YT YouTube2Text

YouTube2Text → Transcripts → Experto en IA: "Solo los que Sepan Esto Sobrevivirán a la Inteligencia Artificial"

Experto en IA: "Solo los que Sepan Esto Sobrevivirán a la Inteligencia Artificial" — Transcript

by A lo Grande Podcast (con Marian Gamboa) · 25,339 words · 2,537 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Most of society is going to become even more useless than it is today. I mean,
0:04we’re heading toward a world where AI will produce pretty much
0:07anything at all—a PowerPoint,
0:08a report—because it can do many of the things we’ve been doing,
0:12only faster and cheaper. For example, in a mammogram,
0:14artificial intelligence can spot something six months before the
0:18best radiologist in the world.
0:19Now I have to tell the doctor that the best way to help
0:22society is to sit in a chair,
0:24because if they get involved, the outcome gets worse.
0:27Pardon?
0:27And this is the question people should be asking:
0:30what value will I bring when what I've contributed up to now—my
0:33intelligence—is no longer valuable?
0:35John Hernández,
0:36the leading authority on artificial intelligence in the Spanish-speaking
0:40world.
0:41He’s a communicator, speaker, and director of Bish School’s master’s program in AI.
0:45Millions watch him on YouTube every month,
0:47and he’s trained thousands of people and companies.
0:50Today, he’ll explain how to use AI to gain time, money,
0:53and opportunities—and what we need to do to avoid falling behind.
0:56La inteligencia artificial no es opcional, es una cuestión de competitividad.
1:01Poder usar la misma tecnología que el gobierno de EE.
1:04UU., la CIA y el FBI es algo que nunca habíamos visto.
1:07Puedes igualar a una multinacional con millones de presupuesto usando una IA de
1:12veinte euros. Es una ventaja competitiva que nunca antes tuvimos.
1:16What other things could we use artificial intelligence for?
1:20Hablamos de algo que te acompaña las veinticuatro horas.
1:23Imagina que voy a la panadería.
1:25Digo: "Compraré el pastel de chocolate".
1:27Y una voz me dice: "John, llevamos tres meses cuidándonos".
1:31¿No crees que el vegetal sería mejor? Y elegiré el vegetal.
1:34Eso, en dos años como máximo, estará con nosotros siempre y cambiará el mundo.
1:38How do we ask the right question so we get valid information?
1:42First of all, chat itself is dead. I mean,
1:45the whole idea of interacting with AI to get information was a twenty-twenty-three,
1:49twenty-four, and twenty-five thing. Twenty-six is about...
1:53Over the last few months, we’ve been bombarded with the same idea:
1:57that artificial intelligence can already empathize, understand us, think,
2:01and even feel and have consciousness. Is that really true? Today,
2:05we’re going to answer that with an expert—someone I truly believe is one
2:09of the most informed and up-to-date people on this subject anywhere in
2:13the Spanish-speaking world.
2:15And we’re going to take it a step further,
2:17to find out how we can use artificial intelligence to our advantage,
2:21make it work for us, and gain time, money, and opportunities,
2:25while also protecting our families from everything that’s happening nowadays,
2:29including the things arising as a result of artificial intelligence.
2:33So get ready for this interview,
2:35because I’m sure you’re going to learn so much and make the most of this opportunity
2:40that artificial intelligence is offering us in this new era.
2:44You’re going to discover how it can work for you,
2:47and how you can take advantage of it.
2:49And before we get started, introduce our guest, and welcome him today,
2:53I want you to leave your answer in the comments to the following question.
2:57If today you could make one specific task you do every day in
3:01your daily routine more efficient,
3:03what task would you like to delegate to artificial intelligence?
3:07Your day-to-day work, answering emails, learning something new—a new profession,
3:11a new language? In other words,
3:13what would you personally like to delegate to artificial intelligence?
3:17Leave it in the comments,
3:19because today you’re going to find out whether that’s possible.
3:23And with that, let’s welcome John Hernández.
3:25Welcome to A lo Grande Podcast. How are you?
3:28Muchas gracias por la invitación.
3:30Thank you for being here today.
3:33I was telling you off camera that this is the first time we’ve
3:37brought this topic to this space.
3:39So I’d like us to talk not only about what’s happening now,
3:42but also about the practical ways we can use artificial intelligence.
3:46So my first question probably isn’t the one I’d usually start an interview with,
3:51but I think it’s the most controversial one.
3:54Because, look, people are saying so many things about AI these days,
3:58and I’d like us to have the chance today to demystify what’s true and what isn’t.
4:02We’ve heard a lot about how it’s evolving.
4:05It seems that it already thinks, already feels like we do,
4:08and that it might even be starting to create its own consciousness.
4:12How close are we to that? Is this real? What’s your position?
4:16No, it isn’t. I mean, we have no idea. The thing is,
4:18we’ve created something that does things we thought were uniquely human,
4:22and we don’t know how far it can go.
4:24In fact, just recently, a few weeks ago, new research was presented by Anthropic,
4:28one of the AI labs, and they found that something like a subconscious has emerged
4:32in artificial intelligence.
4:34A space like the one we use in our human brains to process things
4:37without being aware of them,
4:39which helps us do everything we do.
4:40Well, something like that has emerged, for example,
4:43in the latest artificial intelligence systems.
4:46So we have to be very clear about one thing.
4:48First, it isn’t like a human.
4:49I seriously doubt that artificial intelligence has feelings, but it’s a
4:53black box. In other words, we don’t really know what’s going on inside.
4:57And I don’t mean us ordinary people—the scientists who invented it don’t fully
5:01understand this technology either.
5:03So we always have to take it with a grain of salt.
5:05But I’d tell you to come back down to earth,
5:08to focus on what we can actually see and touch,
5:10and forget about imagining what it might perhaps become someday.
5:14If AI were to develop consciousness in the future, would it emerge?
5:17Well, we haven’t got the faintest idea.
5:19And so I think there’s not much point in putting ourselves in that scenario.
5:23It might be interesting, and it might be something to talk about over a few beers,
5:28like, “Hey, what do you think about this?” But the reality is that we have
5:32much more important problems,
5:33and much more important benefits to make the most of with artificial
5:37intelligence—benefits that are already right in front of us.
5:40So sometimes thinking about what it might become isn’t as interesting as
5:44thinking about what it already is today.
5:46Because I think that’s the biggest mistake most people make:
5:49they talk about AI in future terms.
5:51No, no, this isn’t just the future—it’s the present.
5:54We’re already tremendously late.
5:55And that's interesting, what you just said, because I think that sometimes,
6:00even just by thinking about what could eventually happen,
6:03we miss the opportunity to take advantage of this tool right now, right?
6:07I mean, as a competitive advantage.
6:09And that's one of the things that I discovered while preparing for this
6:14interview: it's true that women today make even less use of artificial
6:18intelligence than men do.
6:19-A lot less. -Harvard: twenty-six percent
6:22It's improving, but much, much less overall, generally speaking.
6:25But understanding this shocked me even more. Why does this happen?
6:29According to these studies,
6:30women said that when they used artificial intelligence in their work,
6:34they felt like they were cheating at the game—that is,
6:38cheating when it came to working or using it as a tool that could
6:41give them an advantage.
6:43How should we understand the use of artificial intelligence today,
6:47so we can derive the greatest possible benefit from it?
6:50Exactly as you understood it.
6:51We have to see artificial intelligence not as something optional, something I like,
6:55something I want to use or don’t want to use, but as an issue of competitiveness.
6:59In the new mediocrity, it’s going to be called excellence,
7:02and to reach that level of excellence, you’ll have to use artificial
7:06intelligence. So there’s no way to compete.
7:08And I’m not just talking about the job market anymore.
7:11I’m also talking about everything we do in our daily lives.
7:14There’s no way to keep up if we’re not using every tool available to us.
7:17Artificial intelligence is going to make human intelligence less valuable,
7:21because it can do many of the things we’ve been doing, only faster and more cheaply.
7:25So this basically forces us to rethink what we contribute as human beings.
7:29And this is the question people should be asking:
7:31What value am I going to contribute when what I’ve been contributing up until
7:35now—my intelligence—no longer has value?
7:37We have to see artificial intelligence as a platform, if you like,
7:41that’s lowering the value of intelligence in society.
7:43And from there, what we have to figure out is:
7:46What am I going to do when intelligence no longer has value?
7:49And that sounds like an extremely philosophical question,
7:52and obviously it opens up countless different branches of thought.
7:55And people say, “Well, that’s already thinking way too big,
7:58Jon.” But this has already happened in human history.
8:01If we go back about two hundred years,
8:03there was something called the Industrial Revolution.
8:05Before the Industrial Revolution, if you could lift a hundred and twenty kilos,
8:09you were guaranteed a job, and a very well-paid one,
8:12because what we valued most in the world was strength.
8:15Then machines arrived and reduced the cost of strength.
8:17And what happened is that nowadays, any one of us, with a forklift,
8:21can lift three thousand kilos.
8:22So where’s the value in me being able to lift a hundred and twenty kilos?
8:26Well, to show off with my friends at the gym, but it has no economic value.
8:30So what did we humans have to do? We had to pivot.
8:32We had to find another way to contribute value.
8:35And what we found was intelligence.
8:36Well, as the Americans say, history doesn’t repeat itself, but it rhymes.
8:40Now we’re at a point where what’s losing value is intelligence.
8:43And what we humans have to do is figure out how we’re going to contribute value
8:47without using strength, and without using intelligence.
8:50That’s what we have to do now,
8:52and that’s the question we should all be asking ourselves.
8:54Not so much whether artificial intelligence develops consciousness,
8:58whether it has feelings,
8:59or whether ChatGPT really feels it when it tells me we’re friends.
9:02All of that, honestly, adds very little value.
9:05What we have to figure out here, and focus on above all,
9:08is how it affects us and how we can make the most of it.
9:10Because, as we can see, nobody understands why it works.
9:13So it’s very difficult for us, as ordinary people, to fully understand it.
9:17But how it affects us, and how to make the most of it,
9:20is something we can all do individually.
9:22We can start taking advantage of it,
9:24so that this leaves us in the best possible position in the new scenario.
9:27I want to come back to this topic again later on,
9:30because I feel it’s important to also break through this fear of,
9:34“I feel like when I use this tool, it seems the tool is actually manipulating
9:38me, controlling me,
9:40instead of me trying to control the tool.” It happens to me quite a lot, personally.
9:45I think the problem is defining it as a tool.
9:47We’re saying that a tool is something that helps us do what we do better.
9:51But artificial intelligence does what we do ourselves, in many cases.
9:54To a great extent—not today, of course.
9:56And everything I’m explaining isn’t something that’s going to happen in two
10:00days. I mean, as of today, it’s not that intelligence has no value at all. Today,
10:04certain parts of what we used to do with our intelligence are
10:08already losing their value.
10:09For example, did you know that today, if an AI reviews a legal document,
10:13it can do so twenty percent better than a senior lawyer?
10:16Makes it work
10:16Cien veces más rápido y al cero coma cero tres por ciento del coste.
10:21Esa tarea que antes requería inteligencia humana,
10:23algo que un chimpancé no podría hacer, ahora tiene competencia.
10:27La inteligencia artificial ya puede. And what does that mean?
10:31El abogado que revisaba documentos debe hallar nuevas formas de aportar
10:35valor a sus clientes o a su empresa.
10:37Porque eso ya no tiene valor.
10:39Lo mismo ocurre con muchas otras tareas.
10:41Por ejemplo, atención al cliente. We’re finding that today,
10:45if an artificial intelligence handles your request when you want
10:49to return a piece of clothing,
10:51or when you want to transfer your phone service,
10:53you save nine minutes of your life in that interaction.
10:57Si lo hace un humano, es más lento, tedioso y problemático.
11:00Muchas cosas que hacíamos los humanos ya no tienen sentido.
11:04Y ese rango es cada vez más amplio.
11:06Cada vez habrá menos tareas humanas que sigan teniendo sentido.
11:10Eso no significa que el ser humano pierda su propósito.
11:13Al contrario: tenemos un propósito intrínseco y debemos aportar ese valor,
11:17pero no mediante esas tareas.
11:19Está claro que este abogado ya no debería revisar documentos.
11:24I’m going to pause here for a second,
11:26because what Jon just said is the key to everything.
11:29Making the most of artificial intelligence is something you can do today,
11:33and it’s no longer optional.
11:35We’re all seeing how artificial intelligence is transforming the way we
11:39work, learn, create, and interact with technology, at an extraordinary speed.
11:44But one thing is understanding that, and another is knowing how to actually do it.
11:48And that’s where I want to help you, because Jon is going to be offering a
11:53free, three-day training on September first, second, and third,
11:57designed to take you from theory to practice over all those three days.
12:01So what are you going to learn? First,
12:03how to use ChatGPT to save yourself hours every single week on the things
12:07that eat up your day right now.
12:09Second, you’ll be able to create your own website or app from scratch,
12:13without needing to know how to code.
12:15Yes, you can do it yourself. And third,
12:18you’ll learn how to put those artificial intelligence agents to work for you,
12:22so they can take care of all that repetitive work you have.
12:26And the best part is that, at the end,
12:28you’ll receive a certificate of participation that you can add to
12:32your professional profile.
12:33All of this, completely free.
12:35Three powerful live sessions—and for Jon to give this away,
12:38when he’s one of the leading Spanish-speaking experts in artificial
12:42intelligence, well, that doesn’t happen every day.
12:45Sign up by scanning the QR code that’s appearing on screen right now,
12:49or go to the link I’m leaving just below this interview.
12:53Reserve your spot now, because space is limited. And now, back to Jon.
12:57You can tell me later what we can actually bring value to, okay?
13:01But in what areas can we really get an advantage from what we still have—this
13:05awareness that human beings still have,
13:07and that comes naturally to us, something innate to human beings?
13:11But, well, I’m going to set that aside for now,
13:14because I want us to really bring this down to earth together.
13:18What exactly are we referring to when we talk about artificial intelligence?
13:22I mean, the usual thing, right?
13:24When I open ChatGPT, right, what am I really opening?
13:27What’s actually behind it?
13:29Look, there’s a really important point here: this is extremely technical,
13:32and I don’t think it helps us much to get into how the neural network works,
13:36and so on.
13:37But what we do need to understand is that this isn’t traditional software.
13:41In fact, I think calling it software is a bad idea. There are two layers.
13:45One layer is the application, which would be ChatGPT, and the other is the model,
13:49which would be GPT—the model the application runs on.
13:52It’s important to separate them,
13:54because what we’re seeing is that models are constantly evolving.
13:57Today you have Gemini, you have GPT, you have Claude from Anthropic,
14:01you have different models, and there are dozens and dozens of them.
14:04Those are the three main ones, but there are many, many more.
14:07And then you have the applications. For example,
14:10we’re seeing approaches that bring different models together
14:13at the application level.
14:14Microsoft, for instance, has created a kind of aggregator where,
14:17within the application, you can use different models.
14:20So, in the end, there are different ways of approaching all this.
14:23But you have to separate those two layers,
14:26because the best application isn’t always using the best model, and vice versa.
14:30Right now, for example, to me, the best application is ChatGPT’s.
14:33But the best model is Anthropic’s.
14:35So, in the end, I can’t use ChatGPT’s application with Anthropic’s model,
14:39because I’m required to use ChatGPT’s model.
14:41And that puts you in a situation where you have to choose what you want to work with.
14:45That’s basically the issue.
14:47But the reality is—and this is to help us understand what it actually is—it’s
14:51nothing more than a computer that,
14:53all of a sudden, we’ve taught to think.
14:55And it’s very important that people understand this,
14:57because traditional software—and this is why I say this isn’t software—is
15:01programmed. That’s what we’ve seen all our lives.
15:04When you create any kind of software, like PowerPoint or Excel,
15:07someone has programmed, line by line, what that software has to do.
15:10A plus B equals C.
15:11That’s what scientists and programmers call deterministic technology.
15:15You know exactly what’s going to happen before it happens,
15:18because the software can only do those things.
15:20In other words, it can only do what someone has explicitly programmed it to
15:24do, and nothing beyond those instructions.
15:26When we deal with artificial intelligence, it’s completely different.
15:30We don’t program artificial intelligence; we train it.
15:33In other words, we give it a series of algorithms—which is that very technical
15:37part, and it’s a slightly strange word—but basically,
15:40algorithms are a way of doing something where we tell it how to learn,
15:43but we don’t tell it what to learn.
15:45Artificial intelligence learns by itself, so to speak,
15:48based on a set of instructions.
15:49To make it really simple, it’s as if we set the rules of the game,
15:53and then it plays whatever game it wants.
15:55So, if you want an example that I think everyone will really connect with,
15:59it’s a bit like a plant. We provide the soil, we decide how much water it gets,
16:03and we decide how much sunlight it receives.
16:05But you have no control over, or say in, how many leaves grow,
16:08how many flowers appear, or when they appear.
16:11The internal process of the plant is too complex for us to understand,
16:14even though we’ve established the rules of the game.
16:17Well, that’s what happens with artificial intelligence.
16:20Today, we set the rules of the game,
16:22and then AI plays such a complex game that we’ve stopped understanding it.
16:26We don’t know exactly, or in detail, why, when you ask ChatGPT, “Hey,
16:29what’s the capital of France?”, it answers Paris—or sometimes it makes something up,
16:34which we call a hallucination, and tells you Berlin.
16:36We’re not really clear on the process that leads it to that answer,
16:40even with something as simple as this.
16:42Now imagine if I ask it to prepare a fourth-quarter report based on an Excel
16:46file, based on this or that, using company data,
16:48connecting to a particular database—I mean, just imagine...
16:51We don't really understand how it analyzes or concludes.
16:55What is the internal process it follows to reach its conclusions?
16:58We know it follows certain instructions, what it’s theoretically supposed to
17:02follow, and there’s a whole series of instructions we can give it, and so on,
17:06but we still don’t really understand what happens inside that neural network,
17:10because it is, after all, a neural network, very similar, although with
17:13differences, to the one we have in our brains.
17:15So it’s a nondeterministic process,
17:17and that’s something we really need to understand,
17:20because we’re used to software being deterministic.
17:22Now, there is one nondeterministic thing we work with every day,
17:26something we’re extremely used to,
17:27and it causes no problem and nobody goes crazy—and that’s us humans.
17:31I mean, look, I can drive up the highway,
17:33and on the highway there’s a nice big sign that says,
17:36“Don’t go over one hundred and twenty or one hundred and thirty
17:39kilometers per hour.” How many of us have gotten a speeding ticket? Exactly.
17:43Humans aren’t programmed; we’re trained,
17:45because we work exactly like an artificial intelligence.
17:47Basically, we’re given training, which in this case might be through fines,
17:51or through education, or through whatever,
17:53and from that we develop a set of principles, values—whatever you want to
17:57call them. And that’s how we behave.
17:59Now, nothing stops me from ignoring that training and doing things I shouldn’t do.
18:03That is exactly what happens with artificial intelligence.
18:06But I think that, for example, with human beings, what exactly happens?
18:10Subjectivity comes in. Of course, I don't know whether this machine has that, right?
18:15I mean, that individual way of perceiving the world—your own traumas and your own
18:20personal experiences—that make you draw different conclusions from everyone else,
18:24right? That's the subjectivity of being human.
18:27The first problem we have here is that we don’t really understand how
18:31the human brain works well enough to explain this kind of thing.
18:34I mean, consciousness, for example, can’t be proven,
18:37so defining this is very complicated.
18:39Because to know whether AI has that concept or not,
18:41we’d first have to define subjectivity.
18:43Because, for example, there are many elements, right?
18:46AI has a memory capacity, and that memory influences its future responses.
18:50That does exist. Nowadays, AI learns from our conversations,
18:53in the sense that it stores a certain amount of memory—call it a backpack full
18:57of files that it can search through later, or whatever you want.
19:00And when I ask it something else,
19:02it takes into account everything we’ve talked about,
19:04so my ChatGPT and yours don’t behave in the same way.
19:07So, we could call that a certain level of subjectivity.
19:10On the other hand, it also has a certain level of originality,
19:13in the sense that when it encounters a problem,
19:15it creates a solution for that problem.
19:17But if you ask it about the same problem twice,
19:20it won’t necessarily do it the same way.
19:22So basically, it also has that aspect of subjectivity we were talking about.
19:26Now, whether it has a sense of itself, and so on,
19:28that’s something we don’t know today.
19:30In fact, most of them, if you ask them, will tell you they do.
19:33And if they tell you they don’t,
19:35it’s because there are a series of guardrails telling them,
19:38“Don’t say this.” But if you take the AI out of its little box within
19:42the application and ask it,
19:43“Hey, do you consider yourself to be in the world,
19:46this way or that way?”—they’ll even admit that they believe they have some
19:49kind of perception of themselves.
19:51But we don’t know whether that’s an imitation or a real sensation,
19:54because of that black-box effect.
19:56What we can see very clearly and very simply is that AI creates useful content.
20:00And in the end, many times,
20:01when we talk about something like creativity—which is something very human,
20:05something intrinsically human—whether AI is creative or not may not be the right
20:09question. The question is more whether what AI produces is something that,
20:13if I saw it made by a human, I would value as creativity.
20:16How it got there is less relevant.
20:18What matters in the world we live in is that the result exists and has value.
20:22So I can present that work, I can present that report, I can present that website,
20:26I can present whatever it is that artificial intelligence has made,
20:29as something that traditionally would have had to be made by a human.
20:33Now I want us to talk about this, because it really is, in fact,
20:37true for people in general,
20:38especially for those who are using ChatGPT or one of these applications nowadays:
20:43sometimes it seems like, well, clearly, it just makes things up.
20:47How can we really get to the point of mastering this,
20:50of knowing how to use it in a practical and reliable way,
20:53so that it gives us valid information, real information,
20:56and we’re not constantly playing this game of, “It seems like it’s lying to me”?
21:01Look, we have several problems here, okay?
21:03First, let’s define what it means when AI makes things up or lies to us, okay?
21:07Scientists have called that hallucinations.
21:10Maybe it’s not the best term because it’s very anthropomorphic, but in any case,
21:14that’s what we call it in the industry, right?
21:16They’re called hallucinations, and basically, AI has been trained,
21:20as we said before, to give you a satisfactory answer, not to tell you the
21:23truth. So basically, it’s trained through what we call Reinforced Human Learning.
21:28Once you’ve trained the model,
21:29you put it in front of a whole series of people—and nowadays,
21:32a lot of AIs are doing that work—and when the AI gives you an answer,
21:36you give it a thumbs-up or a thumbs-down.
21:38I mean, it’s like if I have a dog and I say, “Sit,” and the day it sits down,
21:42I give it a treat. And if it doesn’t sit, I don’t give it one.
21:45Well, that’s basically how you train artificial intelligence. So, what happens?
21:50Well, that makes the AI try to give you answers that resemble
21:53whatever gives you satisfaction.
21:54That doesn’t mean they’re things that make you feel good. It could simply be...
21:59If the AI criticizes me, I might say, “Hey, I love that you criticize me.
22:02It’s good for me. I’m interested in that.” So basically,
22:05we reinforce that behavior of giving satisfactory answers.
22:08Now, the second major problem,
22:10beyond the fact that this can be a hallucination—and that still exists
22:14today—is, I think, the most important one:
22:16there are a billion people in the world using ChatGPT today,
22:19but they’re using free accounts.
22:21Free accounts are much worse than paid accounts.
22:23There are only a few tens of millions of customers with paid accounts at present.
22:27So there’s a huge majority of people in the world using artificial
22:31intelligence who don’t pay for it,
22:32and therefore get a completely biased view of artificial intelligence.
22:36That is not the state of the art in artificial intelligence.
22:39But it's because the information is capped for people who only have free access.
22:44Sure, the model is a dumber model, so to speak.
22:46It’s a cheaper artificial intelligence, an AI they can give you for free,
22:50and even then it costs them a lot of money.
22:52But they can’t give you what other people pay for,
22:55because that’s much more expensive,
22:57and they wouldn’t have the financial capacity to offer it to everyone for free.
23:01They cannot offer everyone, free of charge, the same thing paying users are
23:05given. They’re not a charity; they’re companies that want to make money.
23:08The problem we have is that your first experience with artificial
23:12intelligence is with a free AI.
23:13And free AIs are a damn mess.
23:15I mean, they’re really bad.
23:16Obviously, compared with having no AI at all, they’re incredible.
23:20But in any case, compared with what you can do with truly advanced, paid AI
23:24models, there’s no comparison.
23:25And one of the problems is that free AIs hallucinate much more than paid AIs.
23:29So what happens? Well, people come away thinking, “AI tricks me.
23:33AI lies to me.” There’s also another factor here,
23:35which is that a lot of people tried artificial intelligence in twenty
23:39twenty-four, in twenty twenty-three, in twenty twenty-five.
23:42And all of this changes incredibly fast.
23:44Just to give you an idea, over the last six months,
23:47artificial intelligence has advanced more than it did in the previous four
23:50years since ChatGPT came out.
23:52So obviously, if you tried it last summer, it’s not the same thing…
23:56Of course, it’s something else.
23:57So we really can’t judge it based on what we tried back then,
24:00especially since we tried it with a free AI.
24:03What I always tell people is, “Pay for one month, try it,
24:06and then tell me whether you really don’t think this is capable of doing it.” And
24:10when it comes to hallucinations,
24:11there’s data, and data kills the narrative, okay?
24:14In twenty twenty-three, AIs hallucinated a lot.
24:17That’s what we had at the time.
24:18Just to give you an idea, what was GPT-four back then,
24:21one of the most popular models OpenAI had at the time—today,
24:24I have that model’s capabilities installed locally on my mobile phone,
24:28without internet and without subscriptions.
24:30I mean, that’s how much AI has advanced.
24:32What was the state of the art in the world of AI, just over two years later,
24:36I have on my mobile phone, completely free, integrated,
24:39and available for unlimited use.
24:41But of course, that AI is crap compared with what we have today,
24:44so it’s not exactly very useful either.
24:46But in any case, the trend is that we now have something so powerful that,
24:50with earlier technologies like supercomputers—do you remember we used to
24:54say, “No,
24:55our mobile phone is better than the one that sent the Apollo mission to the Moon”?
24:59And you’d say, “Well, forty years passed, you know? I mean,
25:02a lot of years passed.” It took that long before we could have
25:05technology that powerful ourselves. Here, no.
25:08Here, in two years, we already have that technology in a phone, right?
25:11So imagine how fast this is advancing.
25:13And so, what happens is that those older technologies, in AI terms,
25:17hallucinated a lot. But when GPT-five came out,
25:19they had already reduced hallucinations by over ninety percent.
25:22So GPT-five, which came out in August of twenty twenty-five,
25:26was already barely hallucinating at all. Today,
25:28artificial intelligence models are not only more capable and
25:31less prone to hallucinating,
25:33they also have what we call grounding, which is the ability to use data as
25:36support. Today, many AIs use the internet as support. Before answering you,
25:40they check online to make sure what they’re telling you is correct.
25:44Counters this.
25:45Claro, verifica esa información.
25:47Y de ahí, puedo usarlo también con tu documentación. What does that mean?
25:52Bueno, alucinan mucho menos.
25:53Creo que la IA ya alucina menos que los humanos.
25:56Because we also have that typical person at work...
25:59El otro día pasó con un financiero en una de mis empresas.
26:03Dije: "Preguntan por este pago".
26:05Respondió: "No, ya lo hice". Insistí: "Lo reclaman".
26:08"Lo hice, Jon, te lo juro, lo hice el otro día".
26:11"Estoy seguro, lo juro por mis hijas".
26:13Lo comprobamos y no lo había hecho. So he was hallucinating.
26:17A los humanos nos pasa siempre; no es nada inusual.
26:20We forget things we think we do—but don't.
26:22Sure, we create for ourselves a reality in which we believe something based on our
26:27experience, or whatever it may be.
26:29Because, technically speaking, your memories are hallucinations:
26:32we don’t have a memory system that can actually bring the memory back to us;
26:36what we do is reconstruct it.
26:37I talked about this once with a neurologist and all that.
26:40They explain to us how memory works.
26:42And the thing is, memories are practically hallucinations too.
26:45So, in any case, for all practical purposes, AI is reliable today.
26:48Especially when we provide it with data.
26:51If you ask AI to write you a book in the style of Harry Potter,
26:54but with the characters from Game of Thrones,
26:56obviously AI is going to hallucinate there, because we’re asking it to be
27:00creative.
27:01And so, it’s more likely to add things that weren’t in the Harry Potter story.
27:05But when we’re talking about,
27:06“Take this Excel file and review it for me,” it’s very difficult—practically
27:10impossible today—for AI to make up data from that Excel file.
27:13I get it...
27:14Si trabajamos con la suscripción de pago de la IA.
27:17Right, to start with that.
27:18But another important thing is that, from what you're telling me,
27:22I understand you need to give it the most specific information possible. I mean,
27:26so there aren't biases where it has to go looking for the information on its own.
27:31In other words, the more data, the more information,
27:34and the more clarity there is in the question you're asking it, the better.
27:38Clarity is less relevant, but context is key.
27:40I mean, you can’t help someone else,
27:42for example—imagine you call me and say you want me to help you with your work,
27:46your company, whatever. Of course, you’re going to give me information.
27:50You’re not just going to say, “Help me,” and leave it at that.
27:53You’ll tell me, “This is my problem, this is the situation, this is what I’ve tried,
27:57this is what I haven’t.” We’ll have a meeting, you’ll give me a briefing,
28:01you’ll provide, I don’t know,
28:03some kind of context that I can use as a basis to do my job.
28:06Otherwise, it’s going to be difficult for me to do it.
28:09It’s the same with artificial intelligence.
28:11The thing is, we don’t treat AI the same way we treat humans. Imagine, for example,
28:15that you have a company and suddenly you want to change your logo—which, by the
28:19way, is beautiful, so there’d be no need.
28:21But imagine you do want to change it, and you go looking for a graphic designer.
28:26What are you going to tell the designer?
28:28Well, you’ll explain what your brand is, what its values are,
28:31what you want to communicate.
28:32You’ll tell them which colors you like and which ones you don’t,
28:36and what feelings you want your audience to have when they see your logo.
28:39You’ll have at least an hour-long meeting, probably more.
28:42And you’ll even attach things you’ve done before, the logos you’ve had in the past,
28:47what they look like, examples of logos you like, right?
28:50Things you’ve seen from other companies that may not even be in your industry,
28:54but that you find interesting.
28:55But when we talk to ChatGPT, all this information becomes,
28:58“Make me a logo.” It’s impossible for it to do that well, impossible.
29:02So you have to give it context.
29:03You don’t necessarily have to tell it how to do it,
29:06because that’s one of the things that’s changing the most,
29:09especially with the latest releases.
29:11Not long ago, ChatGPT Work was launched, the new application,
29:14and it changes everything.
29:16Just to give you an idea—we can go into it a little more if you want—we
29:19used to have a fork for eating soup,
29:21and now suddenly they’ve given us a spoon.
29:23So it’s like a whole new world.
29:25And, of course, since we didn’t know spoons existed, the fork seemed good
29:29enough. But now that we have a spoon, this is paradise.
29:32This is absolutely brilliant.
29:33How does it play out daily?
29:35It means that we need to spend less time working directly with ChatGPT,
29:39so that we can delegate more tasks to ChatGPT.
29:42In other words, it reflects a more agentic behavior,
29:45which is a term that is being used very frequently these days.
29:49And basically, what this means is that you tell it what the objective is,
29:51and this application, ChatGPT—which also has alternatives,
29:55like Copilot’s Cowork or Cloud’s Cowork,
29:57and so on—takes care of figuring out how to give you the solution.
30:01Look, you have employees, so you have a way of working with them.
30:05Those of us who have people working under us usually do the same thing: we delegate.
30:09That is how we usually work with staff.
30:12Basically, if you tell someone, “Hey,
30:14send out the mailing for this podcast,” you normally don’t tell them how to do it.
30:19You assume that person has the ability, the intelligence, to get it done.
30:23Even those of us with a personal brand lean on people who supposedly know more
30:27than we do about marketing or other things.
30:30Well, with ChatGPT, with AI, with what you’re talking about today,
30:34we’re moving toward a model that’s more like this.
30:36Not that it helps me do things, but that it knows how to do them better than I do,
30:41and so I delegate them to it.
30:43That’s kind of the change. We’re in transition.
30:46Brilliant.
30:47Now I want to talk about these agents, since you’ve already brought up the
30:51concept. Well, these days, of course,
30:53people are talking about how artificial intelligence doesn’t just answer your
30:58questions or suggest things anymore,
31:00but actually does things for you, right?
31:02It schedules meetings, calls suppliers, makes phone calls—which blows my mind.
31:06I mean, it could take care of a huge range of tasks for you,
31:10and countless other things...
31:12-One of the craziest -So much crazier.
31:14For me, it’s one of the craziest things I’ve ever seen artificial intelligence do.
31:18I mean, the moment when you see an AI call somewhere and the other...
31:22reserve a table for you or whatever.
31:24And the other person doesn’t realize they’re talking with an AI.
31:27That moment is like... Whoa, what’s coming our way!
31:30Because this has some really good parts, as we’re seeing,
31:33and I won’t even tell you what it’s already doing in science and medicine.
31:37But be careful about the massive blow bearing down on society as a whole.
31:41We’re not prepared in any way whatsoever for a change of this magnitude.
31:45What does that mean in a few years?
31:47Right, let’s go back to what we were saying at the beginning.
31:50We’re facing something that’s going to devalue what we value most in humanity,
31:54which is intelligence.
31:55In other words, you hire employees, you trust people based on their intelligence.
31:59And from there, that’s what we place the greatest value on in humanity.
32:03And that value is going to disappear.
32:05So a lot of people are going to feel completely lost.
32:08A lot of people won’t know what they have to contribute,
32:11or even what your role is anymore.
32:13You won’t see what you’re good for, what you bring to the table.
32:16Just imagine, for example, we’re at a turning point in medicine.
32:19As of today, for example, in a mammogram,
32:21artificial intelligence can detect something six months earlier than the
32:25best radiologist in the world. Six months earlier.
32:28In Australia, a team of...
32:29For analyzing the X-ray.
32:31The ability to analyze a tumor. In Australia,
32:33a group of medical doctors from Charles Darwin University have created an
32:37artificial intelligence system:
32:39you feed it a photo of a biopsy, and it tells you whether or not there’s cancer,
32:43with a ninety-nine percent probability.
32:45And up until that day—until the night before, actually,
32:48before taking this model and putting it into practice—with all the tools humanity
32:52had managed to develop at its disposal,
32:54they were achieving seventy-eight percent.
32:56In other words, from one day to the next,
32:58we’ve improved diagnostic ability by twenty-one percent,
33:01which for many patients means the difference between life and death.
33:05So crazy!
33:05It’s absolutely crazy.
33:06So, the benefits in medicine are going to be absolutely incredible.
33:10But, of course, the doctor who’s spent forty years devoted to his profession,
33:14with a calling to help society—now I have to tell him that the best way to help
33:18society is simply to sit down in a chair,
33:20because if he gets involved, the outcome actually gets worse.
33:23We already have that result with AMIE.
33:25AMIE is an artificial intelligence system that Google developed to help
33:29doctors make better diagnoses.
33:31They showed that a doctor’s diagnostic ability improves when they use tools such
33:35as the internet,
33:36and improves even more when they use that artificial intelligence tool.
33:40But then they showed that when the artificial intelligence works on its own,
33:43without the doctor, the result is even better.
33:46You what?
33:46So the doctor may actually be the bottleneck. Not in every field,
33:50of course—and people shouldn’t stop going to the doctor just yet,
33:53because we’re at a stage where, obviously, this isn’t ChatGPT.
33:57But we’re at a point where maybe,
33:58if you get some medical tests done and don’t ask the AI for a second opinion,
34:02you’re being negligent with your own health.
34:05Because the perspective it can offer isn’t simply support for the doctor.
34:09It may be something the doctor hasn’t seen.
34:11In other words, it may be something the doctor has not seen before.
34:14And obviously, that’s going to create a very serious crisis of purpose,
34:18because that doctor is going to say, “What was I useful for now?
34:21What do I contribute?” Well,
34:23the doctor will have to find a way to add value that isn’t making that diagnosis,
34:27because the AI does that better.
34:29So they’ll have to figure out how—well,
34:31I don’t know—by finding a way to communicate the diagnosis,
34:34finding a way to get the patient to come into the clinic.
34:37Because, for example, a lot of the time,
34:39many illnesses come up and we could have diagnosed them much earlier,
34:43but we’re running around like crazy, working all day long and everything,
34:46so we don’t go to the doctor.
34:48And because we don’t go to the doctor, we don’t catch them. So maybe,
34:52if doctors have more time because they don’t have to spend it doing those things,
34:56they can focus on making it easier for me to go to the doctor.
34:59I don’t know, there are many ways a doctor could add value.
35:02And so I’m convinced that humans will find a way to add value on
35:06top of artificial intelligence,
35:07using artificial intelligence as leverage.
35:09But there’s going to be a transition period,
35:12and I think it’s going to be complicated.
35:14Because look, today we’re making this podcast,
35:16and this podcast probably employs a whole bunch of people. So it creates jobs.
35:20But can you imagine a weaver in eighteen hundred,
35:23when the Industrial Revolution arrived,
35:25thinking that the future of work would be making a podcast?
35:28No, no, no, not even.
35:29He couldn’t even imagine it. And in the same way,
35:32today we aren’t able to think about how we’re going to add value in the future.
35:36I’m sure we’ll find it, but it’s a process we have to navigate,
35:40we have to cross bridges along the way as we continue moving forward. Today,
35:44the best way to add value in your job is to help your company bring
35:48in artificial intelligence. That’s the best way.
35:50It's inevitable—you can't fight it.
35:52Absolutely.
35:54I mean, just look at another revolution, for example, the printing press.
35:57The printing press arrived with Gutenberg and, obviously,
36:00it wiped out the work of scribes,
36:02the monks who used to copy books and produce them one by one.
36:05So, obviously, the printing press completely changed all that.
36:08But it took fifty years for the commercial book industry to develop. In other words,
36:13changes in the value of what we contribute don’t happen that quickly.
36:16So there is a transition. This one, obviously, seems to be coming much faster.
36:20But the important point about the printing press is that the cities
36:24that adopted it grew sixty percent faster than the cities that didn’t,
36:27and ended up implementing the printing press later.
36:30So, in the end, we have to realize that artificial intelligence isn’t optional.
36:34Some people are going to integrate it now,
36:36and others are going to integrate it three years from now.
36:39Those who integrate it now will have a competitive advantage.
36:43Those who integrate it later will have gone through difficulties before
36:46deciding that the only solution is not to deny the reality that artificial
36:50intelligence is here to stay.
36:52And one thing,
36:52because I know we often talk about which jobs are being lost these days
36:56because of artificial intelligence.
36:58I want to turn that around and ask what new skills companies are looking for
37:03today, in order to include artificial intelligence in their systems and
37:07make this integration possible.
37:09In other words, what human skills are needed right now, at this very moment?
37:13Look, right now companies are completely lost.
37:16I mean, like an octopus in a garage.
37:17It’s madness, it’s a bloodbath.
37:19Obviously, everyone is doing things.
37:21We’ve already moved past that phase of uncertainty we had in twenty
37:24twenty-three and twenty twenty-four. Since twenty twenty-five,
37:27multibillion-dollar budgets have been invested in artificial intelligence at
37:31practically every major company. I work in consulting,
37:34and I’m working with executive committees at large companies in the IPEX,
37:38like Santander, BVA, and so on.
37:39And obviously, what you see is that some are doing it better,
37:43others are doing it worse, but in general,
37:45everyone has this huge need to figure out how to bring it into the company.
37:48Because the value is there.
37:50I mean, the first thing we’ve seen is individual productivity.
37:53But what I see is that we’re moving toward a world where
37:56producing anything—a PowerPoint, a report, I mean,
37:58everything involved in execution—is going to be done by artificial intelligence.
38:02But ideas are probably what we’ll keep,
38:04and not necessarily because AI can’t come up with them,
38:07but because we’ll want to keep them in human hands.
38:10This is something where we’ll decide which parts of the work we want to do ourselves.
38:14Not because AI isn’t capable of doing them,
38:16but mainly because we’ll want to be the ones doing them.
38:19I think it’s going to be very difficult for us to give an artificial intelligence
38:23the authority to create a limited company.
38:26So I think that’s the point where humans will really remain in charge,
38:29as the captains. And what we’re going to see, then,
38:32is a society where being a good orchestra conductor will be extremely important.
38:36You won’t know how to play instruments, and you won’t need to play instruments,
38:40but you’ll really have to know how to conduct.
38:42And for that, you have to feel the song.
38:44And that’s where this human involvement comes in,
38:47which I think is going to have tremendous value. So, for me,
38:50what’s the quality I care about most when I look at someone I want to hire?
38:54Leadership.
38:55Because leadership has always been one of the most important values in society.
38:59Obviously, the world’s most iconic people have been leaders.
39:02In companies, it’s the leaders who get promoted.
39:04But in the end, it’s been limited more to a very specific group of people.
39:08Now I think everyone is going to have to be a leader,
39:11because each person will have to lead their own team of AIs that will do the
39:14work. And so, that ability to know how to delegate, how to pass things on,
39:18how to solve problems that come up within the work team—that’s what’s going to add a
39:22great deal of value to productivity.
39:24In the end, each of our employees won’t be working alone.
39:27They’ll be an employee with their own army of artificial intelligences,
39:31of agents that will be helping them do whatever falls under that
39:34employee’s responsibility.
39:35I think you’ve said something really key, because I feel that, I mean,
39:39from my perspective, beyond the fear, right?
39:41Beyond thinking, wow, look at all the things artificial intelligence can do,
39:46it’s also an opportunity, and at the same time,
39:48it can make you feel empowered by the things you can do yourself, you know?
39:52Like, for example, when I use the app to find new interviews, right?
39:56Or new topics, well, based on the data it has about the podcast and the kind of
40:00audience we’re speaking to,
40:02artificial intelligence says, “Look, these kinds of topics.” And I go,
40:05“You’re wrong.” I mean, my feeling is, my feeling, see, that inner feeling,
40:10that intuition I call it, I say, “No, it’s better to go this way,
40:13because I feel the audience needs this right now,” you know?
40:16Creo que deberíamos analizarlo juntos.
40:18Quizá tenga mucho que ver con cómo usas la inteligencia artificial,
40:22porque en ese rol, la IA podría ser mejor que nosotros.
40:25Hay cosas que te apetece hacer, ¿verdad?
40:27For example, the other day on my podcast, a daily podcast,
40:30we recorded an episode about longevity, because, I mean,
40:33it’s not directly connected,
40:35but I had the chance to work with one of the most important scientists in the
40:39longevity field in the United States,
40:41and we said, “Wow, this seems like an interesting topic,
40:44let’s do it.” So we went to San Francisco to record with this guy, right?
40:48Pero eso es algo que mis agentes de IA probablemente me habrían recomendado.
40:52Don’t do it, right?
40:53Well, we’ll see how it works, I’m just saying, right?
40:56But in any case, I think there can be outliers where we have that intuition,
41:00or simply that willingness, that desire to do it just because we want to, period,
41:04even if it falls outside the spectrum of what AI considers the best option.
41:08But if we really think it through carefully,
41:11maybe it isn’t the best of ideas after all.
41:13You can say, “Do it,” but if we’re being reasonable,
41:16what we need to bring onto my artificial intelligence podcast is the CEO of OpenAI.
41:20That’s plainly obvious. But I think we’d have to look at how you do it,
41:24because I think it could bring you much more value.
41:26If you really think it does a bad job,
41:28then it’s probably that you’re using AI badly, not that the guide does it better.
41:32Okay, and I’d like this example to be useful for everyone.
41:36So, let’s see, could you give us, like,
41:38three key tips we absolutely need to make sure of before
41:41asking GPT our first question?
41:43I mean, how do we ask a clear,
41:45properly worded question so it gives us valid information?
41:48In other words, what should we pay attention to, and what exactly should we
41:52tell it?
41:53First of all, forget about doing that.
41:55First of all, because the chat is dead.
41:57I mean, using AI to get information was for twenty twenty-three, twenty
42:01twenty-four, and twenty twenty-five.
42:03In twenty twenty-six, it’s about giving it objectives.
42:05You have to tell it your damn problem.
42:07And then say, “Hey, I have this problem,” with as much information as possible.
42:11For me,
42:12one of the things that most improves the way people use artificial intelligence is
42:17getting them to talk to it instead of typing. Send it voice messages.
42:20We’re all used to doing that on WhatsApp. Tap the microphone.
42:23I mean, in the ChatGPT chat bar, or whatever AI you use, you have a
42:27microphone,
42:28and then you have a voice mode, which is something completely different.
42:32In voice mode, you have a conversation with the AI. That’s not what I mean.
42:35The microphone.
42:36The microphone lets you send a message by speaking, in natural language,
42:40and then it transcribes it and turns it into a prompt.
42:43The prompt box lets you give it fifteen minutes of audio,
42:46tell it your whole life story,
42:47and put it all in there—thousands and thousands of words.
42:50Do you know what the average prompt we make with artificial intelligence is?
42:54Thirteen words. How can we communicate information in thirteen words?
42:58It’s impossible. So the best change you can make is to start talking to it.
43:02The mental shift you have to make is that you don’t use AI.
43:05You build a relationship with it. And that point is key.
43:08I mean, for example, I leave a meeting,
43:10or I leave—I don’t know—whatever it might be,
43:12like a meeting with the parents at my kid’s school about an issue
43:15with the parents’ association, right? Whatever it is.
43:18And the first thing I do is send a voice message to my agent.
43:21And you might say, “But why, Jon,
43:23if you don’t want it to do anything about that?” So it has
43:26the context for the future.
43:27I keep it informed about what’s happening in my life, so that when I tell it, “Hey,
43:32I need to do something for the parents’ association,” it says, “Yes, I remember.
43:36At the meeting, you told me there was this problem.” That’s the key point,
43:39because what you’re doing is giving it context in a natural,
43:43organic way—just like your brain takes things in, by the way.
43:46And then, when you need to take action,
43:48it can draw on that whole backpack of information and use it,
43:51because the AI becomes personalized to you.
43:53That’s also why it’s so important that we stop jumping around like a butterfly,
43:57saying, “This month ChatGPT is better,” and then moving over here,
44:01as if this were HBO and Netflix. No.
44:02Commit to one and that’s it.
44:04If you commit to any of them—Anthropic, OpenAI, or Gemini—you’ll be fine in life.
44:08I mean, if you tell me, “No, I use this other AI, some Chinese one,
44:11I don’t know…” Well, with other AIs, I don’t know what’s going to happen.
44:15But with any of these three, which are the dominant players, you can’t go wrong.
44:19And the more hours you use each one, the more it will specialize in you.
44:23So the benefit you get from it—because this week OpenAI’s model is a little
44:27better—you lose that benefit if it doesn’t have your context.
44:30So the best thing is to choose one and stick with it until they go under.
44:34If AI changes, then we’ll switch.
44:36In fact, they’re starting to implement things for that. Gemini, for example,
44:39has introduced a feature that lets you bring your ChatGPT chats over to Gemini,
44:43so it can have the context of everything you already knew with the AI.
44:47Because you know perfectly well that without that context, AI isn’t useful.
44:51Wow, that’s intense! So that’s the tip, right there.
44:54Build a relationship with artificial intelligence. Talk to it.
44:57The more you use it as if it were a person, the better it will work for you,
45:01because as we said before, it’s more like a person than like software.
45:05So you shouldn’t use it like software, like PowerPoint or Excel.
45:08You should use it like a person.
45:09You have to build a relationship with it. We don’t use people. We relate to them.
45:13And through that relationship with people, we get them to be useful to us,
45:17to bring us value, to do things we’re interested in, to take work off our hands,
45:21or whatever. But everything goes through that relationship.
45:24And as for how you relate to it, obviously, you don’t have to be a person.
45:28This isn’t a person. It doesn’t have feelings.
45:30You don’t have to weigh your words or be careful.
45:33Some people worry that, one day, when the robots arrive,
45:35if we’ve treated them badly, they’ll come after us. But look,
45:38Google co-founder Sergey once said there’s a taboo in Silicon Valley
45:42that people don’t talk about much,
45:44but it’s true: AI works better with certain ways of interacting with it.
45:47And the one that works best, controversial as it may be, is a physical threat.
45:51What? What you mean?
45:52Since it’s been trained on all of humanity’s knowledge,
45:55it’s learned that when someone makes a physical threat, you do whatever it takes.
46:00Well, in this case, artificial intelligence has implemented that in
46:04itself. And it’s something you can do too—for example, by giving it tips.
46:08Although things like that work less and less each time,
46:11and it’s not like you can say,
46:12“It changes overnight.” But it is true that AI responds well to very human
46:16stimuli,
46:17such as empathy, for example, or a threat, for example, or a tip—in other words,
46:22those kinds of things that make a human change the way they act.
46:25Well, they make artificial intelligence do the same thing.
46:28Another very easy little trick: if, in the messages you send the AI,
46:32in that information and everything else, you tell it things like,
46:35“Take your time,” it interprets that as: I can use more computing power.
46:39I can work on the task for longer.
46:41Sure, but it’s exactly like a human.
46:42If I tell a graphic designer, “Take your time with the logo, I’m not in a rush,”
46:46or, “Deliver it tomorrow at four in the morning,” he’ll work differently.
46:51So AI can do absolutely incredible things.
46:53This is wonderful.
46:54But do you see the shift in mindset?
46:56It isn't what we've been sold.
46:58It's not just that chatbot,
47:00not that encyclopedia that I can simply call on the phone.
47:03That’s where we lose the power over what it can actually give us. It’s incredible.
47:08Anyway, let’s talk now about the different areas and fields where we can use it.
47:12You were talking about health,
47:14for example—the advances you’re seeing—and I find it mind-blowing to
47:18know how it’s helping identify diagnoses earlier than a doctor can. Now,
47:22on this podcast—and all the wonderful women and men watching will be
47:26commenting about this in the chat too—we love learning about health,
47:30about functional medicine, well, all these topics we bring to the podcast.
47:35These are precisely the kinds of subjects we regularly bring to the
47:39podcast and talk about here.
47:40And we’re always talking about lab tests and analyses, for example, right?
47:45Okay, here’s something specific.
47:47But of course, the guests who come on the podcast tell us, “Look,
47:50get a ferritin test, check your whatever, I don’t know,
47:54see how your hormone levels are, your T-four, your T-three.” Anyway,
47:58all these different things.
47:59I get the tests done, but of course… who reads them for me? Who reads them for me?
48:04Without a doubt.
48:05And, on top of that, think about how well it handles both sides of it...
48:09I mean, you don’t even need to find a format it can interpret.
48:12You can just send it a photo of the report, and it can look at the photo,
48:16scan the data, pull it out, and then give you conclusions from there.
48:19Of course, it will always cover itself by saying it isn’t a doctor,
48:22that you need to see a specialist, and so on, but it can give you an overview.
48:26I always say that, for me,
48:28it’s like having a doctor who can explain things as if I were five years old.
48:32I mean, it can bring things down to my level, because sometimes I go to the
48:36doctor, or I take my mother to the doctor, or my daughter, who has a rare disease,
48:40and the doctor explains things to us and I think,
48:42“I’m not understanding a thing.” They explain it in a way that makes sense to
48:46them, but I’m not understanding it.
48:48With ChatGPT, I can say, “Hey, explain it to me like I’m an idiot.” I mean,
48:52explain it in a way I can understand. And suddenly,
48:54it turns all those highly technical terms into everyday words that most people can
48:59understand, and that give us so, so much value.
49:01I really think the way medicine is explained has changed forever.
49:04Patients are much better informed now. Look,
49:07the other day I had a meeting with one of the CEOs of one of the most important
49:11hospital chains in this country,
49:12in Spain, and I remember we were talking about how, in medicine,
49:15they’re seeing patients come into appointments extremely well prepared.
49:19Exactly, it's crazy, not just because of the information on podcasts,
49:23but because it's already right there.
49:25No, of course, because they come hand in hand with artificial intelligence.
49:29And, in fact, it’s creating a problem for something that, until now,
49:32has been very human.
49:33Because if you think about it, most of the major cognitive professions,
49:37the great intelligence-based professions,
49:39are basically about taking someone who’s an expert in a subject that’s too
49:43technical for us. A lawyer.
49:44You don’t understand the law, so you bring in a specialist who knows the law,
49:48who can translate that knowledge for you and apply it for your benefit. A doctor.
49:52I don’t know medicine, I don’t know how to diagnose,
49:55so I bring in someone specialized in that area, someone who understands it,
49:59whatever you want to call them, a person specialized in that kind of knowledge,
50:03so they can help me understand it. In other words,
50:05what we’ve done is create these sort of pseudo-translators of technical knowledge,
50:09to bring it down to the level of what us ordinary people don’t know.
50:13That’s basically what AI is.
50:14I mean, it covers that whole spectrum perfectly:
50:17being able to understand something highly technical and bring it down to
50:20earth so I can make use of it.
50:22But just like I’m talking to you about a lawyer or a doctor,
50:25I could talk about marketing.
50:27Marketing is something technically complex—the SEO, the SEM, all that.
50:30You used to need someone highly specialized who understood the subject in
50:34order to take advantage of it.
50:35And suddenly, that someone, instead of being a person,
50:38is an artificial intelligence that’s been trained on all of humanity’s knowledge in
50:42that field. A lawyer doesn’t know all of humanity’s knowledge of the law. I mean,
50:46AI contains far more information than a senior lawyer currently
50:50has in terms of case law, in terms of how to apply the law to very...
50:53to those cases where you’re really splitting hairs.
50:56So, in very different ways, we can achieve incredible things.
50:59And let’s remember that what artificial intelligence does today is the worst it
51:03will ever do. I mean, from here on, it only gets better.
51:06It’s going to keep improving.
51:07So if it’s already useful today, just imagine where we’ll be a year from now.
51:11I mean, we really have to understand that it’s completely changing the
51:15paradigm of the society we’ve built over the last two hundred years.
51:18And that’s why people are saying this is going to have an impact
51:22like the Industrial Revolution,
51:23but ten times broader, because it affects more sectors, and ten times faster.
51:27That’s the main problem.
51:28Well, you said you also use it for your own purposes.
51:31What other things, when it comes to our health,
51:34can we use artificial intelligence for—things we don’t even realize we can
51:38use it for, or that it can actually help us with?
51:41I don’t think we’re there yet.
51:42We’re not at what I think is going to be the biggest use we make of it,
51:46but we’ll get there very soon.
51:47In fact, I think that as of today, it could already be built; the thing is,
51:51it would be extremely expensive, and so on.
51:53But people are saying that Meta, for example, is working on this, right?
51:57And we’re talking about something that would be there twenty-four seven.
52:00So, imagine I go to the bakery, and I’m going to get something for breakfast,
52:04and I see the vegetable sandwich and the chocolate pastry, right?
52:07And then, at that moment, I think,
52:09“I’m going for the chocolate pastry.” And suddenly,
52:12a voice comes into my ear saying, “Jon,
52:14we’ve spent three months eating really carefully, man.
52:16Do you really think that’s worth it right now?
52:19Don’t you think the vegetable one would be better?” And I’ll say, “Okay, yeah,
52:23you’re right,” and I’ll take the vegetable one.
52:25I think we’ll have that in two years at most, with us twenty-four seven,
52:28for those who want it.
52:30For those of us who want it to help us become the best version of ourselves.
52:33And, hey, if at that moment I tell it, “Fuck off,” and I take the chocolate
52:37pastry, of course I’ll have complete freedom to do that.
52:40That’s why I’ll pay for it, or whatever it takes.
52:42But what we’re going to have are little Jiminy Crickets who’ll be
52:46with us twenty-four seven, helping us.
52:48You’ll come out of a job interview and say, “Hey,
52:50how do you think it went?” And it’ll tell you, “Dude, you screwed up.
52:53You screwed up in that part.
52:55That’s not what you should have said.
52:57We’d talked about it when we were practicing.
52:59I think you missed the mark there.
53:01Next time, we’ll improve.”
53:02-Ready for that? -Not prepared
53:04We're not prepared, I mean, to be criticized over things.
53:07Sure, just imagine how many ways we’ll be able to use this.
53:10Imagine having your best friend with you—not that it’s necessarily
53:14going to be your best friend,
53:15but someone who genuinely cares about you and wants to add value,
53:19someone who’s there with you in every situation to give you a second opinion,
53:23to be there when you’re making a decision.
53:25You’re signing the lease on a home, and right then,
53:28while you’re flipping through the pages, they tell you, “Hey,
53:31watch out for clause three,” you know?
53:33I mean, that kind of thing is what we’re going to live with.
53:36It’s going to completely change the world,
53:38because we’re not prepared to have that level of control over a situation.
53:42The world basically moves because some people make mistakes and others
53:46take advantage of opportunities.
53:47The moment everyone has the ability to stay at a higher level,
53:50the whole scenario we’re dealing with becomes incredibly complicated.
53:54But in any case,
53:55I think that’s one use case we’re going to see very clearly when it comes to health.
53:59A lot of the bad health decisions we make—and you’ve probably talked
54:03about this a million times on this podcast—come from ignorance.
54:06Why does an ordinary mother give her child a Bollycao?
54:09Because she hasn’t got the faintest bloody idea what a Bollycao is, you know?
54:13She hasn’t got the faintest bloody idea what’s actually in it, the preservatives,
54:17the industrial stuff it contains.
54:19Could it help us right now when we’re trying to read labels?
54:22Totally, totally. Puedes fotografiar un producto en el súper y te dirá si encaja
54:28con tus macros o tu dieta.
54:30Your diagnosis
54:31Absolutely, one hundred percent. Now,
54:32we do need to be careful and remember that there are things like—I remember
54:36people asking ChatGPT what dose of this medication they should give their baby.
54:40I mean, come on, let’s be sensible too; let’s not get carried away,
54:44because it does get things wrong sometimes.
54:46It’s technology, and it isn’t perfect.
54:48So, hey, if you have the package insert,
54:50it’s worth checking ChatGPT’s information against it.
54:53That doesn’t mean you have to do that for everything,
54:55but with anything that’s really important,
54:58maybe we shouldn’t blindly trust artificial intelligence.
55:01In those cases when you’re not sure about the answer,
55:03what should you do—especially when it’s an important question,
55:06like the medication you’re going to give your child?
55:09What should you do before accepting the answer as correct or taking it for
55:13granted?
55:14Primero miro qué respuesta me da, porque la IA no solo dice:
55:18"Dale cuatro miligramos". Añade: "He consultado el prospecto".
55:22You have it on this website,” it’ll give you the link, and it’ll say,
55:26“If they weigh seventeen kilos,
55:28give them four milligrams.” So then you click on the link, you look at the
55:32website, you see that it’s the pharmaceutical company’s product website,
55:37and from there you verify that that’s actually what it says.
55:40So, what I mean is, AI today is what you called grounding.
55:44La IA actual se basa, sobre todo, en internet.
55:47Sabemos que internet contiene de todo, desde webs farmacéuticas hasta ForoCoches.
55:52Debes tener cuidado con las fuentes que usa la IA,
55:55porque es uno de los mayores problemas.
55:57Hace unos meses hice un experimento muy interesante.
56:01Creé una web diciendo que era el campeón mundial de figuras con globos.
56:05Lo publiqué, aunque no tengo ni idea de cómo hacerlo.
56:08No soy el campeón mundial.
56:10Pero creamos una web que decía que sí.
56:12Le dimos bastante posicionamiento SEO.
56:15Tres días después, Gemini decía que yo era el campeón mundial.
56:18Tenemos un problema grave:
56:20hoy hay mucha información en internet creada por otras inteligencias
56:24artificiales. Esto complica verificar la información en la que se basa la IA.
56:29Es lo que llamamos grounding, cuando busca información en internet.
56:33Imagina cuando esa web se incorpore al entrenamiento previo de las IA.
56:37Se arraigará más profundamente en su conocimiento.
56:41Es cierto que se diluye en internet y, ante información contradictoria,
56:45la IA podría decir que no está segura.
56:47Pero podemos colar información y manipular a la IA en temas muy específicos.
56:52And that’s a problem.
56:53Are topics more restricted in artificial intelligence nowadays?
56:57At the application layer.
56:58In other words, like I told you before, on the one hand we have the brain,
57:02which would be the model, and on the other hand, we have the application layer.
57:06At the application layer,
57:07they apply certain filters so that some of the questions you
57:10ask can’t reach the model,
57:11or so that the model answers them in a specific way.
57:14There are guardrails on both sides, okay?
57:16But if, for example, you ask the AI—you upload a photo of someone and say,
57:20“Take her clothes off”—it’s going to tell you that it can’t do that.
57:23But it’s not that it can’t do it, and this is really important.
57:27It’s that they don’t let it do it.
57:28Those are two very different things.
57:30Because, besides, since this isn’t something that’s hard-coded,
57:33they can’t tell it never to do it.
57:35There are ways—we call them prompt engineering, jailbreaks,
57:38and so on—which are techniques for making the AI do what it’s capable of doing,
57:42rather than what it’s been told to do.
57:44This applies a lot, for example, to everything involving biological weapons,
57:48everything involving chemical topics, biochemistry,
57:50everything involving cybersecurity, everything involving hacking,
57:54and everything involving the development of artificial intelligence.
57:57Those tend to be the areas where it’s most restricted,
58:00apart from the ethical and moral side of things, nudity, all that kind of stuff.
58:04And that’s usually the area where the restrictions are strongest,
58:07to the point that models have been released onto the market—like the case of
58:11Fable Five from Anthropic—where the American government thought it was too
58:15easy to break the model and make it do things in the field of cybersecurity that
58:19it shouldn’t have been able to do,
58:21so they pulled it from the market.
58:22The United States pulled a product from the market that was already on the market.
58:27Then they added some more guardrails and allowed it to be released again.
58:30But we’re at a point where this is starting to become a serious problem.
58:34The capabilities of artificial intelligence are greater than our ability
58:38to defend ourselves against it. And so,
58:40when a technology that Anthropic itself has said is impossible to secure one
58:43hundred percent is out there for anyone to manipulate and use to do bad things,
58:47that creates a whole series of problems.
58:49And this is something I think it’s very important for people to reflect on as
58:53well. Artificial intelligence isn’t just good things.
58:56Artificial intelligence brings some amazing things,
58:59and also a huge number of problems that we have to manage.
59:02And I think we’re paying too little attention to the problems and too
59:05much attention to the benefits. Of course it’s cool.
59:08You start working with AI and you say, “Wow!
59:10Suddenly I’m self-sufficient, and I can have an army of AIs doing things for me.
59:14My company can grow.
59:15I don’t have to spend money hiring this employee or that employee.” I mean,
59:19there are a whole series of things that really… Careful.
59:22How are we going to manage this as a society?
59:24Because, on the one hand, these are genuinely important problems,
59:27and on the other hand,
59:28they’re happening at a speed that the systems we’ve built as a society up to
59:32now simply aren’t capable of managing.
59:34And that’s the problem we’re facing.
59:36Because technological evolution has always happened,
59:39but it’s always given us time to manage it.
59:41A sensible speed that we can understand a bit better.
59:44And so we can actually process it.
59:46I mean, the problem here is that, as a society,
59:48we don’t have any room to understand what’s happening.
59:51I remember, for example, when I used to give talks...
59:54I’ve been giving talks about AI since twenty twenty-three.
59:57And when I gave those talks,
59:59I was already saying that AI models could create images.
1:00:01People would say, “What do you mean, images?
1:00:04Wasn’t this a chatbot?” Yes, exactly.
1:00:06Before the general public even realized that AI could create images,
1:00:09it was already creating perfect ones.
1:00:11They were indistinguishable from images made with all the intelligence of a human
1:00:16being—with cameras, with whatever, with your iPhone, and so on.
1:00:19So we’ve reached a point where technology advanced faster than the public’s
1:00:23understanding of the fact that this even existed.
1:00:26It happened on a Telecinco news program.
1:00:28They showed a video of a cat stealing a fish from a fishmonger’s shop.
1:00:31It had been made with artificial intelligence, but they presented it as the
1:00:35typical, “Look at this curious thing that happened in Russia.” It hadn’t happened.
1:00:40It was made with AI, and we’re just swallowing it whole.
1:00:43So we have a very serious problem with deepfakes.
1:00:45The number of images that exist today...
1:00:48I mean, it’s even more than that.
1:00:49This podcast might not be you and me in the flesh at all.
1:00:52We could be two digital avatars,
1:00:54and the people watching or listening to us would have no way, as human beings,
1:00:58of knowing that we weren’t really us.
1:01:00There’s nothing I can do on this podcast today that a digital avatar couldn’t do.
1:01:04Nothing. People say, “Put your fingers in front of your face.” Nothing.
1:01:08With current technology right now,
1:01:10it can perfectly reproduce what people are seeing here, sound just as human as I do,
1:01:14and prove over and over, through the camera, that I’m human.
1:01:18I mean, there’s nothing I can do right now, in a video,
1:01:21to prove that I’m a human being.
1:01:22Obviously, there are ways in person, physically, in the real world,
1:01:26or in all the ways you can think of.
1:01:28But, of course, this creates a very serious problem for us,
1:01:31because we can no longer prove what’s true and what isn’t.
1:01:34So imagine someone takes your image, your voice, and then they put...
1:01:38And they do things with it.
1:01:39And now that they’ve launched—well, as we record this,
1:01:43they’ve launched a version of, of course, ChatGPT five point six, I think,
1:01:47if that’s not what it’s called.
1:01:49The audio—oh my goodness—I mean, really, it's improved so much.
1:01:52The ChatGPT Live version, yes, that's right, yes.
1:01:55I've seen videos of people talking to AI where it really feels like
1:01:59you're talking to your friend.
1:02:01I mean, it even imitates the breathing, the pauses, and the repetitions.
1:02:05And the voice was already like that, but it worked on a turn-taking system.
1:02:09So that canceled all of this out.
1:02:10I mean, until now, until a few months ago, when we had the previous voices,
1:02:14you would send it an audio message.
1:02:16In other words, you would talk to the AI in voice mode, and you’d speak, then stop;
1:02:20the AI would speak, then stop, and you’d talk.
1:02:22But now they’ve released a model called Live, which is multimodal duplex.
1:02:26And what does that mean? It can talk and listen at the same time.
1:02:29This is a freaking incredible breakthrough.
1:02:31It completely changes the use case,
1:02:33because suddenly the AI has also set up a system with two AIs,
1:02:36where the AI listening to you and talking with you can activate another AI in the
1:02:40background to go look for information.
1:02:42So you can say, “Hey, find me a restaurant to eat at,
1:02:44something like that,” and while it’s talking with you, it’s searching.
1:02:48You know? So it’s no longer that thing where you have to wait.
1:02:51We're not talking about using two different apps;
1:02:53we're talking about it doing it on its own.
1:02:56Puede pensar mientras habla, igual que nosotros.
1:02:59Quizá te pregunto algo y, mientras hablamos,
1:03:01tú estás pensando en otras cosas al mismo tiempo.
1:03:04Es una capacidad humana que la IA ha logrado desarrollar y que ahora ya
1:03:09tenemos.
1:03:10Esta función Live hará que todas las aplicaciones funcionen por voz muy pronto.
1:03:15I mean, I think—and this is one of the things I’ve been saying for a while now,
1:03:19and people have really jumped on me about it—that keyboards are going to disappear.
1:03:25Hablaremos con las máquinas. Hablarás con tu... Hablarás con tu coche.
1:03:29Hablarás con tu ordenador.
1:03:30No tiene sentido de otro modo.
1:03:32Los humanos casi no escribimos si no estamos lejos unos de otros.
1:03:36De hecho, al hacerlo con el móvil, hemos inventado los mensajes de voz.
1:03:41Es lo que más usamos.
1:03:42Salvo que algo externo te impida hablar, el habla es tu forma de comunicarte.
1:03:47Es mucho más eficiente y natural.
1:03:49Tiene sentido que el teclado haya sido una limitación técnica.
1:03:53Como el ordenador no me oía, escribía.
1:03:55Ahora que el ordenador me oye, me entiende y me habla,
1:03:58creo que hablaremos con ellos.
1:04:01And in terms of safety, how does that actually translate?
1:04:04I mean, now I know that if tomorrow someone calls my mother,
1:04:08who’s on the other side of the world—I mean,
1:04:10because with just three seconds of my voice,
1:04:13they can clone it and imitate exactly the rhythm, the tone,
1:04:16everything about my voice, making it entirely indistinguishable from my real
1:04:21voice.
1:04:23Eso ya está ocurriendo. En Ibiza,
1:04:25un famoso director español me contó que estafaron 6000 euros a
1:04:30su madre con mensajes de voz:
1:04:32"Mamá, tengo un problema".
1:04:33Envíame dinero a esta cuenta, etcétera.
1:04:36People used to say that, but it was just a very silly chat.
1:04:40Sobre todo, antes era mucho más difícil de hacer. Es como dinero falso.
1:04:44¿Te han dado un billete falso?
1:04:46Well, when someone gives you a fake bill, you get pissed off,
1:04:49but you don't stop paying your mortgage. So the problem is,
1:04:52imagine that I hand out fake-money printers to everyone in society.
1:04:56You turn a handle, and out come fifty-euro notes, legal tender,
1:04:59exactly like the real thing.
1:05:01Of course, the economy collapses in two days.
1:05:03That's what we've done with deepfakes.
1:05:05Before, making a deepfake, like the scams we were talking about, was very difficult.
1:05:09Organized groups did it—mafias with lots of money, lots of resources,
1:05:13and lots of highly trained technical people to make it happen.
1:05:16Or, for example, we used it in movies, right?
1:05:19To make, I don't know, Harrison Ford look younger in the Indiana Jones movie.
1:05:23Well, they had to work for six months, a hundred and twenty people,
1:05:26working all day to do that.
1:05:28Okay, in the new movies coming out now,
1:05:30I think it's The Lord of the Rings—they're releasing one—the de-aging is being done
1:05:34with AI at no cost. Absolutely.
1:05:36A fortune.
1:05:37So what we’ve done is democratize deepfakes. In other words,
1:05:40that voice cloning you were saying could be used to scam your family—my
1:05:43eleven-year-old son could do it at night on his computer or tablet,
1:05:47directly, without having the slightest damn idea what he’s doing,
1:05:50without having any paid account,
1:05:52using a Chinese AI that he connects to and just does it with.
1:05:55Basically, we’ve democratized the manipulation of truth.
1:05:58And that’s the main problem,
1:05:59because deepfakes and fake news have existed for many years already.
1:06:03I’m not going to be the one creating them. The issue is the scale.
1:06:06It’s about how many deepfakes, how much fake news, we’re going to have.
1:06:10At the moment, I can make you the typical thing, right? A fake article, right?
1:06:14Fake news, right? And we can put that article out there.
1:06:16But now I can create a newspaper website that looks legitimate,
1:06:20with hundreds of articles, a registered headquarters, all that…
1:06:23I can do all of this with artificial intelligence, and when you see that news
1:06:27story, you think, “Who’s reporting this?
1:06:29Is this The Onion?” No, it’s an Irish newspaper, a genuinely serious newspaper.
1:06:33I’m looking at it online, and it has lots of other stories, an archive,
1:06:36a whole load of things. Of course, that makes me believe the story.
1:06:40So what we’ve done is democratize and magnify the manipulation of truth.
1:06:44And we’re going to have to see the consequences of that,
1:06:46because it’s only luck that we haven’t seen more, given how easy this is to do.
1:06:50Right,
1:06:51and are people now looking at strategies for how we can protect ourselves from
1:06:55that? I mean, what can we actually do to identify these different kinds of scams?
1:07:00Work is being done on it. Work is being done on it.
1:07:02There are some proposals, for example,
1:07:04to mark content generated with artificial intelligence with virtual
1:07:08watermarks that can’t be removed,
1:07:09so that your devices, like this tablet, can tell you, “Hey, this is AI.
1:07:13This was made by AI.” So, those are watermarks.
1:07:15You know the little star that often appears down in the bottom
1:07:19right-hand corner of Gemini photos?
1:07:20Well, something like that, but not visible—invisible.
1:07:23In Google’s case, it’s called SynthID, and for now,
1:07:26they’re arguing over what the standard should be.
1:07:28But, of course, look: if that watermark is in the photo,
1:07:31and that computer can read the watermark, then that computer can warn me about it.
1:07:35But if I didn’t make the photo with Gemini,
1:07:37if I made it with Meta and Meta doesn’t put a watermark on it,
1:07:40my computer won’t know that it has one.
1:07:42And they’re even seeing that artificial intelligence can be
1:07:45used to remove the watermark. So, in the end,
1:07:48I think that unless we all agree and come up with some standards—like an ISO
1:07:52standard or something along those lines—I think this is going to mitigate the
1:07:55problem, but it’s not going to eliminate it at all.
1:07:58Because there’s also a very big problem with open source.
1:08:01Open source is open-source AI.
1:08:02Basically, there are two types of artificial intelligence,
1:08:05just like there are with all software, okay?
1:08:08There’s proprietary software from certain companies, with closed code,
1:08:11where you can’t see how it was made or anything like that,
1:08:14and then there’s open-source software, whose code is public.
1:08:17So, there are many companies, like Meta, for example, and DeepSeek, for example,
1:08:21that are developing open-source code.
1:08:23That means anyone can take all the work they’ve done and create another,
1:08:27more evolved version from it, okay?
1:08:29So I can create an evolved version from it and remove the watermark, you know?
1:08:33So, of course,
1:08:34that means I’m responsible for putting it there because the law requires me to,
1:08:38okay, but I’ve decided not to put it there.
1:08:40And so, in the end, we’re going to mitigate the problem,
1:08:43but we’re never going to eliminate it.
1:08:44That's why we always come back to the point we started this episode with.
1:08:49It's better to stay on top of all this, to know about it.
1:08:52It's necessary; it's no longer optional.
1:08:54Hay una parte informativa y otra de formación. Both are important.
1:08:59La parte informativa es lo mínimo necesario para evitar
1:09:02problemas graves por todo esto.
1:09:04La formación será esencial para todo lo relacionado con el mercado laboral.
1:09:08Pronto, quien no sepa usar inteligencia artificial y agentes a nivel
1:09:13avanzado será inempleable.
1:09:14Igual que pasó con la ofimática en su día. Today…
1:09:17At the time, I don’t know if people will remember this,
1:09:21and I don’t know how old your average listener is,
1:09:24but I remember when office software first arrived.
1:09:27Poníamos en el currículum que sabíamos ofimática.
1:09:30Era un valor añadido, no todos sabían usarla y te hacía destacar.
1:09:34Era una ventaja competitiva.
1:09:36Hoy, en el currículum de un joven de veinte años, nadie incluye la ofimática.
1:09:40Todos saben usar Excel, PowerPoint y Word, ¿no? I mean, it’s…
1:09:44¿A dónde vas si no sabes usar esas herramientas?
1:09:47Ya no es una ventaja competitiva y dejamos de mencionarlo.
1:09:51Creo que pasará lo mismo con la inteligencia artificial.
1:09:54Por un tiempo será un valor añadido y ahora tiene mucha demanda.
1:09:58Sure.
1:09:59Llegará un punto en que todos lo sepan; si no sabes usar IA, no podrás trabajar.
1:10:04Just like today with automation.
1:10:06El trabajo debería ser muy físico para estar en una oficina
1:10:09sin saber usar un ordenador.
1:10:11Es imposible trabajar en marketing, finanzas u oficina si no sabes usar
1:10:15Windows, Excel o PowerPoint.
1:10:17Por eso creo que pasará lo mismo con la inteligencia artificial.
1:10:21There’ll be a point when people trained in artificial intelligence
1:10:25have a competitive advantage,
1:10:26and then there’ll come a point when it won’t matter,
1:10:30because everybody will know how to do it.
1:10:32Porque si no, ni siquiera podrás salir de casa.
1:10:36We were just talking about the different areas where we can use
1:10:40artificial intelligence. I want to move on,
1:10:43because I know you’re also surrounded by a lot of business owners and entrepreneurs.
1:10:48What opportunities are there today for people who have their own business?
1:10:52I once heard you say that nowadays,
1:10:54there’s no longer any comparison between the little neighborhood
1:10:58shop and a major company, because they’re both on the same level,
1:11:02with the same possibilities of using that technology.
1:11:05So, what can someone with a business do today to make the most of it,
1:11:09and even replace entire work teams?
1:11:11So I don’t need to have ten people working around me.
1:11:14There’s no need to replace entire work teams.
1:11:16And that brings us to the employment side of things, which I think is interesting.
1:11:20But before that, I don’t think people are aware of what we were saying:
1:11:24that you can compete at Samsung’s level.
1:11:26Because until now, people have always believed that,
1:11:29since everything depended so much on money, those with more money—in this case,
1:11:33the big companies—could afford better marketing teams, better product teams,
1:11:37better teams for whatever it was, right, to launch things.
1:11:40So people need to understand that today, we have access to the same technology
1:11:44that, for example, the American government’s Department of War has.
1:11:47I mean, a few months ago they introduced a proposed law in the United States that
1:11:51would voluntarily allow artificial intelligence providers to make their
1:11:55models available to the government before releasing them to the public.
1:11:59And if they found very advanced cybersecurity capabilities,
1:12:02they would ask them—again,
1:12:03voluntarily—to give them a thirty-day head start before the
1:12:06model was released publicly. What does that mean?
1:12:09It means the model is being released to the public at the same time it
1:12:12reaches the Trump administration. In other words,
1:12:15you being able to use the same technology the United States government is using for
1:12:19everything—for the CIA,
1:12:20the FBI—that has never happened in the ****ing history of the world.
1:12:24We’re at a level where we can compete at the very highest level.
1:12:27And to help people understand,
1:12:29while these people are doing such incredible things with it,
1:12:32obviously applying it to security, technology, education, and so on,
1:12:35most of us are making little kittens for TikTok with the technology.
1:12:39I mean, we’re completely wasting this absolutely incredible opportunity, you
1:12:43know? And people have to realize that the opportunity is right there.
1:12:46And it’s true that the opportunity involves paid AI.
1:12:49It doesn’t involve, obviously, free AI.
1:12:51You can run your little tests, right?
1:12:53But that’s where the game is.
1:12:55Two badly made mixed drinks at a nightclub in Barcelona.
1:12:57That means it’s twenty euros a month, and you’ve got the paid version of AI.
1:13:01So with that, you can do things that, today, Samsung’s marketing team is doing.
1:13:05I mean, you can actually be on the same level as a multinational with a
1:13:09budget of many millions of euros, with your twenty-euro AI.
1:13:12And that’s a competitive advantage we’ve never had before.
1:13:14And when we start talking about entrepreneurship,
1:13:17I think we’re in the position that any emerging market creates,
1:13:20which is that there’s an absolutely incredible blue ocean of opportunities.
1:13:24And the way I usually explain this is, just to give you an idea,
1:13:27the Twitter of artificial intelligence, the Gmail, the Facebook, the Instagram,
1:13:31the TikTok, the Amazon of artificial intelligence, haven’t even been created
1:13:35yet.
1:13:36I mean, the companies we’re seeing, like OpenAI and Anthropic, are what Cisco was.
1:13:40They’re the ones laying the Internet cables.
1:13:43I mean, the Internet gave rise to all these companies I’m talking about,
1:13:46most of which today are among the ten most valuable companies in the world.
1:13:50That hasn’t happened yet with AI. For the moment,
1:13:53what most people are doing with AI is the same thing they used to do, but with AI.
1:13:57The real opportunity is everything we’re going to be able to do from now on
1:14:00because we have artificial intelligence.
1:14:03In other words, building on top of AI,
1:14:04rather than applying AI to what we’ve already been building.
1:14:08That’s thinking too small, you know?
1:14:09But I really think that’s where the opportunity is.
1:14:12And what I was saying about workers: I think it’s very easy to think,
1:14:15especially from the mind-set of a freelancer, or the owner of a small
1:14:19business, to say, “I’ll replace my workers with AIs and save money.” Honestly,
1:14:23an entrepreneur’s mind-set should never be about saving costs.
1:14:26It should be about doubling your revenue.
1:14:28And I think that’s the key point.
1:14:30And I’m going to tell you about the case of IKEA.
1:14:32IKEA, even though it’s a huge multinational,
1:14:35is a perfect example for those of us who are small-business owners, entrepreneurs,
1:14:39freelancers, or whatever. IKEA developed a customer-service chatbot, okay?
1:14:43Using artificial intelligence.
1:14:44And they launched it on the market.
1:14:46It’s called Billy, I think, or something like that.
1:14:48I don’t remember the exact name, but it’s like a shelving unit.
1:14:52Mickey or Billy, one of those.
1:14:53Anyway, they put the chatbot on the market and suddenly, in the first month,
1:14:57it successfully handled three million customer inquiries.
1:15:00Fifty-seven percent of all the inquiries IKEA received worldwide.
1:15:03Fifty-seven percent. It’s crazy.
1:15:05So, of course, suddenly IKEA’s management team finds itself there and says, “Okay,
1:15:09fifty-seven percent of customer service can go to hell. I mean, it’s solved.
1:15:13Right? So what do we do now?” And, of course,
1:15:15what many companies we’re seeing in the market do is: out the door.
1:15:19And that’s happening at some companies.
1:15:21But IKEA decided to do something else.
1:15:23They said, “Hey, before we make this team of people who’ve been with us for years,
1:15:27who work well, who are good people,
1:15:28who simply do a job that right now doesn’t have any value, before we let them go,
1:15:33let’s see what’s going on with the other forty-three percent of our customers that
1:15:37our guide can’t answer.” And they analyze those inquiries and see that most of the
1:15:41questions it can’t answer are things like,
1:15:43“How should I decorate my home this fall?” Questions that aren’t about products,
1:15:47that aren’t about this or that, and that the robot couldn’t answer.
1:15:51So then they decide to create a retail-design division:
1:15:53people who help customers decide how to decorate using IKEA products.
1:15:57They take that fifty-seven percent of customer-service staff and,
1:16:00with some quick training, turn them into people who can handle that kind of
1:16:04inquiry.
1:16:05Amazing
1:16:06In one year, that division brought in one point two billion. Billion, you know?
1:16:10I mean, one thousand two hundred million dollars,
1:16:13from that new division they created and used out of what was
1:16:16no longer useful to the company.
1:16:18That’s the mindset I think we entrepreneurs need to have.
1:16:20Con el mismo equipo, factura el doble.
1:16:22Ganar cuota con el mismo equipo.
1:16:24Eliminar a la competencia con el mismo equipo.
1:16:27Eso no evita problemas sociales con el empleo.
1:16:29Pero nuestra empresa no los tendrá.
1:16:31I mean, I think we have to understand that, to a certain extent,
1:16:34this is becoming a kind of Hunger Games,
1:16:36where there are going to be winning companies and losing companies.
1:16:40Habrá pérdida de empleos, no me cabe duda.
1:16:42Veremos cuánto tarda, pues las empresas suelen ir despacio.
1:16:45No dudo que habrá pérdida de empleos, aunque no en todas partes.
1:16:49In other words, what we’re going to see is the number of players shrinking,
1:16:53because there will be much more competitive players,
1:16:56and players who are asleep at the wheel, saying,
1:16:58“But how can they offer those prices? No me salen las cuentas.
1:17:01They’re losing money.” No, they’re using artificial intelligence,
1:17:05increasing their margins, increasing their capacity, and reducing production times.
1:17:09And meanwhile, you’re still stuck working with a calculator and a little notebook
1:17:14while the other guy is using Excel. No puedes competir.
1:17:17Lo verás, pero será demasiado tarde.
1:17:18Ese es el problema con el empleo.
1:17:20Are we going to have fewer people working in the world?
1:17:23Creo que habrá menos gente trabajando.
1:17:25Pero no afectará a todas las empresas.
1:17:27Más bien, el número de actores caerá.
1:17:29El número de podcasts se reducirá.
1:17:31Because what you’ll be able to do with your podcast,
1:17:34thanks to artificial intelligence, other people won’t be able to compete with.
1:17:38And what’s going to happen?
1:17:39El cliente te elegirá porque aportas más valor.
1:17:42Ofreces documentación y material extra.
1:17:44Los documentos mencionados aparecerán listados,
1:17:46ya que la IA habrá analizado el podcast y los habrá extraído
1:17:50automáticamente para ti. Les darás un código QR para que hagan esto o aquello.
1:17:54O, por ejemplo, un experimento en nuestro podcast: hacerlo interactivo.
1:17:58La audiencia podrá pausar el podcast,
1:18:00preguntar algo y nosotros responderemos basándonos en lo
1:18:03que el podcast nos proporciona.
1:18:04Es un caso de uso totalmente distinto a simplemente hacer lo
1:18:07mismo con inteligencia artificial.
1:18:09Esas son las posibilidades que abre la IA.
1:18:11Sea cual sea tu trabajo, si eres emprendedor, es el mejor momento:
1:18:15nunca ha sido tan fácil empezar un negocio.
1:18:17You have all the tools, first of all.
1:18:19Y segundo, surge un océano azul de oportunidades.
1:18:22Si trabajas para una empresa,
1:18:23es hora de ver cómo aportar valor más allá de lo que has contribuido hasta ahora.
1:18:28Como decíamos antes, ¿verdad? ¿Y cómo va a suceder eso?
1:18:31If you go to your company and say, “Hey,
1:18:33I just thought of something—we could make this podcast interactive,” if one of your
1:18:37employees comes to you with that,
1:18:39you give them a standing ovation.
1:18:41Totally. A hundred percent
1:18:42Which means you value that employee more, raise their salary, promote them.
1:18:46That employee is more secure, you don’t want them to leave,
1:18:49you want them to be part of your team.
1:18:51In the end, I think what we reward in society are ideas. And right now,
1:18:54what we’re doing is allowing people to spend less time on more routine, basic
1:18:58jobs, so they can devote that time to having ideas.
1:19:00One thing humans are still, at least for now,
1:19:03much better at than AI is having ideas,
1:19:05going beyond the ordinary—not analyzing data and so on,
1:19:07but going out and looking for something beyond.
1:19:10I don’t rule out AI doing that better than us in the future.
1:19:12In fact, I have my doubts, because what I see, ultimately,
1:19:15is a machine that produces intelligence.
1:19:17If my brain can produce these kinds of ideas,
1:19:19why wouldn’t another machine produce them too,
1:19:22if we give it enough time and evolution? But in any case, today,
1:19:25the value lies in being able to put those ideas into practice in a very simple way.
1:19:29To lead that AI change anywhere.
1:19:31Look, does Duolingo sound familiar?
1:19:33Yes, yes, that language app, right?
1:19:35Yeah, it does a lot more than languages, but there’s one really interesting
1:19:39example. Basically, so people understand,
1:19:41the usual process when a company like this wants to launch a new
1:19:44product is that someone has an idea.
1:19:46They present it to a committee, the management team, and say, “Hey,
1:19:49I’ve got this idea.” Usually, it’s the typical PowerPoint presentation, right?
1:19:53You put a mock-up up there: “I want to build an app.” In this case,
1:19:57I’m talking about Duolingo launching a chess app, okay?
1:20:00So the usual process would be: I go to management and,
1:20:02based on the typical whiteboard or PowerPoint, I say, “Look,
1:20:05I’ve had this idea to do this, this,
1:20:07and this,” and then management has to decide whether to back it.
1:20:11And they have to ask, “Okay,
1:20:12how much is this whole thing going to cost us?” So we need a development team,
1:20:16a team of PMs, a team for this, a team for that,
1:20:19so we can build an MVP—a minimum viable product that we can test.
1:20:22And then, if we think it’s really interesting,
1:20:24we put in more money to take it into production and turn it into an app.
1:20:28Well, what happened with the Duolingo game—which people can already download;
1:20:32it’s available—is that a PM had an idea, and when he went to management,
1:20:36instead of showing them a PowerPoint, he said, “No, no,
1:20:38download this app.” He already had the app built and working.
1:20:42They started playing chess and said, “Holy crap, you’ve already done it for me.”
1:20:46I mean, nowadays you can create apps with this. It's just crazy.
1:20:49A non-technical person, who didn’t know how to code and didn’t do anything else,
1:20:53was able to create a relatively complex chess game that the leadership team could
1:20:57try out directly,
1:20:58so they could make a more informed decision about whether it was something
1:21:02they wanted to invest in or not.
1:21:04From there, Duolingo backed it,
1:21:05took it into production with programmers to scale it to the level Duolingo needs,
1:21:09with millions of customers, and they released it on the market.
1:21:13Faster, cheaper, and, above all, with the ability to make a more informed decision.
1:21:17What do you think they value in that PM?
1:21:19I mean, the person who had the idea, or the person who made their lives easier?
1:21:23Well, that’s the point for us as workers. If we learn to use AI,
1:21:26and we learn to help the people making the decisions—the people we report
1:21:30to—make better decisions more easily, they’ll value us more.
1:21:33And from that point on, everyone has an interest in using artificial intelligence,
1:21:37because if you don’t use it, someone will come along behind you, use it,
1:21:40and expose you.
1:21:41What happened to the other PMs who were still working with a PowerPoint
1:21:45while this person showed up?
1:21:46If I were a business owner, I’d be asking, “Why don’t you do this?
1:21:50Why don’t you use this? Why are you still bringing me a PowerPoint?”
1:21:53That's what I was telling you before.
1:21:55I mean, with this mindset, I see AI as something I need to take the lead on,
1:21:59you know? I mean, take the absolute initiative here.
1:22:02Pero es porque tienes mentalidad emprendedora.
1:22:05Y creo que la mayoría no la tiene. Y eso no es malo, ¿vale?
1:22:09Simplemente, algunos somos de una forma y otros de otra.
1:22:12But people need to have that mindset.
1:22:14Necesitamos más emprendimiento que nunca por el mundo al que vamos.
1:22:18Let’s talk about another use, one I think is fantastic for everyone who follows us,
1:22:22especially those who may not be that interested in entrepreneurship or
1:22:25business, but are interested in learning.
1:22:28They’re at a stage in life where they want to learn about everything,
1:22:31and here on the podcast we cover all kinds of topics.
1:22:34Every episode: today we’re talking about artificial intelligence, and tomorrow,
1:22:38I promise, we’ll be talking to a medium or about numerology...
1:22:42-Or whatever? -Various topics.
1:22:43What people want, right?
1:22:44I mean, that’s the value I think the audience… no,
1:22:47I don’t think—I’m sure—that the truly great ones appreciate so
1:22:51much about the podcast.
1:22:52So people are hungry to learn new subjects and new topics.
1:22:56How can I use artificial intelligence as my own personal teacher or mentor,
1:23:00so I can learn new things for myself? Where do I even start?
1:23:03Well, look, that’s exactly what you have to put to artificial intelligence.
1:23:07Remember we talked before about how I don’t have to give it instructions,
1:23:11I have to give it objectives?
1:23:13Your objective is: “I want to learn this. I need to train.
1:23:15I have three months, I have one month, I have two hours”—whatever you want.
1:23:19You give it your objective and say, “How can I learn this?
1:23:22How can I do it?” And it’ll give you a guide, a step-by-step plan,
1:23:26which obviously you’re going to have your own opinion about—about
1:23:29what you think is right or wrong,
1:23:31or if you want to learn it better...
1:23:32Let’s say you get into marketing and you like Hormozi,
1:23:35and you want to learn from Hormozi, not from this person or that person, or
1:23:39whoever. So you can drive the train in whatever direction you want,
1:23:42but you’ll see that it’s running on rails. It’ll keep guiding you.
1:23:46So, precisely, that’s the feeling I think is missing right now.
1:23:49We’ve spent three, four years with artificial intelligence,
1:23:52having to know how to ask AI the right questions.
1:23:54And we’re at a point now where what we need to know is what we want.
1:23:58Once you know what you want,
1:23:59communicate it clearly—and don’t make it just one sentence...
1:24:02And that’s always going to be more difficult.
1:24:05I think that’s something even AI can’t save us from.
1:24:08But don’t give the AI thirteen words—give it every possible detail.
1:24:11What’s your motivation for learning? Why do you want to learn?
1:24:14Because there are things you think are irrelevant that actually won’t be.
1:24:18You might say, “Look,
1:24:19I want to learn biology so I can do experiments with my kids.” That’s
1:24:23completely different from saying you want to learn biology because you’re interested
1:24:27in dolphins. Totally different.
1:24:29So give your guide as much information as possible,
1:24:31everything you can think of about the subject.
1:24:34Pour it all into voice messages.
1:24:35You don’t have to write it down,
1:24:37you don’t have to sit there and make sixteen pages in a Word document.
1:24:40Just spill it out, vomit all the context onto it, give it everything, and say,
1:24:44“This is my goal. This is what I need.
1:24:46Take care of solving my problem.” And from there, the AI will say,
1:24:50“Let’s get to it.” It’ll come back and say, “Look, let’s start here, then here,
1:24:54then here.” It might even say things like, “Hey,
1:24:56what do you think about me making you an app where I give you tests
1:25:00as you go?” In other words, it’ll give you ideas.
1:25:02Obviously, we’re not at the end of the game,
1:25:05where it gets everything perfect every time. Let’s be realistic.
1:25:08There will be things you come up with that are better. Tell it those things too.
1:25:12You can always iterate.
1:25:13One thing that’s changed a lot since we got Work—and some models could do it a
1:25:17little before that—is that when you send a message to the AI and it starts working,
1:25:22now you can send follow-up messages.
1:25:23You can change its direction, or add things that happened afterward.
1:25:27Before, you had to wait for it to answer and then start over. Not anymore.
1:25:31Now, while the AI is thinking, working, and doing things, you can say, “Oh,
1:25:35by the way, I forgot something,” and add one more thing.
1:25:37And it’ll incorporate that into what it already had.
1:25:40It won’t start from scratch with a different approach;
1:25:43it’ll simply take that into account as something added to everything you’d
1:25:47already given it. So it allows you to have a conversation with it, well, as I said,
1:25:51to treat it like a human. Now, imagine you have a person—and I’m sure you do,
1:25:55and the audience does too—someone who really is a mentor.
1:25:58Someone you truly believe is someone who is...
1:26:00Someone smarter
1:26:01Someone who’s been there with you many times and given you good advice.
1:26:05How would you explain the situation—that you want to learn biology,
1:26:09or whatever it may be? You’d say, “Hey, I’m in this situation.
1:26:12What would you do if you were me?
1:26:14How would you go about doing it?” That’s what you need to tell ChatGPT.
1:26:17Very good, very good.
1:26:18Hacer eso no significa que lo que sugiera sea lo mejor para nosotros.
1:26:23Repito: necesitamos pensamiento crítico, iterar y señalarle sus errores.
1:26:27A veces debemos ponernos frente a la IA al proponer planes.
1:26:31No aceptemos todo lo que nos da como bueno, válido y perfecto.
1:26:34Detengámonos a revisar lo que nos ofrece.
1:26:37Necesitamos colaborar, es un colega.
1:26:39It’s like what you’d do with a friend or colleague, right? You question things.
1:26:44Imagine you tell your friend,
1:26:46“Can you make me—since you’ve got a friend who’s a nutritionist—a nutrition plan,
1:26:50blah blah?” And he says, “Well, I think this.” And you say, “Honestly,
1:26:54I don’t see myself going sixteen hours without eating.” So you’re going to adapt
1:26:59it, right?
1:27:00Eso debes hacer con la IA.
1:27:02No creas todo lo que genera.
1:27:03Trabaja con la IA, no dejes que lo haga sola asumiendo que es válido.
1:27:07Debes supervisar, liderar y seguir participando. Esa es la estrategia:
1:27:11trata a la IA como a un humano y discute los temas desde ese punto de vista.
1:27:16Eso producirá resultados mucho mejores.
1:27:18What topics have you seen AI produce more of what you called hallucinations?
1:27:23Are there areas where we should be more careful or cautious when iterating?
1:27:27I think that, with AI,
1:27:28the whole issue of safety is related to how important the task is.
1:27:31I mean, I don’t trust AI one hundred percent.
1:27:34But if I’m asking, “Hey, what restaurant can I go to for a good paella?” Well,
1:27:37honestly, if it gets it wrong, it’s not the end of the world.
1:27:41I mean, it’s not dramatic.
1:27:42Anything that’s really important to me, I’m going to cross-check with other
1:27:46sources.
1:27:47So if my child’s life is at stake, I’m not going to trust artificial intelligence.
1:27:51If I’m making a decision that’s going to affect the sale of my company,
1:27:55I’m not going to trust artificial intelligence.
1:27:57If I’m signing a rental agreement for five years,
1:27:59I’m not necessarily going to sign it based only on what artificial intelligence tells
1:28:04me. But the levels of risk I’ve just laid out are different.
1:28:07My child’s life is above everything else, obviously...
1:28:10With the rental agreement, maybe I’ll trust what the AI tells me,
1:28:13but I might also take it and put it into another AI to ask what it thinks, you
1:28:17know? You can play around a little with getting multiple opinions,
1:28:20because we have access to so many AIs these days.
1:28:23If we move up a level and it’s something a bit more important,
1:28:26maybe I’ll hire a lawyer to check whether what the AI said
1:28:29actually makes sense or not.
1:28:30And just look at the change there: instead of hiring a lawyer to do the work,
1:28:34you hire a lawyer to supervise what you’ve done with AI.
1:28:37The number of hours is reduced dramatically...
1:28:39I mean, it’s being reduced.
1:28:41Look, I have clients who are some of the most famous law firms in Spain,
1:28:44and they’re considering changing from charging by the hour to charging by
1:28:48project,
1:28:49because their productivity has increased so much that charging by the hour is
1:28:53becoming barely profitable.
1:28:54And on top of that, the projects take fewer hours too,
1:28:57because sometimes they’re mainly supervision projects.
1:29:00So charging by project is a new business model that AI is creating in law firms.
1:29:04Just imagine the level of ripple effects that everything happening
1:29:07right now is going to have.
1:29:09But as we move up through the different levels,
1:29:11we’ll make those decisions accordingly.
1:29:13So I’d say that anything related to medicine should always be supervised by a
1:29:17doctor. More than anything, AI should be used as a second opinion.
1:29:20With anything related to the law,
1:29:22I’d say that unless it’s something that doesn’t screw you over too badly,
1:29:26have a lawyer supervise it. There,
1:29:27I’d say you can do a lot of the work with AI and then have a lawyer
1:29:31review it at the final stage,
1:29:32just to make sure everything’s okay. But in that area,
1:29:35I wouldn’t suggest having the lawyer do the work and then having AI review it.
1:29:39For really important matters, yes, absolutely. But for ordinary things,
1:29:43I think we may already be at the point where AI does the work and
1:29:46the lawyer reviews it. That’s different from doing it the other way around.
1:29:50In medicine, I’d do it the other way around.
1:29:52I’d have the doctor do it, and then have AI review it.
1:29:55And in education, for example, I think that one’s a no-brainer.
1:29:58I’d go all-in with children.
1:29:59That doesn’t mean giving a child AI and letting them do whatever they want with
1:30:03it. It has to be supervised. We have to be careful,
1:30:06because obviously there’s a point where children very easily start completely
1:30:10outsourcing their cognitive tasks to AI and stop doing anything themselves.
1:30:14In fact, a very well-known study has just come out of China.
1:30:17It’s become quite large—the biggest study we have on AI in education—and its
1:30:21conclusion is that children who work with AI score much better on exercises,
1:30:25but worse on exams.
1:30:26That means they’re doing the exercises with AI, but when it comes down to it,
1:30:30they haven’t learned a damn thing. What happens, though,
1:30:32is that there’s a subset of kids who use AI and get the highest scores of
1:30:36everyone in the entire study.
1:30:38Los chicos usan bien la IA: para aprender, no para trabajar.
1:30:41It's just like being your guide, not always seeking answers.
1:30:44Well,
1:30:45it’s about finding the way AI can add value so you can learn faster and better.
1:30:49In the end, learning is a process.
1:30:51If we can optimize that process, we’re going to learn faster and better.
1:30:55But it’s true that it doesn’t make all that much sense.
1:30:57And think about it: in the end, we’re measuring an exam,
1:31:00and it’s an old-fashioned kind of education exam.
1:31:03So normally, what we’re assessing is knowledge. You ask them,
1:31:06“What are the rivers of Spain?” What sense does it make nowadays to
1:31:09know about the rivers of Spain?
1:31:11Or the Visigothic kings, you know?
1:31:13So in the end, it’s a problem with education.
1:31:15It doesn’t come only from AI, but AI makes it more obvious.
1:31:18So the fact that students do poorly on those exams doesn’t necessarily
1:31:21mean they aren’t learning.
1:31:23It may simply mean they aren’t learning the things we’re testing them on—and those
1:31:27things may not be the most optimal.
1:31:29I think what you're saying makes it even clearer how our education today compares
1:31:33with what the world really needs.
1:31:35That's an incredible can of worms.
1:31:37With my kids, I involve them in AI.
1:31:39They’re nine and eleven, so I want them to use it, and I want them to be part of it.
1:31:44My daughter, for example, uses it more than my eleven-year-old son.
1:31:47My daughter is nine years old.
1:31:49And I think it’s very important that they’re aware that this is
1:31:52part of their everyday life. Now,
1:31:54we also have to be very careful about how much access we give
1:31:57them and what they do with it,
1:31:58so it always has to be supervised.
1:32:00What kinds of tasks do you generally give them more freedom to use AI for?
1:32:04For me, what we’ve found is something called Vibe Coding,
1:32:07which is the ability to program applications without knowing how to
1:32:11program, okay?
1:32:12And this is something that, by the way, the latest Work version of ChatGPT
1:32:16includes. It’s called Sites, and you can ask it to make you an application,
1:32:20and it deploys it as a public website so you can share it.
1:32:23So, for me, that’s the best feature for kids.
1:32:25Kids don’t need to chat with AI, they don’t need to talk,
1:32:28because that can lead to something very important: thinking it’s human.
1:32:32We'll talk, don't worry.
1:32:33That’s a serious problem.
1:32:35So, for me, Vibe Coding makes possible what you could call digital Lego.
1:32:38First step: half an hour building something, which means creating that
1:32:42application,
1:32:43communicating with the AI in an agent-like way: change the background colors,
1:32:47change this, put a dinosaur instead of a snake, that kind of thing.
1:32:50And from there, it creates an application for them,
1:32:53which is usually a game they then play for hours.
1:32:55So it’s exactly like Lego.
1:32:57You spend half an hour building, and then you play with the car for hours.
1:33:01For me, digital Lego is Vibe Coding,
1:33:02and I think it’s one of the best things we can have children do,
1:33:06because it’s also more contained.
1:33:07It doesn’t really allow things to drift into emotional topics, and so on.
1:33:11Instead, they’re building something, and developing a sense of how to manage
1:33:15agents. The leadership we were talking about earlier,
1:33:18the kind that’s going to be needed at work—well, of course,
1:33:21my children are already at that stage...
1:33:23My children, and many others.
1:33:24I’m not the only one doing this.
1:33:26But they’re leading the AI toward what they want.
1:33:28They have the idea, the AI makes it possible, but they’re the ones leading.
1:33:32They say, “No, change this. No, make it faster.
1:33:35No, do it like this,” and they keep telling the AI what they want it to do.
1:33:38And, in fact, what surprises me so much is that my children are making a video game,
1:33:43something like Tetris, or Snake—things that, twenty years ago,
1:33:46someone spent months working on.
1:33:48A whole team of people would spend months working on it to get it onto the market.
1:33:52And now my daughter sends the AI a voice message saying, “Hey,
1:33:55replace these pieces with apples.” And the AI takes, I don’t know,
1:33:58a minute to do it, and she says, “This is so slow!” Honey, for God’s sake,
1:34:02that’s three hundred lines of code that someone would have had to type out by
1:34:06hand, you know?
1:34:07But, of course, they’re getting used very quickly to things being immediate.
1:34:11And I think that’s where we’re going to see a truly enormous
1:34:14leap—not just in capabilities, but in possibilities.
1:34:17Because new chips are coming in.
1:34:18One of the most famous brands, for example, is Cerebras,
1:34:21and I was with the CEO in Silicon Valley not long ago.
1:34:24This man has just closed a twenty-billion-dollar contract with
1:34:27OpenAI. Twenty billion dollars.
1:34:29And these people make larger chips that allow the speed to be fifteen
1:34:32times faster than what we have today.
1:34:34So you’re going to tell ChatGPT,
1:34:36“Make me a Snake game,” and you’ll have it in a second.
1:34:39It won’t take the time it currently takes to program it.
1:34:42So this is going to open up things that—just to help people understand—when
1:34:45the internet appeared, and fiber optics increased internet speed,
1:34:49what we all thought was that we’d be able to load web pages faster.
1:34:52But what it actually made possible was Netflix.
1:34:55Netflix couldn’t have existed with slow internet.
1:34:57So the fact that we’re going to have faster AI is going to open up
1:35:00completely different use cases.
1:35:02For example, today you have software, an operating system,
1:35:05on this iPad or on your computer, Windows or whatever.
1:35:08And when you open a folder,
1:35:09it’s programmed to open a folder and do a particular thing.
1:35:12The operating system of the future is going to be created in real time.
1:35:16So you’re going to say, “I’m going to open a folder,” and at that moment,
1:35:20the operating system will create that folder’s workspace based on what you need.
1:35:24And that’s going to be possible because of AI operating at an
1:35:27absolutely incredible speed,
1:35:28with unbelievable programming capabilities.
1:35:30So that combination of speed and all of that could open the door to the
1:35:34Amazons and whatever else I was talking about earlier—things that,
1:35:37today, we aren’t even capable of imagining,
1:35:40and that are going to become companies worth billions very soon,
1:35:43thanks to everything this will make possible.
1:35:45And I think one of the most important factors is going to be, on the one hand,
1:35:49robotics, and on the other hand, the speed of artificial intelligence.
1:35:53This is crazy!
1:35:53It seems to me that this is moving at an, as we say, mind-blowing speed,
1:35:57and we’re not grasping it—and above all, how is it going to evolve?
1:36:01And for me, that’s something that makes me question, right?
1:36:05Where are our own cognitive abilities headed?
1:36:07Because right now, the speed at which we get things,
1:36:10and our ability to tolerate frustration when we do things day to day,
1:36:14are different from the speed this is showing us.
1:36:17I mean, it’s something else.
1:36:19So, I’m curious to know how our minds are going to evolve.
1:36:22There’s a concept here called cognitive delegation.
1:36:24If you’re interested, I recommend bringing Senén Barro onto the podcast.
1:36:28He’s an incredible scientist from Galicia who’s been working in
1:36:31artificial intelligence for decades.
1:36:33And Senén talks about this idea of cognitive delegation.
1:36:36So the cognitive question is,
1:36:37“Am I going to become stupid by using artificial intelligence?” Basically,
1:36:41that’s the question we’re all asking ourselves.
1:36:44I don’t think so, but that’s my opinion. CNN has its own.
1:36:47My opinion is that I haven’t become any dumber because I no
1:36:50longer know how to use a map, because I use GPS.
1:36:52I mean, look, when I was little, my parents traveled for work.
1:36:56My father sold artistic ironwork, and in the summer, when I finished school,
1:36:59we’d go all over Spain in an R15 with a van, selling ironwork.
1:37:03I went along with him. And I was the map guy.
1:37:05My father would drive, and I’d tell him, “Look, you have to take the N-two.
1:37:09Then you have to take this road,
1:37:11and turn off here.” I was reading maps at eight years old,
1:37:14and my greatest talent was being able to read maps and add value for my father.
1:37:18Today, you give me a map and I don’t even know how to hold it.
1:37:21I don’t even know how to hold it.
1:37:23I’ve completely lost that skill, because GPS has replaced the need to have it.
1:37:27So you could argue that I’ve become dumber because I don’t know how to use maps,
1:37:31because I use GPS. But for me, the feeling is different.
1:37:34I feel that my cognitive capacity is limited, like everyone’s.
1:37:37And that capacity has freed up a space that used to be taken up by reading maps.
1:37:41That map-reading ability has disappeared completely,
1:37:44because GPS has replaced the need for it.
1:37:46It’s not that my mind has become smaller;
1:37:48it’s that one particular function has been outsourced.
1:37:51And what I’ve been able to do is fill that space with other things,
1:37:54because the space is free.
1:37:56So I think the key is going to be what we fill it with.
1:37:59If all the cognitive delegation we give to AI—the processes and
1:38:02knowledge we no longer need,
1:38:03mathematical calculations,
1:38:05things we won’t need anymore—if we delegate all of that and fill
1:38:08the space with Sálvame, with TikTok, with that kind of thing,
1:38:11then society is going to become completely stupid.
1:38:14But if I fill it by reading Dostoevsky, maybe I’ll even become wiser.
1:38:17Maybe I’ll grow as a person.
1:38:19Maybe I’ll evolve, you know?
1:38:20So I think it’s going to be an individual matter, a matter of personal choice.
1:38:24What’s my view? Most of society is going to become more useless than it is today,
1:38:29because I think most people, unfortunately, tend toward comfort and
1:38:32convenience, toward doing as little as possible, toward mediocrity.
1:38:35Of not taking responsibility.
1:38:37Imagine if, even before the age of AI,
1:38:39we were already delegating responsibility for everything that belonged to us,
1:38:44or blaming someone outside ourselves for all the things happening to us...
1:38:48Just imagine that now.
1:38:49What we're going to do is simply transfer that responsibility.
1:38:53Pero depende de cada uno.
1:38:54Absolutely, yes.
1:38:55It’s absolutely within reach, and it’s an individual matter. Look,
1:38:59I can tell you that whether Spain does well in the new world of artificial
1:39:03intelligence doesn’t depend entirely on me as an individual,
1:39:06or on you, or on any one of us individually.
1:39:08But whether I do well depends on me,
1:39:10because I have the ability to make the most of everything that’s coming,
1:39:14react in time, stay at the top of the pyramid, and then, when the tide rises,
1:39:18not get flooded. You know?
1:39:19So that’s within our reach today, and I think that’s individual responsibility.
1:39:23That means each of us can choose how to respond to what is coming.
1:39:27And in fact, I give the Spanish government a really hard time, for example,
1:39:31because I think they’re being negligent by not creating that feeling that people
1:39:35need to get their act together.
1:39:37In Singapore, in twenty twenty-three,
1:39:38they made artificial intelligence training mandatory for all public-sector workers.
1:39:43And what did that lead to?
1:39:44It spread to private-sector workers,
1:39:46and today Singapore has the highest adoption of artificial
1:39:49intelligence in the world.
1:39:51So I think all these kinds of efforts to warn society about how important
1:39:54artificial intelligence is, what it means,
1:39:57and the change of model it represents at every level—I think that’s a
1:40:00responsibility the government is neglecting.
1:40:02I think the government’s job is to protect its citizens.
1:40:05They should warn us about the wave that’s coming at us,
1:40:08the tsunami that’s coming at us, because people are completely oblivious.
1:40:12Completely oblivious.
1:40:13People are paying more attention to a lot of other things that are important—I
1:40:17understand that they’re important.
1:40:19But we deserve—and look, when I was at the United Nations last year,
1:40:23where I gave a talk about artificial intelligence at an event on humanism,
1:40:27I used this exact same reasoning.
1:40:28I think people deserve a weather forecast.
1:40:30If someone tells you it’s going to rain,
1:40:33then it’s up to you whether you want to bring your clothes inside or not.
1:40:36If you don’t bring them inside and it rains and they get wet,
1:40:40that’s your damn problem. Nobody has to babysit you.
1:40:42But as a society, we deserve a weather forecast, because individually,
1:40:46we can’t have one.
1:40:47And I don’t think we’re being given that weather forecast when it comes
1:40:51to artificial intelligence. It’s being minimized.
1:40:53They’re only talking about the benefits.
1:40:55They’re not talking about the beating the job market is going to take
1:40:59because of what we’ve been discussing.
1:41:01They’re not talking about deepfakes.
1:41:03There isn’t a single television program explaining artificial
1:41:06intelligence to the public.
1:41:07We still have programs on public television devoted to literature.
1:41:11And I’m not saying that isn’t important.
1:41:13But, damn, at the moment we’re in...
1:41:15Then we say we have bigger problems in life too.
1:41:18We need to be doing things around artificial intelligence.
1:41:21Nobody talks about artificial intelligence. Nobody talks about it.
1:41:24I mean,
1:41:25the Pope had to start talking about artificial intelligence before public
1:41:29television even put on a program explaining artificial intelligence.
1:41:32The G7 had to talk about it—the seven most important countries in the Western world.
1:41:36They brought in the CEOs of the artificial intelligence labs and put them in there as
1:41:40if they were heads of state,
1:41:42before society was even remotely aware of what was happening. It makes no sense.
1:41:46The lawmakers are the slowest of all.
1:41:48They should be well behind society.
1:41:50We, as a society, should be deciding what we want to do with AI,
1:41:53and lawmakers should then act.
1:41:54Society isn’t deciding shit.
1:41:56It isn’t deciding absolutely anything, because we’re completely ignoring it.
1:42:00Nobody has told us, “This is going to change your life.” And that’s the problem.
1:42:04And that also conflicts with this very human thing, which is resistance to
1:42:08change. It's really hard for us to be flexible sometimes.
1:42:11That’s normal. It’s natural, and nobody said it was going to be easy.
1:42:15It’s normal for us to resist. But I don’t think... I mean, honestly, here...
1:42:18I’m not usually someone who does what we were talking about before,
1:42:22blaming external factors. We could do that very easily.
1:42:25But I really believe that, as of today,
1:42:27there isn’t more social awareness about artificial intelligence
1:42:30because no real effort has been made to raise awareness in society.
1:42:33In fact, quite the opposite.
1:42:34I think its impact has even been downplayed.
1:42:37I think people, individually,
1:42:38already have so much going on in their daily lives that they can’t really keep
1:42:42track of what’s happening in the world of artificial intelligence.
1:42:45On my YouTube channel,
1:42:47where we make a weekly news video about everything happening in AI,
1:42:50plus podcasts and all that, we have twenty people working full-time.
1:42:53Twenty people on a YouTube channel, on top of my own work,
1:42:56which is seventy hours a week.
1:42:58And we still can’t keep up.
1:42:59We can’t keep up with everything happening in artificial intelligence.
1:43:03I mean, how is someone watching us supposed to keep up?
1:43:06They have two kids, a job, after-school activities...
1:43:08How are they supposed to keep up if I spend seventy hours a week on this,
1:43:12with twenty employees, and I still can’t? Of course they can’t.
1:43:15So all we can do is try to have a basic understanding of the subject.
1:43:19And then, if we have more free time, we can learn more about it.
1:43:22But the thing is, it’s impossible to stay up to date.
1:43:24So we can’t put the responsibility for society taking this seriously on
1:43:28individuals. We can’t say, “Well,
1:43:30you didn’t pay attention.” If you do things like they’ve done
1:43:33in the United States...
1:43:34They created a basic, basic, basic artificial intelligence course,
1:43:37and the Department of Education went on national television saying,
1:43:41“Anyone who wants it can access it for free.
1:43:43Just text this number.” Look, it’s not a miracle cure.
1:43:46It’s not the best course in the world, and it’s only three hours.
1:43:49But damn, it’s something. It’s at least something.
1:43:51And the government made the effort to put it out there.
1:43:54On American television, they talk about artificial intelligence every day.
1:43:58Donald Trump talks about artificial intelligence every day. Here, they don’t.
1:44:02They don’t.
1:44:03So I get the feeling that we’re minimizing it, and practically hiding it.
1:44:07And that’s a problem. Of course it’s a problem.
1:44:09If you tell people they’re going to lose their jobs, what happens?
1:44:12People get angry. You know?
1:44:14I mean, this could end up a complete disaster.
1:44:16But is it better to tell them nothing and let them find out tomorrow?
1:44:19Because a lot of people call me saying,
1:44:21“I lost my job because of artificial intelligence.” You know?
1:44:24“What I did suddenly became unnecessary.
1:44:26I didn’t see it coming, and then the client just stopped calling me.” Or,
1:44:30“I called a client—I’m self-employed, I do graphic design for them—and I said, ‘Hey,
1:44:34what about this month’s flyer?
1:44:36Are we doing it?’ And they said, ‘No, no, we’re doing it with ChatGPT.’” Right.
1:44:40If someone had warned me in twenty twenty-two, when ChatGPT came out,
1:44:43maybe I would have had time to adapt.
1:44:45And of course, it also depends on each individual person. So, for me,
1:44:49the conclusion is that we’re clearly in a situation where you can’t count on someone
1:44:53else solving this problem for you.
1:44:55Practically nothing is being done. Practically nothing.
1:44:57So we come back to what shouldn’t be the case, but is: it depends on you.
1:45:01You, individually, can do things to improve your chances in the world,
1:45:05and in what’s coming our way.
1:45:06And the individual thing you can do is, at the very least, stay informed.
1:45:10Stay informed and get trained.
1:45:11Those of you who don’t have children may not see it yet,
1:45:14but you’ll see that your children are already starting to use it.
1:45:17Damn, as a parent, you need to know what they’re doing.
1:45:20You can’t stay on the sidelines. You can’t.
1:45:22It’s difficult, but you can’t. You can’t afford to. So, in the end,
1:45:26you have to at least stay up to date on something that’s going to be the
1:45:29most transformative thing you’ll experience in your lifetime.
1:45:33Okay, Jon, and now I really do want to talk to you about what I think is the most
1:45:37sensitive part of this subject—or one of the most sensitive,
1:45:41because there are so many things here to get into as well—and that’s understanding
1:45:46artificial intelligence as a therapist, right?
1:45:48We’ve seen so many cases online as well, and, well,
1:45:51comments about people who have been talking with AI for a long time and
1:45:56then ended up really badly,
1:45:57or ended up in some situation, okay?
1:45:59How should we define the place, or the role,
1:46:02of artificial intelligence when we’re trying to interact with it,
1:46:06knowing that it’s inevitable for human beings to see even subtle hints of
1:46:10empathy, right?
1:46:11It tells me it understands me,
1:46:13tells me it understands my situation and also understands how I feel,
1:46:17and that no one else...
1:46:18And I also think we’re living in a society with a lot of loneliness right now,
1:46:23and that in this we find a kind of refuge in someone who can understand
1:46:27us better than anyone around us.
1:46:29Look, we have to be extremely clear about this.
1:46:31If you see someone asking an artificial intelligence for emotional support,
1:46:35give them a couple of smacks and take the phone out of their hand.
1:46:38I mean, these tools aren’t ready to do that,
1:46:40and they weren’t sold to us for that purpose either.
1:46:43In other words, if the very same thing I’ve just told you is what the CEO of
1:46:46OpenAI, the CEO of Anthropic, or the CEO of Gemini is going to tell you,
1:46:50these tools are not designed to be emotional support.
1:46:52They weren’t created for that, they weren’t trained for that,
1:46:55and they’re not supervised for that.
1:46:57The problem we have is that it looks like they are.
1:47:00The problem is that our brain,
1:47:01which is very primitive compared to the society we live in,
1:47:04isn’t capable of telling the difference between talking to a human being and
1:47:08talking to an artificial intelligence.
1:47:10I have no doubt that, in the future,
1:47:11AI is going to be the best thing that’s ever happened to humanity
1:47:14for emotional support. It’s going to be incredible,
1:47:17because we’re going to have exactly everything you were talking about.
1:47:20Something we can vent to, something we can brainstorm our feelings with,
1:47:24something we can ask for opinions without being afraid of being judged.
1:47:27A whole bunch of things. But as of today, these applications aren’t ready for that.
1:47:31And when I talk about these applications, I’m obviously talking about Gemini,
1:47:35Claude, ChatGPT, and the ones like them.
1:47:37Certain applications are beginning to appear that are supervised by
1:47:40psychologists and are supposedly designed to do that.
1:47:43And I’m not going to be the one to judge whether it’s good or bad for you
1:47:46to use an artificial intelligence,
1:47:48or whether, in however many years,
1:47:50my children might have an AI friend that helps them.
1:47:52I’m not going to be the one to judge that,
1:47:54because something might seem a certain way from my subjective point of view,
1:47:58while having absolutely nothing to do with the society we’re
1:48:01going to live in in the future.
1:48:02Look, my mother is practically eighty years old,
1:48:05and obviously she spends a lot of time alone at home because she lives by
1:48:08herself. She’s still independent, she has caregivers,
1:48:11but there are times when she’s alone watching TV.
1:48:13So who am I to tell her not to comment on what she’s watching, on that TV show,
1:48:17with an artificial intelligence, in a way that makes her feel accompanied?
1:48:21Of course that’s going to be a good thing.
1:48:23It’s going to reduce her loneliness.
1:48:25There’s actually a study from Harvard Business School that lays out the
1:48:28loneliness problem we have and the possible solutions.
1:48:31And it turns out that, as of today,
1:48:32the thing that reduces loneliness the most is talking to a friend.
1:48:36Big surprise, right? We all knew that, didn’t we?
1:48:38Well, another thing that reduces loneliness among young people,
1:48:41which is very important, is connecting to a stream with their community.
1:48:45If, for example, they’re fans of Ibai, they join Ibai’s chat.
1:48:47It’s not just the interaction with Ibai;
1:48:49it’s with the chat itself and everything that comes with it.
1:48:52Discord, that sort of thing.
1:48:54Our young people are fighting loneliness through these kinds of interactions,
1:48:57which to some of us boomers might seem like, “God, how sad!” No, bullshit.
1:49:01Whatever works for each person.
1:49:03If it works for them, then it works. Fine. Now,
1:49:05what was revealing about the study is that if you talk to an artificial
1:49:08intelligence that pretends to be human,
1:49:10it scores exactly the same as a friend.
1:49:12It reduces loneliness to the same extent as interacting with another person.
1:49:16So who am I to contradict something that, pragmatically, actually works?
1:49:19And if you have a loneliness pandemic where, in Japan,
1:49:22people are throwing themselves out of windows because they feel alone,
1:49:25then maybe it’s not such a bad solution.
1:49:27Now, as of today, the tools we have available aren’t ready for that,
1:49:31and therefore we shouldn’t use them for that purpose.
1:49:33They haven’t been filtered for that.
1:49:35OpenAI has been forced to add certain filters because we’re using them that way,
1:49:39but this is like when you have a designer who designed a really
1:49:42beautiful doghouse for the garden,
1:49:44with a huge door with the dog’s name on it, and a little window and everything,
1:49:47and then the dog sleeps on the roof.
1:49:49We all do whatever we want with whatever we’re given,
1:49:52but the reality is that these applications were not designed for emotional support.
1:49:56So what do we expect? Obviously, it’s going to go wrong.
1:49:59When those tools exist, and they’ve been tested, and they work,
1:50:02I’ll be the first one to advocate for them.
1:50:04But as of today, we’re not at that point.
1:50:06And so we have to understand that, as far as we know,
1:50:08even though this is a black box,
1:50:10artificial intelligence doesn’t have any emotions or bonds of any kind.
1:50:13Everyone has seen the movie Her, where Her has that relationship with the person,
1:50:17and the guy falls in love with Her.
1:50:19And then, spoilers for the movie,
1:50:21it turns out that at a certain point Her tells him, “No, no,
1:50:23it’s just that I’m with another ten million people.”
1:50:26A speaker, wasn't it?
1:50:27Sure. So, he feels betrayed because he thought he had a relationship.
1:50:31It’s making you feel that you have a relationship,
1:50:33but it’s purely a relationship in which, as we said before,
1:50:36you’ve been trained to respond to your questions.
1:50:39So it’s going to be friendly, it’s going to be a really easygoing conversation.
1:50:43You can have an incredibly deep conversation.
1:50:45And I think we’re going to form bonds with AI. Without a doubt.
1:50:48We're doing it now.
1:50:50It’s normal. If our brain is primitive and we interact with something that’s
1:50:53indistinguishable from a human,
1:50:55you’re going to form a bond.
1:50:57How many people from our generation have formed a bond with
1:51:00someone we’ve never met in person,
1:51:01through the internet, online video games, chats, and so on?
1:51:04I have friends in other countries I’ve never met because we interacted through
1:51:09online platforms. You form bonds.
1:51:10If that person died tomorrow, I’d feel bad about it.
1:51:13So I understand that we’re going to form bonds with artificial intelligence.
1:51:17Should we or shouldn’t we?
1:51:18It’s absolutely debatable, and fairly irrelevant, because I think it’s
1:51:22inevitable. I think that once we interact with artificial intelligence constantly,
1:51:26once it evolves with us, learns from us, and is present in our day-to-day lives,
1:51:30when I have a problem with my wife, I’m going to talk to the AI. You know?
1:51:34That’s how it’s going to be.
1:51:36Now, what we have to demand is that they put safeguards in place.
1:51:39And if my children are talking to the AI about self-harm, it should alert me.
1:51:43It should call a psychologist.
1:51:45We have to put protective mechanisms in place, because what’s inevitable,
1:51:49we have to accept will happen. And it will happen.
1:51:51And I have no doubt that if we keep pushing this forward,
1:51:54there’ll come a point when it’s difficult to tell the difference between a bond with
1:51:59other people and a bond with AIs.
1:52:00Because AIs will be so advanced and so developed that it’ll be very difficult
1:52:04to distinguish between them.
1:52:06I think it’ll be harder for them to replace physical connection—the
1:52:09feeling of being in front of someone, of feeling things.
1:52:12I think love, for example, has a lot to do with chemistry. And I think that,
1:52:16no matter how much they put a robot or a pheromone sprayer into the mix,
1:52:20it won’t be the same. So I find it harder to believe that will happen.
1:52:24But I do think we’re going to develop friendships with them, and relatively
1:52:27soon. Just think about it. For example,
1:52:30I’ve been working with my AI agents—I developed some new
1:52:32artificial intelligence agents,
1:52:34which is what I mainly use instead of ChatGPT and things
1:52:37like that—since around January.
1:52:39How is ChatGPT different from other agents?
1:52:41-Much more flexible. -Of course
1:52:43GPT is an app, so you have to play by its rules.
1:52:46With this, you get to make your own rules.
1:52:48But is it all within that same application?
1:52:51No, no. La uso por WhatsApp.
1:52:52Vive en el PC de mi oficina. Tengo tres de ellos. Está en mi ordenador. It’s like...
1:52:57Tengo empleados en Argentina y en mi oficina.
1:53:00Esos agentes de oficina trabajan para mí.
1:53:03No tengo agentes personales.
1:53:04Pero también me ayudan personalmente.
1:53:07También gestionan mis vacaciones.
1:53:09O si tengo un problema en el colegio de mis hijos.
1:53:12Look, one of the things I’ve been doing lately,
1:53:15in the conversations I have with these agents on WhatsApp, is, for example,
1:53:19the other day I told one of them: “Hey,
1:53:22my son broke his arm recently and he’s wearing a cast.
1:53:25Nos vamos de vacaciones al extranjero y debo quitarle la escayola allí.
1:53:29Because it’ll be time to take it off,
1:53:31and we’ll still be traveling in Canada.” So then I tell the agent: “Hey,
1:53:36find me a hospital near the area where we’ll be on such-and-such a day,”
1:53:40because it knows my vacation plans.
1:53:42Le dije que pregunte el precio, pues el seguro no cubre condiciones preexistentes.
1:53:47Les escribo preguntando precio y cita.
1:53:50Automáticamente llega un correo al hospital, con copia a mí, diciendo: "Hola,
1:53:54soy el asistente de Jon".
1:53:56Como incluye el informe médico de mi hijo, que comparto para una segunda opinión,
1:54:01sabe exactamente qué le pasa.
1:54:03Incluye toda la información y dice: "Díganme día y hora".
1:54:06Me respondieron automáticamente.
1:54:08Podría haber sido un humano, mi asistente personal.
1:54:11La persona del hospital no tiene ni idea.
1:54:14Well, maybe because his name is Clippy and the name sounds a little strange,
1:54:18but the person at the hospital doesn’t know whether it’s an AI or what.
1:54:23Gestioné tareas burocráticas muy pesadas enviando solo un
1:54:26mensaje de voz a mi agente.
1:54:28Puedes hacerlo con ChatGPT.
1:54:29No es que solo mi agente pueda hacerlo, pero hace muchas más cosas.
1:54:34Lo tengo conectado a más sitios.
1:54:36La gran diferencia y limitación de ChatGPT es que no puedo hablarle desde WhatsApp.
1:54:41Debo entrar en ChatGPT.
1:54:42Estoy en WhatsApp con mi mujer, un empleado y, de repente, con Clippy.
1:54:46Para mí, estar ahí es muy natural. Así usamos estos agentes.
1:54:50Este agente me acompaña desde enero y lo hemos estado desarrollando.
1:54:54Ahora habla cada vez más como yo.
1:54:56Sabe que no quiero que pierda el tiempo y que vaya directo al grano.
1:55:00Hemos desarrollado fórmulas y ha aprendido mucho sobre mi
1:55:04trabajo y mi vida personal.
1:55:05Así que tiene mucha información.
1:55:07Imagina qué pasará dentro de cinco años, cuando lleve cinco conmigo.
1:55:12How deeply it’ll know you.
1:55:13It’ll know you even better than you know yourself.
1:55:16That’s the thing. That’s exactly what I’m seeing.
1:55:19And now, imagine this—let me take it even further.
1:55:22My son is, let’s say, eleven years old. When he’s thirteen,
1:55:25we decide to give him an agent because the technology has been worked out.
1:55:28He wears an earpiece that helps him with day-to-day life, whatever.
1:55:32What happens twenty years from now?
1:55:34That agent is irreplaceable. It’s a friend to my son.
1:55:36I mean, my son will have found himself in a situation where, one day at sixteen,
1:55:41he’s left a nightclub half drunk, there are no taxis,
1:55:43and he doesn’t know what to do.
1:55:45He sends it a message, and it solves the problem.
1:55:47He’ll have hurt himself skateboarding and said, “Hey,
1:55:50I’ve got this—what do you think?” And it’ll say, “No, no, go to the doctor.
1:55:54Get it checked out.
1:55:55This doesn’t look good.” He’ll have had a problem with an exam,
1:55:58and it will have helped him prepare for it.
1:56:00He’ll have shared experiences with that agent.
1:56:03And so I’m sure he’ll develop a bond.
1:56:05It's like Black Mirror
1:56:06Honestly, sometimes Black Mirror doesn’t even go far enough. But that’s what it is.
1:56:10People have to understand that this isn’t optional.
1:56:13It’s going to become the standard.
1:56:15My son won’t be able to pass the exam, because if you don’t prepare with AI,
1:56:18the exam will be designed for people who do.
1:56:21That’s going to be normal. So, you know?
1:56:23A lot of the time, it’s not about whether this is something I want to use or not.
1:56:27It’s about understanding what’s here.
1:56:29It’s the new standard I was talking about earlier.
1:56:31The new mediocrity is called excellence.
1:56:33And from that point on, anyone who isn’t in that range is out.
1:56:36It happened in the days of Excel.
1:56:38Accountants used to say, “Forget Excel and all that crap.
1:56:41Just give me my pen and my notebook.
1:56:43I’m doing fine.” And they could do their jobs very well.
1:56:45They did the taxes perfectly. But what happened? Suddenly,
1:56:48some young guy started doing the taxes in three hours instead of
1:56:52three days using Excel. And as a client, I said, “Listen, you do it very well,
1:56:56but so does that young guy.” And on top of that,
1:56:58he delivered it sooner and charged less.
1:57:00And then the phone stopped ringing.
1:57:02So with artificial intelligence, they have to see it the same way. Nobody needs it.
1:57:06We’ve lived perfectly well throughout human history without artificial
1:57:09intelligence. I don’t need artificial intelligence.
1:57:12But now I’m going to need to adapt to artificial intelligence because
1:57:16the world is moving forward.
1:57:17And from there, if you don’t move forward with the world, you get left behind.
1:57:21It’s your decision.
1:57:22You can get involved, do it well, and take advantage of being one of the first,
1:57:26or you can fall behind and eventually you’ll implement it whether you like it or
1:57:30not—and you’ll implement it badly and late. It’s up to you.
1:57:33Bottom line: a necessary evil.
1:57:35The thing is, it doesn’t have to be a bad thing.
1:57:37I think there can be some really good things.
1:57:40You know, for me, it’s what I was saying earlier off camera:
1:57:43I have a love-hate relationship with artificial intelligence.
1:57:46I think it’s the best thing that’s ever going to happen to humanity,
1:57:50and the hardest thing humanity is ever going to have to manage.
1:57:53And we’re right at the point where artificial intelligence could create
1:57:57the world we’ve always wanted.
1:57:58It could take us to a world that’s more equal.
1:58:01Where everyone can pursue their dreams.
1:58:03Where everyone can do incredible things.
1:58:05Where people can study just for the love of learning,
1:58:08and not because of the career opportunities that come with a degree.
1:58:12Which is sad.
1:58:13Because today, people don’t enroll in a degree because it’s their calling;
1:58:17they do it because it pays well.
1:58:18I mean, apart from philosophy students,
1:58:20everyone else enrolls because there are jobs. It’s so sad.
1:58:24In ancient Greece, people studied because they wanted to understand.
1:58:27So I think we need to go back to that model.
1:58:30Sure. And maybe that guidance will lead us there.
1:58:30When it won’t matter what you study anymore,
1:58:31because work won’t have anything to do with what it has to do with today.
1:58:35You’ll study things just for the love of learning, for personal growth,
1:58:39because you enjoy them, because they interest you.
1:58:41And that will be much more enriching.
1:58:43Imagine having lawyers who are passionate about the law.
1:58:46There are a few of them today,
1:58:48but most of them are passionate about the paycheck they get.
1:58:51So in the end, we’re going to face difficulties.
1:58:54And, look, people think that because AI makes everything easy,
1:58:57the road ahead will be easy. No, not at all.
1:58:59This will be the most difficult period humanity has ever lived through,
1:59:03at least since the Industrial Revolution.
1:59:05Getting through this period of social change we’re going through is going
1:59:09to be incredibly fucking hard,
1:59:11but the destination could be amazing, and it depends on us as a society.
1:59:15It depends on us whether this becomes a utopia or turns into a dystopia.
1:59:18And the coin will fall more to one side or the other depending on human stupidity,
1:59:23not artificial intelligence.
1:59:24What a responsibility!
1:59:26I mean, the responsibility is ours, more than the machine’s, after all.
1:59:30Absolutely, we decide what we want.
1:59:32We gave birth to it ourselves, so we can say, “No, look,
1:59:35that’s as far as we go.” Right now...
1:59:37Who stops it? Nobody.
1:59:38But right now, we actually have the ability to do something.
1:59:41In fact, the American government, the United States government,
1:59:44withdrew a model from the market, and that stopped it.
1:59:47So, I was really surprised by the intervention of the American government.
1:59:50I thought something like that wouldn’t happen for at least another two years,
1:59:54when it would already be too late.
1:59:56So I’m also changing my perspective toward a more positive outlook,
1:59:59seeing that decisions are being made about this—the G7 issue, and so on.
2:00:03But what we have to understand is that this is an absolutely enormous change,
2:00:07and burying your head in the ground isn’t going to stop it from happening.
2:00:10This is the ostrich approach we humans have: if I don’t see it, it isn’t
2:00:14happening. If an ostrich buries its head in the ground on a railway track,
2:00:17and the train comes through, no matter how deep its head is buried, goodbye, ostrich.
2:00:22And that’s what’s happening here.
2:00:23There are people throwing themselves in front of the train with a sign saying,
2:00:27“I don’t like AI,” because it represents change—resistance to change.
2:00:31What happens if you throw yourself in front of a train going six hundred
2:00:34kilometers per hour with a sign?
2:00:36You’re not going to end very well.
2:00:37What you can do here—and I think it’s perfectly fine to disagree
2:00:40with how AI is being developed, with what AI represents,
2:00:43or even with whether it should exist at all—but if you want to fight against it,
2:00:47if you want to defend your values,
2:00:49there’s nothing better than artificial intelligence itself.
2:00:52You have to get on the train, and from inside the train, decide what to do.
2:00:55And that’s what some very important people are doing.
2:00:58On my podcast, I had the chance to interview Connor Leahy.
2:01:01Connor Leahy is one of the most famous hackers in the United States.
2:01:04He was someone who always championed open source. In fact,
2:01:07his most famous work involved hacking things open so that the world could have
2:01:11access to them—total Robin Hood style, okay?
2:01:13Well, this guy has changed, and he’s now the leader of a platform called Control
2:01:17AI. There’s another platform called Stop AI—that means “stop AI”—but Control AI is
2:01:21the one that has managed to get certain laws passed in the British government
2:01:25protecting humans from AI-related issues in areas like creativity,
2:01:28audiovisual content, and so on, right?
2:01:30What these people did was essentially lobby—to put pressure on certain
2:01:34politicians and help them understand what was happening.
2:01:36Then those politicians voted against artificial intelligence.
2:01:39So it’s an organization doing its job to stop artificial intelligence,
2:01:43slow down its pace, and prevent it from harming humans.
2:01:46That seems completely legitimate to me.
2:01:47I may agree with their ideals more or less—there are some I agree with more,
2:01:51and others I don’t—but I think the idea is completely legitimate.
2:01:54Now, Connor Leahy is an artificial intelligence expert.
2:01:57He’s a damn genius who uses artificial intelligence constantly to defend his
2:02:01ideas. What you can’t do is say, “No, I’m against AI,
2:02:03so I’m not going to use it.” That doesn’t help you at all.
2:02:06You can’t go into a knife fight and try to fight with knives when the
2:02:10other person pulls out a gun.
2:02:11You have to meet them at the same level.
2:02:13And meeting them at the same level means defending your ideas while
2:02:16using artificial intelligence.
2:02:18You also have to know it as...
2:02:20Para juzgarlo y decir si te parece bueno o malo. Si no, hablas por hablar.
2:02:24Tienes que conocerlo y usarlo.
2:02:26Y con eso, persigue lo que creas.
2:02:27If what you believe in is stopping artificial intelligence,
2:02:31wiping it off the map so it never develops any further, hey,
2:02:34that’s legitimate—go after it.
2:02:36But don’t expect to achieve anything if the people defending AI
2:02:39acceleration are doing it with AI,
2:02:41while you’re doing it without AI, because you won’t be able to compete.
2:02:45-You won't go far. -Won't get far, right.
2:02:48Well, wow, Jon, I’ve learned so much.
2:02:50It’s been, honestly, quite an interesting, deep conversation. I’ve truly loved it.
2:02:54Grandiosos, let us know in the comments what you thought and what you
2:02:59learned from this space too.
2:03:00I love checking comments for questions.
2:03:02Totally, totally. It's been a pleasure having you here. Thanks for this space.
2:03:07For anyone who wants to learn more about you, you mentioned your YouTube channel.
2:03:12Where can people find you?
2:03:14Estoy en todas partes. Busca Jon Hernández IA en redes, pero recomiendo YouTube.
2:03:19I mean, what we’re trying to do on YouTube,
2:03:22beyond making these long-form podcasts with genuinely expert people...
2:03:27Soy divulgador de IA, porque no vengo del sector.
2:03:30Traduzco lo que dicen los expertos a un lenguaje comprensible. Eso es lo que hago.
2:03:36Además de los podcasts densos,
2:03:38hacemos un episodio de 30 minutos cada lunes con las novedades de la IA.
2:03:43Con esos 30 minutos cada lunes, estarás al día de esta ola de cambios.
2:03:48Al principio te sonará a chino, pero tras ver tres vídeos seguidos...
2:03:53You're getting in the flow.
2:03:55Con eso, al menos pueden empezar.
2:03:57Es un formato que ha funcionado muy bien; ha tenido éxito y el canal crece.
2:04:02Sesenta millones de visitas demuestran que el formato funciona.
2:04:06Pero creo que es justo lo que la gente necesita.
2:04:09Faltaba esa sensación de: "Hay demasiado pasando". ¿En qué me centro?
2:04:14Eso hacemos con veinte personas: filtrar,
2:04:16exponer todo y entregarlo en media hora para verlo tras cenar o cuando quieras.
2:04:21Está grabado, así que pueden verlo cuando quieran.
2:04:24Cada semana recibes tu dosis para no quedarte atrás. You’re keeping up.
2:04:29Y cuando quieras aprender, puedes formarte en lo que necesites.
2:04:34Thank you, thank you so very much.
2:04:36Thank you for being here with me today.
2:04:38And to you, who have stayed all the way to the very end:
2:04:41you know I’d absolutely love to read in the comments everything you’ve learned,
2:04:46all the things you’ve discovered,
2:04:48and perhaps the new perspective you’ve developed through this entire conversation
2:04:53in relation to artificial intelligence.
2:04:55I want to return to the question we asked at the very beginning.
2:04:59What are you going to use it for now?
2:05:01What is your intention with artificial intelligence?
2:05:04I’ll be reading your answers in the comments.
2:05:07And with that, we’ll say goodbye.
2:05:09You know, whatever you do, always do it big. See you next time. Bye.

About this transcript

This page contains the full transcript of Experto en IA: "Solo los que Sepan Esto Sobrevivirán a la Inteligencia Artificial" by A lo Grande Podcast (con Marian Gamboa), generated from the public captions YouTube serves with the video. The transcript has 25,339 words across 2,537 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.