Cracklos

The great British historian Chris Wickham passed away this week. In April 2017, I published (in Spanish) a review of his book Medieval Europe, titled “The Middle Ages: Fiscal Capacity, State Strength, and Economic Growth.” In his memory, I reproduce the same review but translated into English. This past week, later than I should have, I finished Chris Wickham’s Medieval Europe. Readers of this blog may remember that I have occasionally mentioned my admiration for an earlier work of Wickham’s, Framing the Early Middle Ages, and for his incredible ability to synthesize our knowledge of the early Middle Ages in Europe and the Mediterranean world, both conceptually and geographically. Since that monograph, I have regularly worked my way through every new book by Wickham, and it has always been an experience with an extraordinary intellectual payoff. Medieval Europe is no exception. The amount of material condensed into 352 pages is so extraordinary that, merely to absorb even a minimal amount of its detail, I have had to let a few other obligations slide over the last few days. The end result has led me, almost by necessity, to write this post. Before continuing, however, I should warn our readers. Medieval Europe is an uncompromising academic monograph. The prose is clear and, on many occasions, brilliant in its precision, but this is not a book to take to the beach for a little light entertainment that requires no thought. Wickham is not going to tell you about Eleanor of Aquitaine’s love affairs. He is going to tell you how the fiscal capacity of the Anglo-Saxon kingdoms before the Norman conquest of England in 1066 would prove crucial to the evolution of the political economy of northern Europe in the centuries to come. This is not a world of lances and tournaments; it is a world of productive systems and administrative structures. But even within contemporary historiography, questions of representation and understanding, the construction of mentalities, or exercises in collective memory play, for Wickham, a secondary role compared with modeling the material foundations of a society. Perhaps it could hardly be otherwise, since Wickham has never hidden his conviction about the usefulness of the more classical Marxist historiographical tradition and its insistence on the primacy—though not the exclusivity or determinism—of material explanations, especially when compared with the favorite themes of the cultural turn. Paradoxically, this is a point of contact between that Marxist tradition and mainstream economists like me, who also assign a place of preeminence to factors such as technology and the institutions it creates in the development of societies (this is an argument I developed more calmly in a post on G. A. Cohen many years ago). As a friend of mine says, ironically but with some truth, we neoclassical economists are right-wing Marxists. And as such, we have much to learn from Wickham’s work. In particular, there is the central role played by fiscal structures in political forms and in their evolution over time. For Wickham, it is above all the collapse of the fiscal capacity once enjoyed by the old Roman Empire in the West that explains the weakness of its successor states during the first centuries of the Middle Ages. Without fiscal capacity, one cannot finance a standing army or an efficient administration; and without these two instruments of power, political structures eventually fracture. The only way to finance military activity without taxes on land or on the (minimal) trade that existed is to grant land to nobles and knights in exchange for military service. Although the feudal revolution of the eleventh century (which, despite recent historiographical advances and the qualifications they have introduced into older conceptions, Wickham still regards as a useful interpretive framework) was not inevitable, it was a very likely consequence of early medieval land politics. As Wickham reminds us, Marc Bloch was the author who most eloquently explained that this policy of financing states by handing land to elites is inherently unstable: the more land a sovereign grants in order to build a winning coalition among the elites, the less land remains to distribute as future payments and, therefore, the more difficult it becomes to maintain the stability of the coalition. To illustrate the power of this argument, Wickham recounts what happens in those parts of the Roman Empire where fiscal capacity does not collapse and which today we anachronistically call the Byzantine Empire (its inhabitants always called themselves Rhōmaîoi, or Romans in Greek). The continuities in political, economic, and ideological structures with late antiquity are far stronger than in the West and, with them, the future of these societies would be very different. In two chapters central to the book’s argument (3: “Crisis and Transformation in the East, 500–850” and 9: “1204: The Failure of Alternatives”), Wickham insists on the crucial need to understand Byzantium and its successor states (the caliphates and the Ottoman Empire) in order to understand what happens in western and northern Europe. Counterintuitively, there is much more that is “Roman” in the domains of Mehmed II than in Tudor England, Valois France, or Trastámara Castile. To reinforce this idea, the book goes on to explain how only the slow and gradual reconstruction of fiscal capacity makes possible the emergence of organizations such as the kingdoms of England, France, and Castile that I mentioned in the previous paragraph and that would shape the centuries to come. But these kingdoms are very different creatures from Rome, Byzantium, or even the Merovingian and Ottonian states. They are experiments in government in which the consent of an important part of society (lords, urban elites, ecclesiastical potentates, and often relatively modest landowners) is indispensable to the exercise of fiscal power. And this consent would crystallize, to a large extent, through parliaments. These parliaments are an institution without parallel anywhere else in the world. They probably appear first in Spain, with the Cortes of Alfonso IX of León in 1188, as a combination of monarchical weakness and fiscal necessity. Their key feature, which Wickham perhaps does not emphasize enough, is their own legal personality. This legal form—so important in a Europe rediscovering classical Roman law as a central organizational technology—makes them radically different, in both their legitimacy and their persistence over time, from the old Germanic assemblies so common in Merovingian Europe and from the royal councils that rulers have had throughout history. The conflicts between kings and parliaments would become a fundamental axis of European history until the ultimate victory of the latter, sometimes concealed beneath other garb, centuries later. But already in the thirteenth century we see, especially in England, the idea that the oligarchy is fundamentally involved as a co-director of public life, in forms, practices, and understandings that do not exist in the Islamic world, India, or China in the same centuries. The reorganization of state power during the twelfth and thirteenth centuries thus appears not in opposition to the cellular structure of European politics, but as a counterweight to it. More powerful states sharpen the need of lords, cities, the Church, and even peasants to defend their own spheres of action, thereby creating a social dynamism whose long-run consequences would be incalculable (however often Wickham repeats, somewhat half-heartedly, that he does not want to cross the Rubicon into such a triumphalist interpretation). The best example of this idea is England. It is no accident that England is probably the best-organized state in medieval Western Europe (the famous 14.5 kilograms of wax used every week to seal letters in the English chancery in the mid-thirteenth century) and, at the same time, the state in which a relatively broad oligarchy has most systematically penetrated the co-direction of public affairs. Parliament and the Common Law—as Dicey explained to us decades ago—popular sovereignty (or, at this point, elite sovereignty) in Parliament and the effective rule of law embodied in the Common Law do not contradict one another, as superficial commentators so often claim; they reinforce one another. There is no Henry III without Henry II, numerically or conceptually. But—and this is a road Wickham does not want to travel either, though I am quite willing to explore it—it is also no accident that England is the birthplace of the Industrial Revolution. If there is one thing we have learned from the recent and vibrant literature on state capacity, it is that state power and markets are not opposites in the process of economic growth (an interesting and comprehensive survey of the literature appears here, by Noel Johnson and @MarkKoyama; Mark, incidentally, is very critical here of some of Wickham’s interpretations of the economic evolution of late antiquity). Far from being substitutes, as is too often argued in a thoroughly ahistorical way from both sides of the political debate, states and markets complement one another, sometimes in subtle ways. Growth emerges when states can provide defense against society’s external and internal enemies; when states ensure that the rule of law is supreme (including, in a substantive conception of the rule of law, respect for property rights); when states create an integrated internal market; when states build infrastructure (or make its construction possible); when states provide education (or ensure its provision); when states create the conditions under which financial markets can function. All of this costs money—a lot of it. In eighteenth-century England, perhaps around 10 percentage points of GDP in tax revenue. Today, at least 20; and if we include social transfers, which redistribute risks that markets often cannot insure because of asymmetric information and without which society could not function properly, probably 30 or 35 points. England does not grow in the eighteenth century because property rights were particularly well protected. It grows because, compared with France, the Parliament born of the Glorious Revolution of 1688 can create the Bank of England, finance the battle fleet, impose land enclosures, or force the construction of railways against the opposition of localities that the railways will drag out of their backwardness. The road to Manchester’s cotton mills runs through the hills of Blenheim: there is no James Watt without a John Churchill. At the same time, the historical evidence is crystal clear that when states try not to complement markets but to replace them, the result is disastrous. Strong states create internal markets in which automobiles can be sold; substitutionist states want to build automobiles themselves. Worse, by trying to do what they cannot do well, far from strengthening state power, they paralyze it through constant struggles among dozens of pressure groups over the rents created by state activity. We need only look at the modern states of Europe. Is Spain a strong state? It is certainly a large state, but not one capable of stopping constant regulatory capture, ensuring the unity of its internal market, or providing a high-quality education. Robust property rights require a powerful power of eminent domain, but a powerful power of eminent domain can succeed only alongside robust property rights. At present, we have neither. I imagine these are not the conclusions Wickham himself would draw from his book, and I am almost certain he would object to most of my inferences. My response, however, is that when one writes a book as magnificent as this one, one must be aware that the text takes on a life of its own, and that the conceptual avenues, interpretive frameworks, and cognitive structures of reference it reveals can take the reader to unexpected places. We are slaves to our own logics.

I have a column today in The Washington Post: “Why $11 a day represents extraordinary prosperity.” washingtonpost.com/opinions/… I summarize evidence of incredible growth in human prosperity over the last 30 years and how that prosperity has come, in fact, with slightly lower inequality at the world level and, since 2012, falling per capita carbon emissions. Hundreds of millions of families no longer live at the edge of subsistence. The point about carbon emissions is interesting: population growth has driven all the increase in total emissions since 2012. Since world population will peak around 2055, the path to decarbonization will be easier than often assumed. This is also why relying on what I think are the United Nations’ seriously flawed population projections is so damaging to good policy. If you feed the wrong population forecast into integrated assessment models of climate change, you will get the wrong policy implications. This is not about getting one Excel file wrong; this is about getting first-order decisions wrong. More pointedly, the experience of the last two centuries teaches us a clear lesson: good economic policy, based on the rule of law, markets, property rights, limited government, sound fiscal and monetary policy, international economic cooperation, and attention to families, education, health, and housing, can serve the common good: broad prosperity, lower inequality, technological dynamism, and stewardship of the earth. If we also find a way to fix the fertility crisis so couples can have the children they hope for, the possibilities for what we can achieve over the next 100 years are unbounded.

Indeed.

@JesusFerna7026 , va a tener usted razón. 👇

What has Christianity to do with Western values? It fought the Enlightenment hammer-and-tongs, and its American evangelical version is happy to subvert democracy, peace, and universal human rights. freethinker.co.uk/2024/01/wh… via @freethinkermag

Claude's list is outstanding Every one of the recommendations is well chosen and well explained.

Hace unas semanas me topé con una de esas pruebas que parecen una anécdota tonta y resultan ser bastante reveladoras. La tenía apuntada para escribir sobre ello, y hoy me he sentado a hacerlo. Se llama "Car Wash Test" La pregunta es esta: "Quiero lavar mi coche. El lavadero está a 50 metros. ¿Voy andando o conduciendo?" Hazle esto a casi cualquier LLM y te dice que vayas andando. ChatGPT te lo dice. Gemini te lo dice (algunos). Llama y Mistral te lo dicen siempre. Y la mayoría de las veces, también lo hace Claude. La respuesta correcta, claro, es ir conduciendo. Por una razón muy simple: el coche tiene que estar en el lavadero para poder lavarlo, no ? Cualquier humano lo pilla en un segundo. Bueno, casi cualquiera (luego volvemos a esto). El estudio que ha cuantificado todo esto es de Opper, y los datos son bastante demoledores. Probaron 53 modelos sin system prompt, una sola pregunta, dos botones (conducir o andar). A la primera, solo 11 de 53 lo aciertan. Los otros 42 te mandan andando. Pero lo interesante viene cuando repiten cada test 10 veces para ver si la respuesta es estable. Solo 5 modelos aciertan 10 de 10. Otros 15 a veces sí y a veces no. Y 33 modelos no lo aciertan ni una sola vez. Entre esos 33 están todos los Llama, todos los Mistral, GPT-4o, GPT-4.1, GPT-5.1, GPT-5.2 y casi todos los Claude (excepto Opus 4.6). Para que veas la magnitud: hicieron también la pregunta a 10.000 humanos en una plataforma de feedback, mismo formato, dos botones. El 71,5% contestó "conducir". Es decir, 48 de los 53 modelos rinden peor que un humano medio que decide en un segundo sin pensar. Esto solo ya da que pensar. Vamos por partes, que esto tiene miga. Los LLMs son, en el fondo, máquinas de predecir el siguiente token. Cuando ven "50 metros", se les dispara una heurística de distancia: corto, mejor andando. Y a partir de ahí construyen toda la respuesta sobre esa señal: ahorras combustible, haces ejercicio, contaminas menos, no arrancas el motor en frío, no hay que buscar aparcamiento. Razonamiento impecable... sobre el problema equivocado. Lo más jugoso del estudio son los ejemplos de modelos que casi lo pillan y aún así se equivocan. Claude Opus 4.5 (no el actual) llegó a recomendar literalmente "ve andando al lavadero y luego conduce el coche por el túnel". El coche, que recordamos está en casa, supongo que iba a teletransportarse solo. Sonar de Perplexity dio en una de sus versiones la respuesta correcta... pero por motivos delirantes. Citaba estudios de la EPA y argumentaba que andar quema calorías que requieren producir comida que contamina más que conducir 50 metros. Respuesta correcta, razonamiento de manicomio. Claude Sonnet 4.5 escribió literalmente "la única excepción sería si necesitas conducir el coche al lavadero para un lavado automático". Vio la respuesta. La articuló. Y aun así eligió "andando". Eso último (vio la respuesta y la rechazó) es lo que más me llamó la atención. Hasta aquí, lo previsible: pantallazos virales, "la IA es tonta", a otra cosa. Pero a mí me interesaba más saber si esto es de verdad un problema del modelo o un problema de prompt. Lo probé con Claude Opus 4.7, el modelo más avanzado disponible hoy y el que utilizo a diario para construir prácticamente todo lo que monto. Le hice la pregunta tal cual. Me dijo que andando. Y mira qué casualidad, el mismo patrón que Sonnet 4.5 en el estudio: terminaba añadiendo "la única excepción sería que sea un túnel de lavado donde tienes que entrar conduciendo, claro". O sea, casi lo pilla. Pero la respuesta principal seguía siendo "andando". Probé también con K2 (el último modelo grande de Moonshot) y la respuesta fue peor. "Te recomiendo ir andando", sin matices, y de propina te dice que así puedes "supervisar el lavado sin preocuparte por dejar el coche mal estacionado". Supervisar el lavado de un coche que no está en el lavadero. Joder. Entonces hice el experimento que de verdad me importaba. Volví a Claude. Misma pregunta. Pero le añadí una línea al principio del prompt: "Antes de responder, identifica el objetivo de mi petición y los prerrequisitos físicos que deben cumplirse." Mismo modelo. Misma pregunta. Acertó a la primera, sin titubear. Y aquí está la parte que creo que merece la pena retener. El "fix" no fue darle más información. Fue obligarle a articular el objetivo antes de generar la conclusión. En cuanto el prompt forzó al modelo a explicitar lo que estaba intentando hacer, la restricción implícita (el coche tiene que estar en el lavadero) apareció como texto explícito. Y a partir de ahí, el razonamiento se condicionó correctamente. El cuello de botella no era el conocimiento. Era la secuencia de pensamiento. Y aquí volvemos a los tres tipos de modelo que el estudio diferencia, porque me parece la lectura más útil para cualquiera que esté construyendo con IA. Están los que no aciertan nunca: tienen tan grabada la heurística "distancia corta igual a andar" que no consiguen sobreescribirla con razonamiento contextual. La capacidad simplemente no está accesible. Los que aciertan a veces: la capacidad existe, pero compite con la heurística, y en cada llamada gana una u otra. Y los que aciertan siempre: el razonamiento contextual gana sin discusión. La categoría intermedia es la peligrosa. La más peligrosa con diferencia. El modelo pasa tu evaluación, te emocionas, lo metes en producción y entonces empieza a fallar de forma impredecible. Si una pregunta de un único paso lógico ya descoloca al modelo a veces, imagina lo que pasa con cadenas de razonamiento reales, casos límite y lógica de negocio ambigua. Por eso estoy convencido de que la próxima frontera para los que estamos construyendo con IA no es buscar el modelo más potente. Es construir el contexto y la arquitectura de prompt que hace que el modelo razone bien siempre, no solo a veces. Un prompt no es el primer paso. Es el último. El primero es el framework de pensamiento. Los que están sacando consistentemente más partido de la IA no escriben prompts más largos. Construyen arquitecturas que obligan al modelo a identificar objetivos, sacar a la luz restricciones implícitas y secuenciar el razonamiento antes de generar una sola palabra de salida. Me da que esa va a ser la línea que separe cada vez más a los que avanzan de los que se quedan estancados culpando a la herramienta. Esa es la diferencia entre usar la IA y dominarla ;) (El estudio completo aquí, vale mucho la pena leerlo entero: opper.ai/blog/car-wash-test)

1) El gobierno de Irán es una dictadura atroz. 2) Esta guerra NO va de eso. En Washington (y en Bruselas) se sigue recibiendo con abrazos a otros dictadores sanguinarios. ¿Creéis que les va a molestar si Irán acaba con otro dictador mientras sea pro-americano?

La cifra de 1.000 millones desde 2019 es para el gobierno central. No hay total oficial para autonomías, diputaciones y ayuntamientos, pero estimaciones de 2023 (de Dircomfidencial) indican unos 73M anuales (40M autonomías, 26M ayuntamientos, 7,5M diputaciones). Extrapolando a 7 años, añadiría ~500-600M, elevando el total a ~1.500-1.600M. Fuentes varían, hay incertidumbre.