Generative AI: What Has Happened (and What’s Coming Next)

Introduction

Good afternoon again. You are thinking, who is this annoying guy that is always speaking?

Change of Plans

We have had a last minute dropout speaker, so I will be handling the theoretical part.

AI Usage Before 2020

OK, so I want to begin with one question. How many of you have used AI before 2020?

OK, probably if you haven't raise your hand, it's because you don't really understand what is AI.

I'm going to set the timer. I have to timekeeper myself.

Generative AI vs Other Types of AI

So, CGPT is generative AI. Gemini is generative AI. But AI, it's also Alexa, Siri, Google Assistant, and many more.

How many of you have used any other kind of AI that is not generative AI? And the one that don't raise the hand is because you are lazy or you have been in prehistoric age.

Okay, whatever. Let's go ahead.

The Impact of AI on Our Lives

I'm going to do a really quick view of what's AI and how it's going to impact our life in the past, in the present, and in the future. And I have 15 minutes. 14.

Okay, so I want to begin with one shocking fact that I discovered in March, and it's that in March of 2025, we reached, in Hugging Face, one of the leader platforms for uploading AI models and LLMs, the same amount of AI models done in whole 2024. 12 months, three months. So for me, this is impact grow in the number of AI models in the market.

So you could think that it's only quantity and it's not quality, but if you have been using chatGPT, paid chatGPT or any other tool of paid, I think you can realize that it's make a difference. Okay.

AI is not new, in 97 some AI machine learning that it's also AI start working and the Turing test start to be speak about. But I want to take a bit for speaking, 2016, how many of you know what it's go? we are in AI field, so it's not impressive, that we have listened to this before.

Really quickly, the mind of Google built an AI that could beat the AlphaGo, the Go championship, that it was like, this is a really human skill. AI, a machine, cannot build it, cannot be better than it is.

Generative AI vs Machine Learning

AI is much more of generative AI, and I want to do a quick point that in some scenarios, you don't want to need AI, generative AI. We have some clients or potential leads that they come to us and they tell us, okay, I want to use AI. For what?

Some people don't know, and some people know, I want to use it for this, okay. but do you mean generative AI or machine learning? Because it's not the same for all the parts.

So if you are in a technical field, you probably know that generative AI is not the only one, but if you are not, I just propose you to have a quick conversation with a GPT or your LLMs of trust and ask them what other types of AI are available and for which use are better.

Applications of Non-Generative AI

For example, in economical field, if you want to do trading or a stock market, you really likely are not going to want to use generative AI because then you are not going to have sustainability of the results. And this is one of the biggest gaps of generative AI.

Autonomous driving, I think all of you might listen it, maybe seen it, maybe been in the car. This is also AI. AI that involves a lot more image than other type of AI, but it's not a generative AI.

Virtual assistants, that Juan has been speaking to us, it's also AI that it can be generative or not.

Understanding Transformers

Okay, so how many of you know what a transformer it is? I'm not speaking about the film of the robots.

Okay, so I'm a business tech, a geek, so this is for me an amazing example of how a small startup, a small company can innovate more than a big one.

ChatGPT is very in transformer technology and Google invented. Google discovered it in a paper in 2017, 2016, I'm not sure, but they didn't take a step for using it until OpenAI did it. Just a curiosity that I'm good on that part.

Okay, GPT is, it means Generative Pre-trained Transformer.

Transformer is, for a speak, explain it really quickly. One or a technical guy can explain it much better than me. But you use many computers and you take advantage of the computational power. You're not using one computer, taking the query and taking it to many, many computers or graphics or so on.

Pretrained is because ChatGPT is a model that took a really high part of the internet and responds based on the probabilities on the next word. For example, hello, good morning, it's probably really likely to be the answer, or good afternoon. So it's going to respond based on the information it's on the internet.

And generative because the purpose of generative AI is generate things, generate content, generate image, generate videos.

AI and Economics

GPT in economy. For understanding how big this technology revolution it is, GPT is not only generative pre-trained transformers, it's also an economic term used for technologies that are going to disrupt the world.

For example, electricity, nuclear energy, these kind of things before GPT and OpenAI were named GPT.

AI Limitations

Okay, quickly, AI limitations.

How many of you really likely in the past has asked something to touch GPT or other LLM or image generator and said this is bullshit, this doesn't work.

Okay, this is because as we have speak, the LLMs are trained based on huge amounts of information. So this is one of the reasons it cannot always respond good.

And the other reason is that Generative AI, it has in their DNA the mistake because it doesn't know, it's just probability.

So if it doesn't know, if it haven't been with that information for training, it's going to tell you something false. But it's never going to tell you, I don't know that. They are trying and it might be sometimes, but it's not the usual.

Okay.

Quick test, how many of you use the first version of ChatGPT?

Okay, this is the lowest question I have ever received today.

Evolution of AI Parameters

So, we can see that the amount of parameters used have been increasing, but not in a few thousands or in a few millions, in a few billions, in a few trillions. It has been increasing how we are counting them. And that's the reason, one of the reasons, we are getting better details, better responses every single time.

Advancements in Image and Video Generation

I want also to speak about, I don't know if you are seeing it well, about image generation, okay? This is mid-journey.

I think no one thinks this is a real person. But we go through the months, through the years, we start seeing a really high quality.

This is V7. We could even think if we see it quickly that it's moving.

Okay, and how many of you have been seeing recently VO3 videos? Okay, how many of you seen at the beginning thought it was a real video? This is dangerous, but we have to be accepting to everything we listen.

As fun tip, with your family, with your dear people, speak about a word, a phrase or whatever, that if you speak with them by phone, something is weird. Okay, how is something? I really recommend it, I have with my family.

Future of AI

Okay, about the future. AI, we are here, we are at the beginning of the mountain and we might go all the way up till the infinite or destroy the universe or not work again. Who knows?

But it could also be like that. We are here, we might go high a little bit more and then we are not going to be high anymore.

AI have a challenge, and it's data. So if we don't generate enough quality data, we are going to run out of improvements in generative AI field.

So when someone asks about the future, expert with the same years of experience can tell you one thing and the opposite, because the truth is that no one knows.

And you can disagree with me in that. It's totally OK.

Okay, so what is coming next? We are going to see an accelerated process for the moment.

We are going to see how governments, how different private companies are investing more and more in technology. Deep-seek, data drop in the stock market of USA, huge.

Because really cheaper than the other models, we're doing something even better. We can later, if you want, discuss about conspiracy theories about deep-seek, but for the moment we are going to think it's real.

And if you use AI properly, you are going to reach the goals faster and better. And this is something of future, because not all the people are doing it.

But the people that are actually using AI right now are actually there. This is the high-in-phase metric that we have speak before.

Geography and AI Development

And I want to speak a little bit of geography. The biggest majority of AI's we are using, where are from?

USA, right? Some of you might be using from China.

How many of you are using from Europe? Mistral, two people. And if we go to a leaderboard of LLMs, Mistral is not in the top.

And it's the only AI serious company that we have in Europe. We could speak about ILEA, but I'm speaking of serious companies that do good work.

Okay, so on top of that, I don't know if it's the next one. No, I'm going to go to another slide.

Private Companies Driving AI

I have compared before AI with nuclear energy, with nuclear bombs, if you want. But in this case, the Manhattan Project is running completely by private companies.

You can like it more, you can like it less, but it's a fact. that the objective of having the best AI is not under a government that is supposed to look to the good of the people, even if you believe it or not.

We are not going to enter in politics.

Okay, so what's next?

The trends of AI. GPT-5, they say it's gonna be huge.

I think it's not the biggest disruption in AI. How many of you are using agents? Many of you, okay.

I want to make a point, because depending on which professional you are talking to, they are going to have a different opinion. Agent, there are agents that it can be a GPT, or a C-Stand of Cloud or a Gem of Gemini, and you answer and they reply without connecting to other source. Depending on what professional do you listen, they tell you this is an agent.

From my point of view, it is. And there are autonomous agents that is, okay, I want this task done. Okay, I'm going to do in the background different process and I'm going to give you the final answer.

This is an agentic, an autonomous agentic that is going to be coming next. We have two ways of doing it.

One is connections and API like MCPs that Anthropic is doing a really good work. And another way it's letting the AI take control of your machine.

How many have used or listened of Manus or OpenAI Operator? The fact is that if you use it, they are not as good as they promised yet. But if they do what they promised, it's gonna be amazing.

I have run out of time, but I'm my own timekeeper, so I'm gonna keep. Okay, really quickly, okay.

Specialized AI Applications

Specialization, AI is not going to be a general AI, it's gonna be, okay, AI for pets, AI for engineers, AI for economical people, whatever.

And we are going to see more and more regulation, if we don't like it or if we do.

Mainly from you, for the moment.

AI-Powered Robotics

Okay, AI power robotics. We have seen robots before, but this is two things that make it different.

One is humanistic robots that have anthropomorphic form, and multimodal, on the edge. Multimodal on the edge.

Multimodal Technology in Robotics

Multimodal in one part is, for example, with voice agents. We can speak to them. They translate it in the background. They give it to AI and a voice model answer to us.

This is one way of doing it. And the multimodal is they understand what you are telling to them in real time.

And if you use ChagiPT, Paid version I think or maybe free now and you use the camera you use the microphone You can activate the camera and it's seen in real time with multimodal features But this is going to come with the multimodal is linked to On the edge on the edge is that I do a LLM that is quite small LLM or AI Or I can put it in a really powerful computer so there is going to be A robot that doesn't have to connect with cloud to answer, interact, and move in real time.

The Future of Multimodal Robots

That is going to be the rapture. And some people tell that in 2030, you are going to see more than one of these in Castellana Street. Maybe it's too early, or maybe it's too long for arriving.

I have to speak about that before, and I'm going to stop that for the moment.

Finished reading?