Privacy
We use necessary cookies for the living book to work, and we'd also like to use anonymous analytics to improve it. Your reflections are always private and we never analyse them. More details in the cookie policy.
Reading
Section01
Let me tell you why I am here, and why you are reading this book.
I have always been one of those people who need to touch, test, and experiment. If a new technology product or service appears on the table, I cannot help wanting to get my hands on it before anyone explains what it is for. So when, in 2022, the first GPT-based systems began to circulate in very limited circles and, shortly afterwards, the first access to ChatGPT arrived, I was there: exploring in the dark, with no manual, no reference community, and hardly anyone to compare notes with. It was almost virgin territory. I tried the obvious things: simple questions, translations, summaries. None of those first interactions truly shook me. Until one day, driven by that mix of curiosity and healthy skepticism every engineer carries somewhere inside, I decided to give it a real challenge.
As a chess player, I know the eight queens problem well: placing eight queens on a chessboard so that none threatens any of the others. It is not a trivial problem. It has more than one solution, and solving it requires a recursive algorithm that explores and discards possible combinations. Any experienced programmer knows that this kind of logic requires thinking through the structure before writing a single line of code. I wrote the prompt in a completely plain way, with no tricks, without specifying the language in detail, without defining the architecture: "I want you to write Java source code that solves the problem of placing eight chess queens on a board so that none of them threaten each other." Nothing more. And, within seconds, there it was. I copied it, pasted it, compiled it, ran it... and what appeared on my screen left me frozen: a simulated chessboard, eight queens perfectly placed so that none threatened any other, and a button that allowed me to navigate through the different valid solutions to the problem. I sat there in silence for a while, staring at the screen and checking the validity of each possible solution.
That silence had a particular weight. Because in that instant I understood that what had just happened was not a sophisticated autocomplete exercise. That machine had understood what I was asking, even though I had not been especially precise. It knew enough chess to understand how queens attack. It had grasped the complexity of the specific problem. It had designed and implemented a correct recursive algorithm. It had translated that logic into functional Java code. And, on top of that, it had built a user interface with navigation logic that responded exactly to what my question, deep down, actually needed. It had not only solved the problem. It had interpreted it for me.
FormatsBuy
Each format has its place. The print book to underline and revisit. The eBook to carry always. The web app to live the living edition.
Coming soon
We're closing distribution. In the meantime, open the living edition or activate your code.
Print book
Print edition via Amazon. The full reading experience: underline, annotate, return to the pages that matter.
Coming soonNotify meeBook
Kindle edition. Synced with your library, ready to highlight and export notes.
Coming soonNotify meeBook
Read on Android, web, or any device connected to your Google account.
Coming soonNotify meWeb app · Living book
The living layer: navigable index, infographics, tools, mentor, and news that update the book as reality shifts.
Open living bookLiveThis is already happening
Facts from the present that illuminate or challenge this chapter's thesis.
Google DeepMindJul 21, 2025Supports the thesis
En un año, DeepMind pasa de plata a oro: el modelo resuelve los problemas en lenguaje natural, sin formalización Lean, dentro del límite de tiempo oficial de la competición.
Editorial take
El delta es de 12 meses. Para los lectores: este es exactamente el ritmo del que habla el libro — capacidades que el año pasado eran "casi humanas" pasan en un curso a "top humano". Si tu hoja de ruta tiene horizonte de 3 años, replantea.
OpenAIAug 7, 2025Supports the thesis
Mentor
Ask the mentor about the demo. It only answers from what this chapter says.
Start here:
I remember thinking, almost feverishly, about everything this implied beyond the chessboard. If it could do this, what could it not do with a legal, medical, financial, educational, or industrial problem? How many professions, how many processes, how many certainties built over decades were about to be reconfigured from the ground up? It was not an orderly or calm thought. It was more like an avalanche. The same sensation I imagine the first engineer must have felt when he saw a steam engine move a piston and understood, in a fraction of a second, that the world would never be the same again. That was 2022 for me. The compass had turned, and it never went back.
I admit it: it was not the first time I had felt that kind of jolt. There are moments in the life of a technology professional when the ground shifts beneath your feet and you know, with a certainty that needs no external validation, that something fundamental has changed forever. The first of those moments came for me in the year 2000, with the Internet. I remember the feeling perfectly. Suddenly, all the information in the world was beginning to connect: accessible, alive, in motion. Distances collapsed. Intermediaries became less necessary. Business models that had worked for decades began to creak. At first, it was a silent revolution, almost invisible to most people. But for those of us working in technology, it was absolutely clear that this was not a passing trend. It was the beginning of a reconfiguration of civilization.
Then came the second wave: mobile devices. The Internet was no longer something you connected to from a desktop computer in a fixed place. Suddenly, the Internet was you. It lived in your pocket, followed you everywhere, and brought with it new platforms, new economies, and new social behaviors that nobody had fully anticipated. Each of those waves was enormous and transformative. Each created winners and losers on a scale never seen before in modern business history.
But when I compare those moments with what I was seeing in 2022, there is one truly consequential difference. It is not a difference of nature. It is a difference of speed. The Internet took years to penetrate the everyday lives of most people. Smartphones needed roughly a decade to fully redesign the global social and economic fabric. Generative AI, however, did not offer us that courtesy. ChatGPT reached 100 million monthly active users in just two months, according to UBS estimates cited by Reuters, and for a time it was the fastest-growing consumer application ever documented. Two months. That is not an anecdotal figure, nor a number designed to impress in a presentation. It is the clearest signal that this time the rules of the game are radically different. There is no time to watch from the sidelines. There is no maturity cycle slow enough to allow us to adapt later. The window between the moment something appears and the moment it has already changed everything is, right now, smaller than ever. And that realization (the awareness that the clock no longer runs at the rhythm we learned to read) is what drove me to write this book.
GPT-5 reemplaza a GPT-4o como modelo por defecto de ChatGPT. Cambia hacia un sistema de "clases de inteligencia" que enruta automáticamente cada consulta al sub-modelo apropiado, y reduce alucinaciones en torno al 45% con búsqueda activada.
Editorial take
Para los lectores: la noticia no es el modelo, es la cadencia. GPT-5 llega menos de 18 meses después de GPT-4 y, en los meses siguientes, OpenAI ha publicado 5.2, 5.4-Cyber y 5.5. Si tu plan de IA se basa en "el mejor modelo del mercado", lo perdiste antes de aprobarlo.
NvidiaMar 18, 2024Supports the thesis
Blackwell ofrece 30× más rendimiento en inferencia LLM frente a H100, con consumo y coste 25× menores. Adoptada por AWS, Google Cloud, Azure, Oracle, Cisco, Dell y Supermicro desde su anuncio.
Editorial take
La curva exponencial del coste por inferencia se consolida. Para los lectores: cualquier modelo de negocio basado en "la IA es cara" caduca cada 18 meses. Recalcula tu unit economics asumiendo que el coste por consulta cae un orden de magnitud cada generación.
OpenAIJan 21, 2025Adds evidence
Joint venture con compromiso de 100.000 M$ inmediatos y 500.000 M$ a cuatro años para construir centros de datos AI en suelo estadounidense. Trump anuncia el proyecto en su segundo día de mandato.
Editorial take
La IA se ha vuelto infraestructura crítica al nivel de la electricidad o las telecomunicaciones. Para los lectores: cuando un país compromete medio billón de dólares en cuatro años, es porque ya no es una apuesta — es una carrera militar-económica con presupuesto de defensa.
NatureMay 8, 2024Adds evidence
Google DeepMind e Isomorphic Labs publican en Nature una arquitectura de difusión que mejora drásticamente la predicción de interacciones entre biomoléculas. AlphaFold Server abre acceso gratuito para investigación no comercial.
Editorial take
La química computacional cruza una frontera: ya no "sugiere candidatos", construye el complejo molecular completo. Para los lectores: el ciclo del descubrimiento de fármacos se va a comprimir de años a meses. Los modelos pasan de oráculos a colegas de laboratorio.
NvidiaMar 18, 2025Adds evidence
Anunciado en GTC 2025. La nueva generación, orientada a inferencia de razonamiento, multiplica por 50 los ingresos potenciales de una AI factory frente a Hopper. Despliegue previsto a finales de 2025.
Editorial take
12 meses entre Blackwell y Blackwell Ultra. Para los lectores: la cadencia es ya anual y, si compraste hace un año, tu hardware acaba de bajar al segundo escalón. Los CapEx de IA tienen una vida útil más corta de lo que dicen los modelos contables.
Lawrence Livermore National LaboratoryOct 15, 2025Adds evidence
Lawrence Livermore confirma su décima ignición. El experimento de abril 2025 entregó 8,6 MJ desde 2,08 MJ de láser (ganancia 4,13), récord histórico. El de octubre 2025 produjo 3,5 MJ. El programa entra en fase de optimización industrial.
Editorial take
La fusión deja de ser ciencia ficción y pasa a ser ingeniería de yield. Para los lectores: una vez que algo se vuelve repetible, el calendario lo pone el capital, no la física. La conversación cambia de "si" a "cuándo en la red".
GoogleNov 18, 2025Adds evidence
Gemini 3 Pro y la variante Deep Think suceden a Gemini 2.5. Deep Think alcanzó nivel de medalla de oro en la Olimpiada Internacional de Matemáticas 2025, una capacidad que antes estaba reservada a humanos top mundial.
Editorial take
El razonamiento matemático "de élite humana" deja de ser exclusivo. Para los lectores: si tu ventaja competitiva descansaba en la complejidad de tu análisis, prepárate. La frontera no se mueve hacia la creatividad o la ética, se mueve hacia ahí.