. So what is artificial intelligence, really? And the thing is, we all tend to think of Terminator or perhaps chatgpt when we talk about artificial intelligence. But nothing could be further from the truth. The reality is that artificial intelligence has been with us for a very long time, and you have been using it without realizing it. Look, I want you to imagine artificial intelligence as a city, okay? And let's think of any city. For example, Barcelona. Barcelona. And all cities have an old town. Notice that I have Pointed it out here. And this would correspond to the original AI. That is a set of programs that follow fixed rules that humans have given them. But let's put Pac man here so that you can understand it in a very simple way. Look, I have put Pac man here, a game that most of us are familiar with. So here the ghosts do not learn to corner you, but rather follow rules that the programmers have given them. So they've told him that ghosts always have to chase Pac man along the shortest path. That if the distance between two routes is the same, the ghost must choose to go up first or they can never go back. Or also that when they're scared, which makes perfect sense, they move randomly. Randomly. So knowing this helps us understand how revolutionary things happened later. So inside that city, we're going to go a step further. Up until now, we had our old town over here, which would be our original AI. But this city is modernizing and starting to build on the outskirts. And here we have our modern neighborhood, which would be this word, these words that we often hear that sound so strange, which is machine learning. In this case, it's not the human who tells it exactly how to behave, but here, the artificial intelligence. In this case, the buildings are not built following set rules, but rather they learn by seeing many examples. And to make this very easy to understand, I've given you two examples here. For example, Spotify or Netflix. Those playlists that Spotify or Netflix suggests to you based on what you've watched before, you haven't told them what you like and what you don't, or at least not indirectly. But they've learned to make suggestions by seeing many examples of what you've watched before. So in these cases, artificial intelligence using machine learning improves with experience. In this case, Spotify or Netflix will make better suggestions the more you use them. And now we're going to move on to the latest revolution. We've seen that we had our old town in this part here. Then we saw that we had a modern neighborhood here. But notice that now these things we've called skyscrapers have appeared here.

Okay.

We'Ve gone a step further. These skyscrapers have emerged from our modern neighborhood. And it's called deep learning, which is deep learning. This is a type of machine learning, as we say. It's an extension of that modern neighborhood that uses layers or floors to recognize much more complex patterns. And that's where ChatGPT, Gemini Cloud, Grok would come into play, as you can see here. So what makes this so special? Right? How does this deep learning Work. What's inside these skyscrapers? Well, I want you to imagine a system of pipes. And I'm sure you've heard this about neural networks, which sounds super scientific, but we're going to simplify it this way. A system of pipes where information will go from the bottom up through layers until it comes out here. So an example of this would be, for example, Siri or Alexa. What happens when you send Siri an audio message? In one layer, it will recognize that someone has spoken to it. In the other layer, it will recognize whether it's a female or male voice. The other layer will associate it with you. And at the end of the day, when it addresses us, it already knows that it's us. So far we've seen something that seemed very complex in a very simple way. We have our city, our map built. We have our old town, we, which is that original AI that we've been using for so long and that we've said is a like Pac Man. Then we have that modern neighborhood that would be machine learning where AI learns by seeing many examples. And then we would have an evolution of that, those skyscrapers, that deep learning, which works well with a network of pipes that allows information to pass through layers. And therefore the result that artificial intelligence gives us is much better. And here comes a precaution that I want you to take into account, and that is that we have two traps with this artificial intelligence, with deep learning, and what we all know and use today. And they are both hallucinations and biases. I am not going to go into too much detail. We already have a video about this on the channel. I will leave the tag here, but to give you an idea, we have hallucinations and that is that AI lies. It is not that it does it on purpose. It is due to a question of probability, which is how it works. As I said, watch the video that I have linked to you. And here you must absolutely always verify the data that AI is giving you. It is true that after the latest update of chat GPT5, they tell us that the level of hallucinations has dropped significantly. But even so, this is a precaution that absolutely everyone should take. And then we have biases and the eye has prejudices, and this is very normal. It is a machine trained by humans. Humans have prejudices and therefore the machine also carries them with it. So if you stick with what we have just seen, with that map of the city and with these two precautions, these two traps that you have to be very careful not to fall into, you are already ahead of 90% of people who currently use it completely blindly. But we are going to go a step further because understanding what comes next will help you know which artificial intelligence model works best for you based on your case, and will give you much more criteria. Look at how AI learns. AI has three ways of learning. In the case of models like ChatGPT, Gemini, Cloud, and Grok, they use all three. But let's do a quick review.

Okay.

So first of all, we have supervised learning. And this is when we give artificial intelligence examples. And the correct answer, I tell it, this is right, this is wrong, and this is, for example, the way it uses to detect emails that are spam or not. So, as I said, the important thing here is not having a lot of data, but having the right data. In this case, then we have unsupervised learning, which is this one here. And in this case, we don't give it examples and tell it what's right or wrong. But look here, I've given you the example of the typical cards that we all used as children, where we had fruits and we had to group all the ones that were the same. Well, here it's the same. What it does is find patterns and similarities, and from there it makes groups to be able to categorize them. Then, as I was saying here, humans give it meaning. And then we have reinforcement learning. And here it learns with rewards and punishments. I've given you the Mario Brothers game and games as an example because it's a very graphic way of seeing it. So I want to give you, for example, a case that you've probably heard of, or maybe not, but an artificial intelligence beat the best go player in the world, one of the most complicated games in the world.

Okay?

And it's not because the eye is smarter. It's because the eye has been able to do more trial and error tests in less time and therefore has been able to find moves that a human can't possibly do in such a short time.

Okay?

So it tries many times and sees through trial and error what works and what doesn't. Now we come to the part that we all know by now, which is generative AI and LLMs.

Okay.

So this generative AI will be an artificial intelligence that generates new content. It can be text, it can be images, it can be audio, it can be videos. There are some of them that are multimodal. This means that they can do both text and images. This would be the case of ChatGPT, for example, or Gemini. And then we have those LLMs that are that type of generative AI that artificial intelligence capable of creating? Exactly. They would be what we just said, chat, GPT, Cloud, Gemini, Grok, for example. So those would be LLMs. But here I want to make a very important distinction, right? All generative AI are LLMs. LLMs are in this case trained with enormous amounts of text and predict the next piece of language.

Okay.

Then we have other models that do not generate text, but for example, only images or only videos, for example, this is the case with the images in midjourney. These are not LLMs, these are diffusion models. And this is an important distinction that we must make. That said, we're going to move on to the last and very important part, which is going to have a direct impact on how you use artificial intelligence in your day to day life. And that's the context window and tokens. I want you to think of the context window as the working memory of an LLM. That is the working memory that ChatGPT, Gemini, Cloud, Grok might have.

Okay?

I'm sure that on some occasion you've sent it a message, given it some instructions and it stops halfway, doesn't respond properly, or forgets what you told it at the beginning. And this happens because we've exceeded its working memory. Working memory is measured in tokens. Now right away we're going to see what they are. But here I've given you a little guidance so that you can see, for example, ChatGPT5, the latest update has approximately, because this changes, and it also changes in each model a context window that is a working memory of 400,000 tokens, that would correspond to 700 pages. With Cloud we have 200,000, 1,000. Gemini we have 1 million. Grock, we also have 1 million. Therefore, I want to make a small note here. If we know that Gemini has 1 million working memory in that, for example, ChatGPT5 has 400,000, 1,000 working memory tokens. And in our day to day work with very extensive documents, with many pages, very long reports, it will probably be much better for us to choose our reference artificial intelligence than Gemini to give you an example, right? So that you can see how this can impact the selection of the most suitable artificial intelligence for you. And now we go to the topic of tokens. And the thing is that a token does not have to be a complete word. They are fractions of words. Here I have given you, for example, an example with a sentence. Here I have told it. Marta and Juan are going to go on vacation to the Maldives in September. As you can see, despite the fact that these are 67 characters, this corresponds to 18 tokens. If I were to send this prompt to the artificial intelligence and ask it for something, this would count as 18 tokens. Therefore, it helps us to get an idea with what we saw earlier, which is 18 tokens. And here I see that I have 400,000 or 200,000 or 1 million working memory. In any case, it is not extremely important, but if you are interested, you can check how many tokens are in the proms that you are using. I will leave you here the OpenAI page with the section where you can paste your proms and it will tell you the number of tokens that are in that prom. Okay, and now we are going to the last part which is these working proms. Now that we know what the map of that city looks like, we know how artificial intelligence learns and the precautions we should take. Now we know what generative AI is, what an LLM is. We also know what the context window is and what tokens are. And now that we know this, we must also know how to communicate correctly with artificial intelligence. I'm not going to go into too much detail. If it's something that interests you, let me know in the comments so we can do it in a future video and go into much more depth. But I do want to leave you with a little guide so that in your day to day you can have a good structure and achieve good results from artificial intelligence. And before I forget, I'm also going to leave you the QR code here so you can subscribe to my newsletter where every day I send you an email with valuable and useful news and content so that you can advance in all this artificial intelligence stuff. Okay, let's go with these tips. So the most important thing and what you have to remember is that you must have a defined structure. There is a saying that says garbage in, garbage out. This means that if we are not communicating with artificial intelligence in the appropriate way and it is not understood obviously the result it will give us will not be good either and we will get frustrated, we will blame it. But the reality is that we are not communicating correctly. Therefore the main structure, then we move on to the next part, which is the role part. I am sure you have heard it and right now there are many who tell you that the role is no longer necessary, while others say that it is. Well, I am going to tell you that it is optional. And this is not something that I am making up. But companies like Antropic, which is the company behind Cloud, or in this case Google with Gemini, say that using the role in specific tasks or those that require a lot of detail improves the response that artificial intelligence gives us. On the other hand, we also have, for example, the case of OpenAI where it tells us that the role is not necessary. If we are using models that reason, like ChatGPT5 thinking.

Okay.

So depending on what we want, and we want it to act as an expert in something, and we need a very detailed response, very, very, very specific tone and certain criteria, I am going to advise you to use the role. If you need something not so detailed, it will not be necessary. Secondly, we are going to give it an objective and this is the most important thing and it is what it must do. On the other hand, we are going to give it the context, which is very important. We are going to give it context, this is essential. And then we are going to give it criteria, that is, if we want it to maintain a type of tone, if we want a length, if we want it to have some elements that are important to us and that it must take into account, or if on the contrary, we have some restrictions. And then we're also going to ask for the format, whether we want it to be in text format, an image, a table or a graph. Do you remember that feeling we mentioned at the beginning of feeling like a digital illiterate? That party you didn't feel invited to? Well, with this map you know the entire city. You have all the information you need translated so you never feel left out again. With this information you're ready so that every time a news story comes out or you hear someone, you know exactly what they're talking about. And above all, so that you have the criteria and knowledge to be able to choose from all that sea of artificial intelligences and applications that we come across every day. If you found this video useful, don't forget to share it. Give it a like. Leave me your comment because I read them all and I love them too. And I'll see you next week.