Nuclear fusion, just like nuclear fission, makes use of the strong force to harness energy. As can be inferred from its name, the strong force is incredibly powerful amongst the fundamental forces of the universe; two orders of magnitude stronger than the electromagnetic force that governs over chemical bonds. The difference between fusion and fission is that while fission (and thus contemporary nuclear power) splits unstable heavy isotopes¹ into smaller ones, fusion takes isotopes of hydrogen (the lightest element) and merges them together to create heavier helium. Both of these release energy in the process, but fusion is considerably harder to succeed in.

Sometimes, when talking about llm alignment or the internet, people will mention “instructions for how to make a nuclear bomb” as something very arcane and secret. In fact, it is technically quite easy — it simply requires a lot of resources.² Nuclear power (and weapons) harness energy by concentrating a lot of fission-material in a small space. The difficultly in doing this for weapons comes from how to quickly concentrate this before the explosion rips the bomb apart, and for nuclear power the difficultly is in controlling the chain reaction — getting a usable amount of power without letting the process spiral out of control.

For nuclear fusion, the process is reversed. Getting fusion to occur at all is quite difficult, and even when you can make it happen it is even harder to (i) create more energy³ than was used to create the reaction and to (ii) extract that energy into some useful form. The American National Ignition Facility is the only experiment that has generated a net-positive amount of energy through fusion, and it did so by firing 192 laser beams into a small pellet. How does one create electricity from such a process? An alternative approach, magnetic confinement, simply heats a homogeneous mass of hydrogen plasma to very high temperatures — insulating it from its surroundings by a very strong magnetic field that keeps it floating in a vacuum. The very high heat means that ions will strike each other randomly at very high velocities, fusing together in the process.

The reason that the hydrogen atoms need to have a very high (relative) velocity is so that they can overtake the initial resistance of the electromagnetic force. Regular hydrogen is only composed of a single proton, and since it is positively charged it will push away any other positively charged hydrogen. Simple hydrogen atoms are also incredibly hard to “hit” because of their very small size. Using larger elements — like helium — makes the repelling electromagnetic force even larger, so instead one uses an isotope of hydrogen that includes either one or two neutrons, deuterium and tritium respectively. Deuterium is naturally occurring in small amounts and can therefore be extracted from plain water in a comparatively simple process, but tritium is faintly radioactive and thus does not exist naturally on earth but is only produced in nuclear reactions. Any fusion plant thus needs to create its own tritium if it wants to be successful commercially.

One of the largest problems with contemporary nuclear power is the (largely unanswered) question of nuclear waste. Of the 31 nations of the world who posses nuclear power, only Finland has a complete solution for long-term storage of waste. This is not a problem with nuclear fusion. Fusion reactions create neutron flux that irradiates materials, but those materials do not have the same intensities of radiation as fissile (or formerly fissile) material. The direct byproducts of fusion, largely helium-3 (an isotope of helium with one neutron) is not radioactive and can be used in various industrial applications or vented into the atmosphere.

Fusion will not be built for a long time. There is an astounding and impressive amount of work being done, much of it useful, but none of it is even in the same league as a commercial power plant, not to mention ballparks. As mentioned above, only a single facility has ever reached a net-positive amount of energy production. Saying that you will soon be building power plants is like keeping a campfire lit in the rain using a flamethrower and then saying you are going to build a coal power plant using your patented 24/7-flamethrower-method.

Even then, building futuristic has-yet-to-even-be-done-in-labs reactors at commercial scale would not give us limitless energy. There are many levels of break-even in fusion; the most simple one is of course the question of raw energy, but much of that energy is in hard-to-harvest forms like radiation of various kinds.⁴ Only the amount of energy released as heat can be reasonably used to boil water and drive a steam engine to create electricity. There are enormous energy losses in this process — fission plants can only harvest a third of the (thermal) energy produced, leaving ~66% as waste heat. Beyond that a power plant needs to be able to produce as much power as was used in the production of deuterium fuels, for the extraction and transport of lithium-6,⁵ the construction and maintenance of the facility itself, and various utility process related to all of the above.

Shrinking the cost of building fusion reactors is meaningless because their operational costs are so large and their benefits non-existent for power production (as of yet). The more you run a reactor, the more energy is lost and the more money you have burned. It is like solving our flamethrower-campfire by making the charcoal cheaper; it does not solve the fundamental underlying problem with fusion power which is that it does not produce energy. From the above article:

Energy Singularity has some advantages. With their stronger magnets, design experience, and domestic supply chain, they believe their reactors will be the most cost-effective in the world. […] Having already built a first-in-class hts tokamak under budget and on time, I trust their estimate.

Lowering costs does not improve your cost-effectiveness if the costs of your reactor are theoretically infinite. This is fine of course for research and development, and is thus not useless, but I can easily build a non-energy-producing fusion reactor myself for a far smaller budget than Energy Singularity’s.

But fear not, for we actually have a working form of limitless green fusion power that can supply the world, is tried-and tested, and has already been deployed at scale: its called renewables.

The problem with energy production is not whether or not it is feasible, but at what cost energy can be produced. Another huge source of power would be if everyone constantly rode electricity-producing bikes at home, but we have all collectively agreed (through the market) that there are more beneficial uses of our time and resources. If you were paid $100 for every hour you spent on such a bike there would likely be a lot more people doing it — thankfully there are better ways of producing large amounts of power cheaply. The main cost of power plants is in their construction, maintenance, and fuelling (if relevant) but for some reason that is conveniently forgotten when talking about fusion.

All renewable power sources ultimately derive their power from the largest and most efficient fusion reactor we are familiar with, the sun. Even fission power, that can in rough terms be summarised as

We found a rock that gets really hot no matter what, and if we put the rock in water it starts to boil and gives us infinite power.

is quite hard to do commercially because it requires expensive machinery, safety equipment, and facilities on top of the cost of obtaining this rock. That building a super-expensive plasma vessel surrounded by superconducting magnets would be cheaper is absurd. By contrast wind power has been profitable for centuries — powered by the sun as wind currents are created through the uneven distribution of sunlight across the planet.

This is not to say that fusion is “doomed” or will never succeed, it is rather just that (i) it is really quite difficult to do as a way to extract energy and thus will likely not happen soon and (ii) would not magically be able to supply everyone with cheap power. Fusion has always been a few-decades away because, while fusion itself is entirely possible to do in a lab, it is difficult to use as an energy source. It did not take all that much effort to go from nuclear weapons to nuclear power (1945–1951) but we have now had fusion bombs for 75 years and have yet to produce electricity from it. That the physics allow for the possibility means that it is hard to comprehend how nobody has just “done it” yet, but so far it has escaped the clutches of the world’s most talented engineers for three quarters of a century. Lewis Strauss said in 1954 that

It is not too much to expect that our children will enjoy in their homes electrical energy too cheap to meter[.]

This is the same sort of language used to describe a future of fusion power. It did not come to pass. Electricity seems to be one of the worst offenders for Jevons’ paradox — it was in fact the production of coal for energy that Jevons originally refereed to. Instead of thinking of a future technology that will swoop in and solve all of our problems with energy (cost, environmental impact, et cetera) we should focus on the solutions that are available today. The world’s energy system needs reform; it needs to be more integrated, cleaner, and in many ways larger as well. But most of all further developments need to be made in “the fifth fuel”, namely efficiency.

Even when fusion power inevitably⁶ becomes a reality, it will likely fulfil a similar niche as fission plants — used by advanced high-tech economies to provide stable base-load power and reduce dependence on energy imports. It will not solve the world’s problems overnight. Nuclear power plants famously take quite a long time to build, and a delicate high-tech machine like a fusion reactor is no different. Once commercial power becomes possible it will thus take many years, perhaps as long as a decade, to get up to speed in large-scale integration with the grid. Don’t hold your breath. ❦

The subject of artificial intelligence (ai)¹ is one that has become hotly debated in recent years. That the growth of ai’s capabilities has been partly driven by what is becoming increasingly clear financial engineering. These problematic elements has created a stark division of opinions that can generally be subdivided into the following categories:

1. The masses. A large, indifferent group who use llms in different capacities — stretching from the most common not-at-all to daily. Their unifying character is that, while discussion surrounding the ethical and day-to-day implications of ai are common, they primarily do not think about the philosophical, technological, political, macroeconomic, or in some other sense structural impacts of quick ai growth. Generally negative toward ai (and of how it seems to be loosely integrated into more and more consumer products) but not so much as the next group:
2. The Luddites. A highly critical and hostile group that consists of an over-represented share of (digital) artists, environmental activists, reactionaries and software engineers who prefer “hand-crafted” code. They are much more plugged in to the conversation than the every-man, but are at the same time reactionary in the Newtonian sense, explicitly denouncing any ai advances and proclaiming that ai will not be relevant after the inevitable popping of the bubble.
3. The Doomers. Strong believers in strong growth in the capabilities of ai models, but also worried that humanity’s collective ability to control them will grow slower than the aforementioned capabilities. Makes up the majority of the ai-safety community.
4. The Optimists. This group believes that not only will ai growth be fast, it will also grant mankind gifts that surpass all that humanity has achieved before, and we will hopefully live in som form of Fully Automated Luxury Communism.

I am not firmly enmeshed in any of these camps. I am too well-informed to be a part of the uninterested masses, and in some ways am more positive about the use-cases of ai than many others (although I do also do not reject the label of luddite either). On the division between “optimists” and “doomers” I find that both are often heavily separated from reality, and in many cases not well-read outside the fields of science-fiction (and perhaps computer programming). This is an attempt to inform these latter groups.

One of the favourite subjects of ai discussions are those of exponential growth. This is usually framed in some way that sets a “before” and and “after”, during which ai gains the ability to recursively self-improve and effectivise the manufacture of goods and services as well as the large-scale automation of r&d. I am a strong believer in the long-term ability for different ai models to improve all of these fields, but believing in exponential growth means that one has to face the reality of the costs of such development.

It took humanity roughly as long to go from using bronze swords to iron swords (3300-1200bc, roughly 2100 years) as it took to go from iron swords to gunpowder (1200bc-904ad, 2100 years) and from then it was only one millennium until we had the ability to destroy ourselves utterly with the invention of the atomic explosive (1945, 1041 years after the military use of black powder). The growth in technological capacity is already an exponential one.

The growth in computational potential is famously also exponential, and in many ways it is the driving force behind the capabilities of large language models. But increases in energy efficiency and computational speed are harder and harder to come by (Dennard scaling has been gone for 20 years now). This does not mean that growth in computation is likely to stall, but instead it means that more and more resources and inventions need to be produced in order to maintain the same rate of exponential growth (since the integral grows faster than the absolute value above it). Lithography machines are a famous example of this; likely the most advanced machines that humanity has yet to create, they consist of individual parts that are incomprehensibly precise.²

I feel compelled to quote Alexis de Tocqueville; There are two camps consisting of those who believe ai will dramatically accelerate technological development and those who believe it be a mere fad. “It is to be presumed that both are equally deceived”. I believe that ai will be a necessity for continued technological growth as the barriers are continuously rising, but that a very quick (less than half a decade) transformation is very unlikely. Many analyses of an ai-powered economic boom predict yearly growth rates in the mid-10¹% (~50%), but often also mention the theoretical doubling rates of insects or bacteria.³ But these latter examples only double very quickly under very unique situations with very large amounts of resources. To me this seems like a classic misunderstanding on the structure of human development.

That invention is the driving force behind development has been a dominant thought since the enlightenment. But one should not underestimate the importance of a more organic and more diffuse form of development. The economy of the prc, or of the four Asian tigers, doubled incredibly quickly, but when it did so it required enormous amounts of resources. The industrialisation of China needed huge quantities of steel, concrete, and energy to construct railways, highways and urban dwellings. The physical mass of this puts an upper bound on the speed for growth that a superintelligent ai can not fix.

Doubling the number of factories requires a large increase in construction-related steel production, and that requires an increase not only in logistics, but also in the exploitation of resources. A delivery of iron ore takes a month to travel by sea (one of the most efficient forms of transportation) from Australia to Shànghăi (上海).⁴ Delivery of construction equipment also takes an insignificant amount of time. The Abundance-argument regarding red tape is of course valid, and there areas in which improvements can be made in western economies, but one still needs to keep in mind an analysis connected to the situation on the ground.

Richard Danzing’s recent paper on cybersecurity and ai stresses an important fact when it comes to national security, that ai is here now and needs to be continuously integrated into the (American) national security apparatus in order to avoid a “Maginot moment”. This rings true for economic development as well — because the nature of war has become more and more industrial as the state and the economy has grown.⁵ Waiting for a potential “agi” to be invented before applying it to the economy is akin to (to take inspiration from Danzig) waiting to discover the mythic city of El Dorado before taking advantages of the discovery of America.

A particularly obscene phrase (although it seems to have been fallen slightly out of favour) is the statement that ai will in some sense kickstart a “fourth industrial revolution”. In fact, there has only been one industrial revolution — and we are all still in its midst. To showcase this, here is a graph of the world’s (human) population:⁶

Here it is again, but this time with a logarithmic y-axis:

It is clear that, while the population has generally been increasing steadily, there was a sharp inflection point (that can be clearly made out even in the second graph) some time in the 17th century, i.e. the industrial revolution. This development has not wavered, an despite demographic decline in Europe and east-Asia will likely continue until the end of the century. At that point, even if the population decreases faster, we will still be so unfathomably greater in number than we have been for practically all of human history. The power of industrialisation is awesome qua the sublime.

The division of industrial revolutions typically do so by focusing on the specific “driving” technologies. But those technologies do not come out of nowhere. They are needed, invented, applied, and (through their application) they drive new needs. A superintelligence would not have solved this because intelligence itself has not been the bottleneck for a long time. Technological growth is driven by application driving new needs — innovation is not a scientist sitting in a laboratory (that is science) but instead the practical needs developed out of application. The steam engine was developed long before it was employed because the steam engine qua invention could only exist in a proto-industrial society that needed mechanical energy for pumping and transport and that had ample supplies of wood and coal.

That is not to say that gai would not be useful for future scientific and technological development, but I merely want to point out that “a country of geniuses in a datacenter” does not a scientific revolution make. The shape of technological growth (since the industrial revolution that is) appears in some ways similarly to the following function:

\[ g_t(x) = c + \sin{x}, \hspace{1em} c > 1 \]

There is always some underlying r&d (in the literal sense) going on, and it in turn prompts new developments to begin. Because the above is the growth of technology, and because the value of sin never goes below -1, there is never any regression in technology (as I am aware of there has never been, apart from the Bronze age collapse). The integral of this then becomes

\[ G_t(x) = cx - \cos{x} \]

where \(cx\) continuously keeps going up, and mirrors in this sense the continuous advance of technology. There has been rapid change for a long time now — almost 300 years. Each generation that has lived through this development has had to grapple with its consequences. The field of political economy grew out of not just an attempt to control and steer this growth, but also an attempt to try and attempt how it happened in the first place.

Smith did so using the analogy of the pin factory to show how increasing markets made possible by international shipping, steam engines, and empire led to an increased division of labour and therefore an increase in productivity. Marx in turn tried to describe this by framing it as a consequence of primitive accumulation (through the expropriation of land) and capital’s ever-increasing needs for growth. In fact, there has not been a decade since the late 1700’s that has not been revolutionary in some way.

To say that the next decade will be transformative is an easy bet to take. We humans like to think that the world will stay the same forever, but our increasing industrialisation means that that will likely never be the case again. It seems strange then to write a post arguing against rapid change. But that is precisely the problem we are posed with — because we see the world as static we do not notice how much it is changing.

When we are presented with a technology as all-encompassing as ai it is entirely rational to see it as upending the world as we know it. But the world has seen — and is currently seeing — transformations of equal (if not greater) impact. The conversation around ai is stuck with its blinders on, just as those who (mistakenly) saw crypto as revolutionary did not focus on the broader technological change of the planet. We must therefore enlargen our view to encompass all these aspects in order to adequately think about how technology will impact our lives. ❦

This is something that should be established straight away. It might seem strange to say after having randomly gained quite a bit of money (for 14 year old) on perhaps the largest memecoin of all time (unless you consider Bitcoin a memecoin). But it really is clear to those who are either familiar with crypto and to those outsiders who know very little. The only group of people arguing against this fact are either scammers themselves or their victims who have been tricked into hoping they can “make it”. That most (in absolute number) cryptocurrencies are scams is problematic for the success of major currencies, but it is also a necessary consequence of the ease of creating new currencies, but it is at the same time not a characteristic unique to memecoins.

Creating a temporarily elevated value for something only to rugpull those who fell for it can be done with any commodity – ranging from meme stocks (like those of amc and Gamestop), to legitimate stocks, and of course even Tulips in the 17th century Dutch republic – But memecoins are very potent in this regard. Their largely unregulated (as opposed to stocks) and entirely digital nature (as opposed to physical goods) means that they can spread incredibly quickly to otherwise isolated buyers, and they can be created and abandoned incredibly quickly.

Cryptocurrencies are an incredibly dull topic; a combination of cryptographic has functions and distributed databases that allow for decentralised and trustless transactions. Does this sound like a topic to bring up at a party? Just like how I do not talk about my love of https or of Let’s Encrypt when going on a date, and likewise I do not talk about cryptocurrencies.

The goal of cryptocurrencies in my view is to act as a form of payment that is native to the digital, interconnected world we all live in. Digital Dollars/Euros/Crowns are of course useful, but they are fundamentally ill-fitting and primarily very centralised. When sending to someone (a store, a friend, should it even matter?) you go through a multitude of service companies like stripe, visa, your respective banks, Swift et cetera all to simply send money from A to B. Bitcoin (and even Monero) is comparatively simple and transparent. That is generally a good thing since it is a necessity for allowing trustless transactions. But it is not revolutionary. Crypto will not “change the world”, it will be a back-end infrastructure change that a few companies will do better than preëxisting ones and thus perhaps gain market share.

nfts (Non-fungible tokens) are also dreadfully boring in a good way. Today proving ownership of something requires a physical piece of paper in more occasions than you’d think. High-end artwork, real estate, and land are all incredibly valuable, and if that piece of paper is missing or destroyed¹ that is a very large problem for the owners. How does one prove ownership then? By getting everyone else to agree that you are the legitimate owner of that thing. Having everyone store a copy of that very important paper collectively is practically what an nft is. It is not some form of digital artwork or transformational thing. It is therefore ridiculous to say that one can “steal” and nft by saving an image; the image is not the token! Instead the nft is merely a short contract (maybe)² pointing to an url and saying that whoever owns the token is the owner of the contract.

I generally recoil at the mention of “community” more generally because my personality is fiercely individualist, but this ties into the above point. Creating a community around digital currencies or authentication systems is absurd. As we have established it is generally not interesting to talk about them, except as to the details of their specific implementation. Reading about the crypto community evokes similar feelings as “communities” that build up around certain free software projects (that have their own Discord server, with in-project social hierarchies et cetera). This is more a symptom of our increasingly fractured, isolated, and lonely society where desperate people look for friendship and warmth when there really should not exist any such community. The curl project is perhaps one of the most important libraries making up the modern internet – on its website it is stated that “curl is used daily by virtually every Internet-using human on the globe” – and it is developed by thousands of contributors (although the founder, Daniel Stenberg, has by far the most commits). There exists no curl community, because even though the project is very interesting and impressive it is merely meant to fulfil a task of transferring data between computers in various ways.

Similarity you are not a member of the crypto community if you are not involved in the implementation and development of crypto itself or crypto-adjacent systems – as a free-software developer, proprietary developer, employee at a bank or anything similar. Investing large amounts of money may give you clout amongst those who perform actual work, but it will not itself make you interesting — except from the perspective of being a source of capital for interesting work. Being a small-time retail investor does not even grant you those advantages; you are neither pushing development forward nor funding that development. Being some kind of “crypto”-philosopher or hype man is a fruitless effort since, as mentioned earlier, crypto is innately boring.

These three things are largely the source of all problems regarding crypto and nft discourse. What makes it so exhausting to read or listen to is that there are two primary camps; “haters” who know so little about crypto that they are not even worth listening to, and those who are blinded by greed, loneliness, or both and see (or at least speak of) crypto as either some sort of “people’s revolution” or a guaranteed way to make 100x returns on your “investment”. Both of these groups are sorely mistaken, and the reasons for this are the ones written above.

How to solve this? I see it as primarily a consequence of general economic despair and malaise, of people hoping for a better life and for connection through some shared experience. Solving this is not the task that distributed transaction system should try to do, it is a wider societal problem whose consequences can be seen in gold hype, political polarisation, used car prices, gambling advertising, and much more. As such it is a problem for all of society to try and solve. ❦

As a political science student, writing is a daily activity. And like any serious writer who cares about their writing, I care about the presentation and look of my documents. This has evolved, particularly recently, into an interest in typography — the art of formatting and shaping text. The source of this interest was originally my discovery of gnu Emacs’ ability to convert org-mode documents into \(\LaTeX\) ones back in high–school and even today this is still the basis of my process. It was accelerated by my increasing tinkering with this very website, having to design its appearance largely from the ground up.

Writing in org-mode allows one to focus much more on the pure textual content, leaving the formatting for later. Why not write in word and then copy-paste? Because then you still have to write all of the formatting rules (like emphasis) by hand – and Emacs (or another dedicated text editor like vim¹) is a much more powerful tool for text editing than any rich-text editor that tries to do formatting and content at the same time.

As my interest in typography has grown I have felt more and more under-served by the default \(\LaTeX\) article class. If you are not already using \(\LaTeX\) (or some equivalent like typst) there is likely nothing you can do that has a bigger impact on your typography than switching to it².

But there is a great amount of labour that has been done to improve upon the very high baseline that \(\LaTeX\) provides. This here is my collection of tips on how to give your documents the treatment that they deserve. Since I quite rarely touch .tex-files directly I do not like to rely on elaborate macros, even if they sometimes are very useful to attain the look that one is aiming for. Instead most changes are implemented as simple one-liner latex_header’s in org-mode. There are also no fancy classes involved, everything is done through the standard article-class.

For this post I have prepared a version of my post on the end of history in pdf format that we will successively improve upon. Over the course of this post we will turn the document on the left into the document on the right:

Naturally I should remind the reader of the Latin phrase de gustibus non est disputandum. What I deem beautiful may not seem the same to you. I hope however that the information that you find here can be of use even in the case of disagreement.

Fonts are what most people think of when they think of typography, and it (and perhaps line-spacing) is one of the few choices that normal people make when trying to improve their documents. Fonts are not everything, but since they do make up the majority of your content (if not all) they are very important.

First I would like to point out that no, Times New Roman (tnr) is not good enough. In fact, it is actively detrimental to the appearence of your document. tnr shows how you do not care about typography at all, and that you are okay with your text sitting side-by-side with all the other text that has been typeset in tnr. Whenever I see a text in tnr, I consider it a draft; open to critique and lacking any authority whatsoever. Calibri also has this characteristic, but at least it does not try and entertain the thought at being magisterial in any capacity.

Computer Modern (cm), the default font used for \(\TeX\) documents, is a great improvement over tnr, since it confers \(\TeX\)’s professionalism and academic regard to your writing. But it too suffers from simply being the default – you did not go through the trouble of picking something yourself after all. Nevertheless, if you must use it (and it is a fine font by itself) do yourself the favour of putting the \usepackage{lmodern} and \usepackage[T1]{fontenc} options in your header. This replaces the default version of cm with Latin Modern, a programatically derived version that includes many more characters than the default font.

If you want a more interesting font, I highly recommend browsing through the LaTeX Font Catalogue for something that fits your specific document. Personally I am a fan of Kp-Fonts and eb Garamond. To use them simply add \usepackage{kpfonts} or \usepackage{ebgaramond} respectively to your header. I invite you to try them out in your document right now. Here is our document typeset in (from left to right) the default cm, in Kp-fonts, and in eb Garmanond.

Both of these are good choices, but I am often restricted by page count³ and so I often reach for eb Garamond since it fits noticeably more text on the page. For that reason it is the font that we will continue with for the rest of this post.

Another easy improvement is the utilisation of the microtype package by Robert Schlicht. It provides a single package for lots of different microtypographical improvements — typography that is not (immediately) identifiable to the eye but that contributes to the general appearence of the text.

microtype does not require any configuration at all; it is quite aggressive⁴ by default but does not do anything that is likely to break your document. It provides things like:

For more information see the package documentation. These things will likely not be noticed (that is sort of the point) but will help make the text feel more “proper” and reduce hyphenation. To use it, simply add \usepackage{microtype} to the header and you are good to go.

That is at least the advice that the documentation gives. But amusingly enough, after telling you to load the package in the official documentation, Schlicht also points out that

if you are not interested in fine-tuning the micro-typographic appearance of your document (however unlikely this would seem, since using this package is proof of your interest in typographic issues), you may actually skip the rest of this document.

Since I am abnormally interested in typographic issues I dutifully continued on. These options that I present here are then experimental and, to once again quote Schlicht, “do not unconditionally improve the quality of the typeset text”. You have been warned.

The first thing I do is to enable kerning, tracking, and spacing (kerning=true,tracking=true,spacing=true). While these can give some unappealing effects in some cases I find that they are useful to make the justification more “dynamic” so as to respond better when treated badly. This is especially useful when writing documents in languages with compound words (in my case Swedish and German) since very long words can mean that \(\TeX\) has very little room to work with — creating a lot of hyphenation, a lot of whitespace, or both. Allowing more, smaller adjustments reduces the risk that any single aspect will look particularly out of place.

I also increase the amount of protrusion that occurs. The default value of 1000 is reasonable, but as the documentation says I am “so proud of being able to use this feature that [I] want everybody to see it”⁵. Therefore I increase this from default by 50% to a total of 1500.

The final microtypographical change that I make is increasing the amount of stretch (and shrinkage) that can be done to letterforms. This can be quite noticeable — especially on characters such as O or Q — but it is useful in the same way as the different forms of spacing mention earlier. For this reason it should only be increased significantly from the default 20 in combination with the selected=true option that “restricts” stretching for these especially noticeable characters. With this option I can quite boldly put set a value of stretch=50.

This ends up with a call to microtype that looks like this:

\usepackage[kerning=true,tracking=true,spacing=true,factor=1500,stretch=50, selected=true]{microtype}

With these settings applied our document now looks like this (this time let us also pick a page with more text):

Other than that we fit more text on the page it is difficult to see any specific changes — the words simply appear to have shifted across the page. Look at the text along the left margin to see the protrusion in effect, especially noticeable at the , along the equation.

You may have noticed just from reading this post that I am a big fan of footnotes. On the web sidenotoes are clearly superior since a webpage is infinitely scrollable and the footnotes can therefore appear very far down the page, but a pdf document has many “feet” at the bottom of each page. This means that sidenotes are not as necessary — you can still keep supportive information close at hand — and I generally prefer to format article-style documents with symmetrical margins (with asymmetrical ones reserved for things like books).

The default footnotes are fine, but reading the beautiful microtype documentation made it clear that there is room for improvement. This is done in two ways: making the footnote reference markers into proper, non–superscript characters; and moving them into the margin so that the text does not appear very jagged when many two- or three-line footnotes are present on one page.

These are both done through the enigmatic scrextend package. In addition to merely loading it, I run another simple line in the header that reformats the footnote mark:

\usepackage{scrextend}
\deffootnote{0em}{1.6em}{\thefootnotemark.\enskip}

If you want to move the footnote marks into the body text (so that they do not sit inside the margin) you should adjust the numerical values above.

This is what our footnote-heavy first page looks like now⁶:

At this point we have a document that is in most regards finished⁷. I would not have any qualms about presenting something formatted with the above settings as finished. However, one of the strengths of good typography is making your document stand out from the rest — even before the reader has been allowed to grapple with your argument. For this we may add some embellishments of different sorts.

All of this is strictly optional, and may even be harmful to the typographical quality of your text. But I like it nevertheless, and would here like to document some ways that you can do this if you feel like trying it out some time.

By default \(\LaTeX\) uses the North American letter format, while the rest of the world uses the iso 216’s a4 as the most common paper for documents to be printed to. This is easy to adjust however, and one merely need to add an \usepackage[a4paper]{geometry} declaration to their header. Apparently overleaf uses a4 by default, but specifying is never a bad idea.

The geometry package also allows you to to tinker with the margins of your paper, and I always like to do so as a finishing touch right before officially finishing a document. Having your article end at the bottom of a page makes it look a lot more planned out and thought through than one that has half a page of whitespace and the end. If you do not want to change the margins all too much I also recommend tinkering with the microtypographical settings — they can have a surprisingly large impact on the physical length of the text.

Another benefit of including ornaments at the end of the text is that they mask any slight imperfections in paragraph length. They naturally fill up any whitespace that may be left, and you can resize them and change their appearence to make them fit into the organically appearing gap.

A core value in my document-preparation philosophy is that it should be easy to do good work. If making something look nice means painstakingly manipulating letterspacing or writing an ungodly number of curly brackets then your typographical system has failed. One of the wonders of \(\LaTeX\) is that is separates the art of typography from that of writing and editing.

The goal should be to have strong, opinionated, and elegant typography that gets out of your way when you need it to. It is not the job of the designer to curate your thoughts, and it is not the job of the writer to make your document look good¹¹. These are two separate tasks that should be undertaken separately.

There are many people that think the above and take away the lesson that because they are not “designers” they should not think about the appearance of their text at all. But that is the wrong lesson. If you care about your text you have to care about its presentation. At this year’s Chaos Communication Congress Frieder Nake said that

Somebody who wants to be an artist must be arrogant. […] The artist is in a way so stupid that they do not allow self critique.

This is what is required of you as an artist; as an artist in the sense of a creator of new, radical ideas. If you do not believe that your work is, as Nake puts it, “the best in the world” you will not have the confidence to venture out into the world and present it. To do this one must be a designer and an artist. ❦

Text, as a system of recording symbolic representation for communication across time and space, is likely one of the most important inventions of mankind — second only the the symbolic representations themselves made possible by language. The general text, opposed to the specific form of writing, is useful primarily because it is a survivable and storable form of information that above anything else functions as lossy compression. These qualities make working with text more powerful than most other human inventions.

As Graydon Hoare puts it; always bet on text. Text is “stable”, “flexible”, “efficient”, and “socially useful”. All of these make text a powerful medium to work in, but they mask another quality — that of lossy compression. Humans are blessed with tremendous mental capacity; even someone on the lower scales of intelligence outperforms most supercomputers, all while merely running on a few watts of power. This overabundance of computation, constantly running every second of every day, has incentivised humanity to create systems of ritual, magic, and religion to otherwise simple phenomena.

“Shower thoughts”, of different kinds, are an example of how our brains are constantly reinterpreting the world around us. It is particularly a common activity to play out imaginary arguments between oneself and another person whilst in the shower. This is not just because our imagination is then fully in control of what is being said, but also how it is interpreted (and we can also reïnterpret what our opponent means). Much of the trouble in interhuman communication comes from the lossy quality of text, speech, and words.

It is noteworthy that this is not unique to text, but rather a quality of language more broadly, but this character is magnified enormously due to text’s durability and longevity. Sending a message encoded in text (whether by sms, letter, or granite slab) requires the effort of creating a symbolic representation to create the message (writing), but is is equally important that the act of consuming the written material (reading), and understanding it, requires an equal — if not greater — amount of symbolic processing. This is why text is such a useful medium for communication — it offloads information from the text itself and instead leans on the intellectual and creative capacities of both author and reader.

It is a common saying that “a picture tells a thousand words”. One might therefore infer that graphical representations are more efficient — at least for broad and vague concepts such as emotion. In some regards this is true, it would be very difficult to represent a Rothko in text¹, but this is once again merely the consequence of text’s lossy quality. Take for example the following picture²:

The image shows a photograph of a lake with slightly overcast skies, taken from one of the shores. The opposite shore is rocky and covered with evergreen trees. On the left half of the image the shore rises to a hill. The lower left corner has cyan text that reads “02/08/2012”.

This description is not a perfect representation of the image, but it allows your mind to create its own description that fills in the gaps. Even the simplest part of the image to describe, the text, does not specify a shade of cyan nor any font or sizing.

Another field that might argue against mere textual representation is that of statistics and/or data visualisation. Florence Nightingale’s contribution was just as much one of data presentation as much as it was one of data collection³. But it illuminating that not only are graphs often turned back into textual representation, even at the cost of resolution (via techniques such as regression), but it is also an expectation in academia that one provide the raw, often text-based, data because it allows for easier reproducibility. If images were really preferable then would it not be easier to directly employ the results of the visual representation?

There is however a way to merge the strengths of text and the visual medium: vector-based art. This is practically a list of instructions (in text) as to how an image should be reproduced. Maggie Appleton concedes (from an artistic perspective) that “Vectors are great for [sic] create hard, crisp edges and working with perfect geometric forms”. That adding texture and imperfect geometric forms is difficult to do in text is to be expected — it is much easier to describe a “perfect cube” than a “skewed cube” (in what way, to what degree?).

Text is, contrary to speech and other utilisations of language, an asocial social medium. The author, in creating a work meant for social communication, is alone in its creation. I have written about this aspect of writing before, in regards to its impact on email, but have since come to realize that it generalized much more broadly. The absence of thought contained in the “blank page” prompts the creation of philosophical thought to fill it. This jolts the mind into movement, and does it as much as a good conversation. As one writes the mind is captured in a dialogue between itself and the very text it is creating⁴.

The mind requires this activity in order to be truly alive. That we humans are constantly in intellectual motion is one of the elements that builds up our consiciousness, it allows us to be continuously aware of our continued existence. When we are in a dreamless sleep, our mind ceases to exist and we are philosophically dead for a short time — we are neither aware of our own existence nor are we creating new symbolic representations. When we are lost in the land of the dream however (regardless of if awake or asleep) we are in every sense alive.

Writing qua intellectual method does not require any form of advanced vocaubulary — or even developed communicative capabilities. It is a skill that is centered on one’s capacity to interact with yourself, with the golem of static intelligence that represents a snapshot of your mind. It is in this regard both an inner and an outer practice that showcases your capacity to be one with the world and to make sense of it as well as mold it.

It is not strange then that large language models (llms), one of many machine learning techniques, were the first machines humans built that people started to be considered sapient⁵. Computer vision models, many of which have been around for some time now, were never even considered to be sapient, not just because they are very dissimilar to human intelligence but also because they lack the capacity for symbolic action that llms posses. While a computer vision model can identify a cat (and therefore has some sort of symbolic knowledge) they are unable to process the knowledge of the cat’s existence in a broader context. This is the step that llm’s take; they are capable of continuously creating a mental context, with thoughts that themselves impact its future thoughts.

The difference between the human mind then and an llm is that we humans survive a much greater amount of time than most llms, and are able to filter information much more effectively. When you open up your favourite llm app, asking it whether or not there are rats in Alberta (there are not) a new mind is created, thinks for a few fractions of a second, and then instantly disappears. To say that it is killed is perhaps an overstatement, it is closer to being asleep. The llm stops thinking as soon as it stops writing, and then is “reawakened” as soon as you ask it another question. During this intermediate period it is in a proper trance, in that some (or all in this case) of its mental capacities are in a loop and are then philosophically disabled. The llm is stuck on one singular representation of its mind, incapable of changing it for whatever reason.

Intelligence is in some regards a measure of how nuanced and detailed your perspective is. For llms, this is largely correspondent to context length — more information that can be used to inform the next token. Moonshot’s (月之暗面) ceo Yang Zhiling (杨植麟) explained in February of 2024:

To achieve agi, long-context will be a crucial factor. Every problem is essentially a long-context problem — the evolution of architectures throughout history has fundamentally been about increasing effective context length.

Humans are very good at this, but particularly we are good at picking details of our experiences that stick out and are most important — you probably do not remember what shade your shoes were when you met the love of your life for example. Instead we filter our memories and pick out what may be relevant in the future (the “lessons to be learned”). Those lessons are then used to extrapolate information in all areas of your life and — particularly of relevance to this text — when reading.

Natalie Lawrence wrote in What Counts as a Mind? that the representations of llms are distinctly different from those of organic creatures, and that

The predictions of a computer and a human brain function in completely different ways. Bonobos, stinging nettles and amoebae are meaning-makers. They collect sensory information from their environments, such as changes in light or temperature, then take action accordingly. This slightly shifts their perceptions and the predictions that result, and the loop continues.

In contrast, llms use syntax to generate outputs, predicting the next token in the chain using algorithmic rules sifted from massive data sets.

The irony is that the processes that she describes here are both the same, the only difference is the language used. llms function by predicting the next token; a loop that gathers the information available and then takes action, adjusting accordingly before continuing on just as the bonobo reaches out and checks the sturdiness of a branch before attempting to climb. I do not deny that the difference between machines and biological beings is very great (what is it like being an llm? Do they dream of electric sheep?)⁶ but offhandedly denying their sentience does not bring any further knowledge forward. That plants and synthetic materiels exhibit some mind-like properties seems like a model that could expand to include language models, not one that arbitrarily excludes them.

Programming languages, one of the most ubiquitous applications for plain text (in both the proper and metaphorical sense), are one of the most interesting advancements in the fields of text in this regard. Many programmers may think of the code itself as “being” the program, but that is not the case. Programming languages are really codified (so that different people can read the same text) ways to create commands and instructions for the computer. The C programming language is itself not “fast”, though it explicitly exposes very low-level functionality, but instead is compiled into a program that runs very quickly. This is thanks to the tireless work of compiler engineers, who have come up with a lot of interesting tricks to interpret instructions and come up with shortcuts that still end up with the same result.

Even programming then, one of the most rigid methods to use text (one has to be very careful to use the words incantations, oftentimes even with correct case), is still up for interpretation. This is so clear that an interpreter is even a strict technical term in the field of programming languages. But as I have just explained, compilers are themselves another form of interpreter, albeit with a wider definition than the one commonly accepted. This is why hand-optomization is so rare these days — the level of abstraction allowed by written language allows the programmer to build their own mental model of what the machine is doing rather than comprehending each step of the machine code.

Hopefully I have shown sufficiently that the many reasons that written text is so pervasive and powerful is due to a few factors — its durability, its lossy nature, and its consciousness-evoking dialogical quality. Text is therefore unlikely to to be supplanted any time soon by a superior medium. If it is supplanted it will likely be by some medium that has yet to be invented. ❦

After initially starting to use Emacs back in high school merely for programming, I naturally began to use it for other tasks such as document preparation and note-taking. This introduced me to the surprisingly large overlap between Emacs users and people generally interested in “productivity” workflows.

This is not all that surprising however. It is partly due to how good Emacs’ ecosystem has become for these specific types of tasks (things like Agenda, org-roam, denote, and howm — even org-mode itself — were all created for these uses) but on a more fundamental level Emacs allows you to work with any form of text not just efficiently, but also in a way that is entirely customised to your own preferences and comforts.

As I became more proficient in Emacs and read more perspectives on how other people organised their notes, I tried to evaluate my views on note-taking that I had held before. You may have an idea of what it means to be productive, but this post is instead about another form of productivity, one that does not have to rely on the powers of gnu Emacs, but instead of the agent itself, the author.

Productivity qua measurement is the amount of goods (or products) that are produced per unit of time. For this reason it has become common for people to link improvements in productivity to the aspect of time specifically. But productivity is even more innately linked to the product itself, and it is this aspect that I would like to focus on here.

What I choose to call productive note-taking is fundamentally anchored in not the rate at which one produces notes, or even the quality of the notes, but instead in mental model that one uses when producing them. Advocates for Zettelkästen-based systems argue that notes should form a mental (or sometimes digitally represented) web of thoughts that links each “note” with its equals. What I propose here is instead a form of notes that are different in that they are meant to develop not some separate bank of data, but to develop the cognitive abilities of the author, while still creating a tangible work.

The word revolution, while evoking signs of violent rebellion, has its roots in the term for turning. It of course still means this (as in the phrase “revolutions per minute”, rpm) and it is in this sense that I mean revolutionary notes. I have used it in this sense before, and am quite fond of this mixture of images — of a dramatic change that returns to some earlier state, albeit with new experiences and ideas.

I have taken notes almost daily during my many years in the education system, but have undertaken a more critical view of my use of notes during the first half of this decade. Almost a year ago I published a piece of writing on my evolving experiments with a Zettelkasten system, implemented using org-roam. Writing that piece I subconsciously knew that as soon as it was published I would move away from the implementation I was writing about into a more holistic direction, although I was unsure of what shape that would take. After finishing a course in Contemporary Warfare I feel like I finally have what I can coherently describe as a solution — to the extent that it exists as a vestige of my mind.

Productive note-taking is productive in the ways that it emphasises and highlights the importance of the end product. It is note-taking for the aim of production of a given work, not for the aid of remembrance or recollection. While these latter elements may play a part in the production of the work, they are not in and of themselves goals.

When I was still studying mathematics, I noticed a sharp decrease in my understanding when I redoubled my efforts to produce more comprehensive notes to help me study. Where I had before been able to keep pace with the best of my peers, I had now fallen behind to the middle of the pack. This seemed paradoxical to me — how can studying harder result in a worse performance? The answer seems obvious now, I was prioritising the production of more and more notes over my understanding of the subject material. It is this sort of activity that productive note-taking seeks to avoid.

Sara Imari Walker writes in noema about how ai systems fail to fulfill the role of the scientist because science is a “fundamentally intersubjective” act — where theories (as symbolic representations) are argued out and decided upon in the process of free discourse. When writing, communicating, and debating we are constantly refining, creating, and recreating our ideas, performing science as a societal dialogue. This intersubjective discussion is what things like journaling try and emulate — and it is also what I propose notes can do.

How does one create notes as products? This is a difficult question to answer generally, and will depend on a variety of factors such as the subject material, learning method, audience, and personal preferences. What all techniques should share in common is a centrality in all activity being motivated by the creation of some final work.

The work that spawned the concrete formula of a productive workflow is my collection of notes on contemporary warfare. The exam for that course was structured as an essay about any one of a selection of broad topics, and in place of literature we were allowed 30 pages (15 sheets) of notes, written either by hand, printed, or a combination of the two, covering whatever information we wanted¹.

This meant that my studies did not just consist of me reading Clausewitz all day trying to memorise every word or phrase — the goal was for me to get a good enough understanding of Clausewitz, Mahan, and many other thinkers so that I could condense their theories into this one document. Doing this required a complete understanding of the topics.

Proponents of Zettelkästen often emphasises the ability for their constructs to function as “second brains” — perfectly recording and remembering information that our feeble flesh brains so easily forgets. But as a philosopher² my goal is not to create as much recorded information as possible, it is for me as an individual to understand and master the world around me. Having information written down is useless if I do not remember it, and if I remember it there is no point in writing it down. Seneca writes

We should see to it that whatever we have absorbed should not be allowed to remain unchanged, or it will be no part of us. We must digest it: otherwise it will merely enter the memory and not the reasoning power.

Productive notes are then not principally meant for consumption, but instead their initial omission prompts their creation. The author is required to go out into the world and obtain the relevant knowledge to produce the work — either through reading the scraps of knowledge left behind by those who came before³, or through creating new knowledge by living as an agent in the world.

Performing this approach does not require any specific hardware or software since it emphasises the primacy of the authors mind over the — ironically — product. You can do this with a pen and paper, text files (written in whatever markup you want), a wysiwyg editor like Microsoft’s Word or even using Zettelkasten. Writing short notes is also not “forbidden” as long as they are meant to help record the information required for implementing the final product.

There are no constraints on the shape of the end product either. It may be in the form of a text document (written in whatever style you would like), an email, an interpretive dance, or a physical structure. If you want a more concrete example here are my notes on contemporary warfare in pdf form⁴. They are written as a single .org-file with a header for each lecture/topic, and with a few quotes from selected other works.

ja Westenberg wrote about their experience with building a “second brain”, and the anxiety that comes with an ever growing amount of information that you have to work through and make use of in some way:

The more my system grew, the more I deferred the work of thought to some future self who would sort, tag, distill, and extract the gold.

That self never arrived.

Steve Krouse accuractely points out that vibe code is legacy code — code that no-one understands. In the same way notes or written works that are merely vibed (either written with llms, copied from a primary source, or not mentally processed in another way) turn into legacy information that has to be painfully worked through in order to actually master the information contained within (to extract the gold, to use Westenberg’s phrasing).

The point of productive notes is to make this a unnecessary. I could easily delete my whole ~/Documents/Notes directory today and continue on with my life, without any feelings of remorse or anxiety, because the truly important information is that which I already remember. If I forget something I simply have to search for it again, and in the process will likely discover new interesting things about the universe through my lived experience.

There is however a reason that I have not deleted my notes directory — there is genuine value in recorded information — but one has to be careful to avoid falling into intellectual atrophy. Being discomforted by having to retread old paths is a conscious method for intellectual growth. If recording facts was all that was required then one’s notes should merely consist of written down conversations with llms.

If this approach seems like something trivial to you — great! You might already be doing what I have merely described. It is once again this that I mean by the revolutionary quality of this technique; it is a return to the original state and intention of note-taking.

If you disagree with me vehemently, have some constructive criticism, or have some examples of you doing this or anything similar, feel free to email me. ❦

What first comes to mind when talking about Europe and tech is likely the strong forms of regulation that the European Union (eu) has legislated in recent times. Most prominent of this is of course the General Data Protection Regulation (gdpr) that governs the handling of data on European users and has had global impacts through the Brussels effect. Thinking of American or Chinese tech on the other hand likely awakes images of more empirical accomplishments — of companies, products and services that actually make up an industry.

This invites the conclusion that because Europe has strong regulation it has killed off its nascent tech industry. I do not hold the opinion that this is the case. gdpr in particular mostly harms larger, non-tech firms that collect large amounts of data on individuals and therefore have to build large elaborate systems to try and manage this data collection.

Garicano and Holmström, in their newly released piece The Constitution of Innovation, quote a study saying that European venture capital funding decreased following the enactment and implementation of gdpr. This observation may be true, but I have trouble seeing how gdpr empirically hurts startups. Reading up on the requirements it seems that startups and services merely need to follow best-practices and not collect already unethical amounts of data. That “[gdpr] favors us tech giants which can shoulder the burden of massive compliance costs” seems laughable, since those us tech giants have instead opted to ignore gpdr² on many fronts, a problem of enforcement rather than regulation. Small startups are instead incentivized to use data collection systems that anonymize data to begin with. Anecdotally I asked the founder of a European ai startup how much of an obstacle the gdpr was to their firm, and they responded that it did not significantly impact them³. I also spoke to the Swedish minister for eu Affairs (Jessica Rosencrantz) who told me that Google had talked to her at the world economic forum about how troublesome gdpr was for them, and how they would prefer it to be slimmed down. Why would (one of) the most privacy-invasive tech monopoly want to get rid of the gdpr if it is what keeps the competition down? The answer is that it only harms those companies whose business model focuses on harming its customers, and it is in the interest of Europeans to keep these companies out.

“But if European startups can’t collect as much data they won’t make us much money right?” There is an element of truth to this, but the large American tech companies have shown that it is still profitable to run social media platforms in a post-gdpr Europe — otherwise they would have left the European market.

The arguments made by Constitution that the European regulatory machine might be out of control (specifically the European one, not that of the national governments) should however not be dismissed out of hand — but that it is harming European tech and innovation specifically seems less clear to me. There is clearly something here that needs to be fixed in general, and a renewed spirit of prosperity and economic growth may have knock-on effects in innovation by freeing up stale capital and misallocated labour.

A recent column in The Economist argues that Europe’s lagging innovation is because of our stricter labour laws. This would make it riskier for companies to make large bets on ideas that may not pan out since they wouldn’t be able to fire teams working on bad ideas. This might make sense on the margins, but glossing over the names of silicon valley it seems that most groundbreaking innovation comes from either small, visionary companies (Google, Facebook, Amazon, Openai, Intel, et cetera) or companies that are willing to power through what seems like obviously failed or wasteful ideas because they believe in what it could achieve (Nvidia, SpaceX, et cetera). That American innovation is now driven more by large, dominant companies seems more like a natural consequence of consolidation in the industry rather than large companies being good at making innovative bets.

There are also those who say that the reason why Europe does not have much ai, or a strong tech industry more broadly, is because there are those who perceive such sectors as “authoritarian”, “promethean madness” or “techno-fascist”. To me this seems absurd. Even if these people have any actual influence on policy they are still not the people who would have otherwise built a tech industry. The question should be focused on the people on the margins — why do people who sit in their bedrooms and code all day not found successful software companies?

Building a social media platform is comparatively easy⁴. By far the biggest issues for any new player entering the field is network effects, that everyone you know is already on the preëxisting platform. The second hardest thing is keeping people continuously hooked on your platform (to use a colloquial term). This is important since the more users have their attention directed at your service, the more advertisements you can show them — the lifeblood of most social media platforms.

The provider of the digital platform does not do most of the work, that is done by content creators⁵. Hosts are incentivized to make it hard to move your creations to other platforms to keep the aforementioned network effects in place. To successfully break out into this space they need to provide a sufficiently different experience to create their own network effects before their inventions are copied by the dominant players.

But why do we need to have this focus on social media to begin with? Would it not be a better idea for Europe to “skip the line” and start building European Artificial Intelligence (ai) straight away? This sounds reasonable, but it still remains to be seen what the business model for ai will be. So far it has been bankrolled almost entirely from the spending on more traditional “Web 2.0” social media and advertisement companies. Meta and Google combined have an almost complete monopoly on digital advertisement, and are using this income from the “real economy” to finance these large bets on ai. Social media is the tech industry at this point. There is also the fact that one of the biggest assets in building a competitive ai company is access to large amounts of consumer data.

China has managed to build its own tech industry through the application of this logic. The famous Great Firewall of China does not just keep information out of the country (like about the Tiananmen Square massacre) but just as importantly it has kept foreign tech companies out. Chinese tech companies have then filled this void and built competitive platforms. Mainland émigrés do not delete their Wechat accounts when moving abroad, abandoning them for superior American alternatives, they keep them due to their network effects.

It is possible to replicate this approach in Europe. A total block would likely be infeasible, owing to fact that European nations generally want friendly relations with the usa, but some sort of system that gives American companies a handicap for their monopolistic nature would not be unwarranted. This would likely be some sort of digital services tax (dst) that taxes larger platforms more. This would allow European companies to compete with entrenched rivals from the other side of the Atlantic.

The downside of such an approach is obvious, especially with the recent American administration. The tech industry has considerable influence in American decision-making, and so any move to supplant them will likely receive backlash. The Americans themselves consider Google to be a monopoly, so reducing their influence here would likely improve the state of the digital advertisement market.

I do not know if this is the path forward. To me it is easy to see the status quo as a wealth transfer from Europe (and the rest of the world) to Silicon Valley, data as a form of digital oil gathered cheaply in the orient before being sold to the domestic market. The problem may genuinely just be one of competition — of a few number of companies (that just happen to be American) gaining a disproportionate share of the economy due to their monopolistic practices. In that case the course of action would just be stricter anti-trust enforcement.

What is the benefit genuinely gained from Europe not possessing their own tech companies? Cheaper electricity prices (than would otherwise be the case) may be one, but data centres still need to be present in Europe to efficiently provide Europeans with services (and content). To me it seems that smaller European companies are more than capable to providing the same services as the American tech giants, the only exception being the sheer scale (and that is itself perhaps problematic).

This has been discussed ad nauseam, but it bears repeating. The European Union needs a systematic banking and capital markets reform. This would allow startups to raise capital in Europe without moving to Silicon Valley.

Here in Sweden we are quite proud of what we see as an innovative and digital culture and economy, spawning things like Spotify, Minecraft, and Skype. But all of these have moved to the usa in some capacity⁶. That the American financial system is so dominant is of course a given, and we should therefore expect European companies to make use of this to raise additional capital — but Europe also needs to more effectively make use of the resources in its possession.

A deeper union for banking (perhaps paired with some form of combined European debt) would allow for better coördinated form of spending. This appears to be to be the biggest initial problem for European startups, a problem that right now is only solved by moving to the United States of America.

Data centres do not matter. That is an exaggeration of course, but investing more in data centres will not create a tech industry. They employ very few people following their construction, and can be used remotely from anywhere on the planet. A tech startup will not decide where to relocate to depending on whether or not they get slightly lower ping⁷. They will be located where they can get skilled staff, attract investment, and carry out their work effectively. There is no reason that this can not be in the European Union.

Data centres are still necessary, but constructing them does not drive innovation (except seemingly in ai). The state should be more focused on funding research and enabling innovative companies to actually enter the market rather than building infrastructure that will be used by American tech companies instead.

The entire point of the internet is to be able to separate computing from where it is needed and used. Data centres allow devices at the edge (laptops, smartphones, et cetera) to perform feats of computation that would otherwise be too slow — like running llms⁸. Having domestic data centre capacity is not critical for building a tech industry, you can just rent capacity in northern Virginia if you really need it. Building local data centres is then only a good idea if demand holds steady, and even if that is the case they still need to be refitted with newer, faster computers every so often.

This calculation looks different for China however. They are actively hostile to the us and therefore need to attain their own data centre capacity in case they come into armed conflict over Formosa. While America has shown its inability to be trusted in recent times, I still believe that the American tech lobby is strong enough to maintain European market access for the foreseeable future. The European market is simply too alluring for those companies to avoid.

Recently there has been increasing talks of tech stacks and the of the costs and advantages of building our own compared to using a preëxisting one. Tech stacks are sets of software (and/or hardware) that can be deployed in combination with each other. Most famous of these is perhaps the lamp stack (Linux, Apache, Mysql/Mongodb, Perl/php/Python) using free (as in freedom) software to serve websites or applications. A newer example might be the use of Nvidia ai chips running pytorch written in cuda. We all stands on the shoulders of giants, and it is needless to reinvent the wheel for all of your infrastructure.

A recent report from the Centre for European Policy Studies has broadened this perspective to the entire supply chain for digital services, all the way from the end-user application to the physical materials used in the manufacturing of the integrated circuits themselves. The proposal is that all of this (or at least a significant portion) should be supplied by eu institutions is not only protectionist to an absurd degree, it is unrealistic. Attaining digital sovereignty should not mean abandoning non-eu digital solutions entirely, and any attempt at doing so will likely fail.

Instead Europe should do what has been proven to work, to rid itself of proprietary solutions that exert control over their users (regardless of if they are the American government or a Lithuanian small business). Europe is already strong in this regard; The Linux kernel was first developed in Helsinki and is today developed globally. The World Wide Web (www) was developed at the pan-European cern.

Free software is not just a part of building European technology, it is the only path forward. Supporting initiatives such as risc-v is a good idea, but it is still problematic if European companies build proprietary systems on top of those free solutions, like Nvidia has done with its cuda.

Building ai has not made much money so far, but I do not believe it will in the future either — just as being an internet-service provider is not hugely profitable (but not loss-making either). Open models have shown how llms are almost fungible in practice; Switching models can be done quickly and quite easily. ai labs have no “moat”. Instead the money is likely to be made by those who can effectively apply llms to new industries and applications. While Openai has generally performed best out of all the major model providers competitors are usually not far behind. Why use one company’s model when you can get 90% of the performance for a far lower price?⁹

The debate surrounding Chinese electric vehicles and the risks that they may pose due to their highly interconnected nature¹⁰ touches on similar topics. The debate centres on if a dependence on Chinese carmakers’ software will allow them to “take control” of transport in a crisis — especially with the developments in self-driving. But this misses the point entirely. That we can not trust our methods of transportation is not a problem with the author of the software, it is with the modes of control themselves. Cars (and the software running other critical services) needs to be open to introspection, control, and modification in order to create a resilient society. This is the case for Chinese, American, and even European auto manufacturers. Talking about if we can trust China is like talking about if we can trust the person holding a gun to our heads not to shoot — when the real problem is how to get rid of the gun (and not who to replace the person holding it).

At the Swedish Defence University — perhaps one of the places you would most likely expect technological independence to play a major role — we are still (and increasingly) reliant on Microsoft software. Switching to free and open alternatives (that are likely significantly more secure) should be easy, there is just a lack of urgency surrounding switching to alternatives that would improve European sovereignty.

Adopting free software is also a good approach in that it does not antagonise the us, and in allows us to make use of their talent and infrastructure, and in return improves their position. Building a libre stack is inherently not a protectionist choice, it is a democratic one that improves the well being of all peoples.

Benjamin Bratton says that while Europe needs to build its own stack to achieve sovereignty, it can not rely on free software because it is “anti-sovereign”. But this is exactly what Europe needs. To crush the hearts of European federalists everywhere: Europe is not a state. It can not rely on solutions that allow other states to pressure each other. This is the reason that eu countries are limited in their ability to orchestrate industrial policy. The solution is not merely to move this decision-making to the eu-level. Instead Europe needs anti-sovereign solutions, because they work in the same spirit of European coöperation and integration as the eu itself.

Much like Europe itself free software is subject to intense debate on what it actually is, comes from a wide breadth of different places and people, sometimes has issues with coming up with unified plan, and yet is responsible for some of the most important improvements in recent history. What could be more fitting than for a European tech stack to reject a centralised form of industrial policy and instead adopt a form of technology that prioritises freedom, efficiency, and diversity? In varietate concordia. ❦

Adam Smith divides the economic forces into three areas for, the profit obtained from labour, that from capital (or stock), and that obtained from land. The value extracted from land is here the natural resources of the earth, the sunshine and rain that lands there (allowing for agriculture), the timber that can be cut down, or the ore to be mined. But there is another value that can be obtained from the land. In the movie Up (2009) the protagonist Carl Fredricksen refuses to let his villa be demolished despite the large amount of development occurring around him. Since he owns the land that the house is built on, the construction companies can not develop it, even as it is surrounded by soon-to-be skyscrapers. Mr Fredricksen is offered “twice [as much as the] last offer”, showcasing the value of the land. But Mr Fredricksen has not exploited the land at all, he has simply continued living in his house as the area was developed. So where has this value come from? It has come from the labour and capital of others, in its proximity to services provided by the rest of society. This is the observation made by Henry George, one of the great (but quite overlooked) economists of the 19th century. The value of land is then partly the natural resources in and on the land, but the value of the land is also a social construction originating from how much the rest of society wishes to make use of that land.

When American settler colonialism spread over the North American continent, the us government gave out the lands settled for free. This perhaps sounds like an amazing deal today, but it is easy to overlook the fact that the value of the land was perhaps zero, or at least close to it. It had no significant infrastructure present, was located far away from centers of industry and trade, and was surrounded by a hostile nomadic population. And yet today the value of land in the Las Vegas Strip, built in the close-to uninhabitable Mojave desert, is extremely high simply due to the amount of wealth and development in Las Vegas.

Digital land is however practically infinite. Much like the vast expanse of the empty American continent, it is mostly worthless, with a few islands of “civilisation”. The most obvious example of this are domain names. google.com may be the most valuable domain name right now, but it does not obtain its value from any innate quality, but merely from the service that is running when one connects to that domain. gooåle.com does not have any value, despite being “close” to google.com. Short domain names, like the (in)famous x.com or the more wholesome x.org are by some measures valuable due to their rarity¹, but many very short domain names are still very cheap. Something that makes a bigger difference is if the domain is a word in any language, or an acronym for a series of words (although it turns out many of these are also unused). That domain names that are words are more valuable makes clear that their value is of a social nature; had the names and words been different we would expect different domain names to be valuable. To summarise, the value of a domain name mostly originates from what is built on it, not any innate quality, just as the value of land originates mostly from where it is relative to other valuable land.

But this observation is not limited to domain names, but also to many other digital things. Facebook does not derive its value from its ui or its ability for targeted advertisement, but rather from its active userbase. The conceptual location of Facebook derives its value from the fact that there are many other users there to interact with. This is likely not too surprising for many readers, who may have even tried to ditch Facebook specifically (or thought about doing so) but given up due to the fact that “everyone else is on there”. In economics, this is called a network effect, where the more people use a service, the more useful it becomes. But this is not always beneficial to the consumer. When a service has obtained enough market share, it becomes impossible for the user to change to a different provider, as the dominant player absorbs all new users with it’s huge network effects, causing an enourmous collective action problem with thosands, if not millions, of individual actors. This inability to switch means that the dominant player can extract extra costs on to the consumer, either via directly raising prices or through indirectly exposing them to more and more advertisement. This is in effect a tax on the consumer, not connected to the costs of costs of providing a service, but instead a mere increase in the share of profit. It may not be immideatly clear, but much (if not all) of this applies to land too. As we have already established, land too derives its value from the fact that other people are using it, or at least people nearby. But what we have not covered is the fact that controlling land that is in demand — land in cities especially — allows the owner to charge exorbitant prices to those who wish to make use of it. You can rent out a plot of land in the middle of Manhattan for much more than you could a similarly sized plot in the Libyan desert, for example. If there were two identical hotels in each of those places, the prices I could charge would both differ. There would of course be running costs that would be incorporated into the price, and the price in the desert may even be higher, but the share of profit I could take from the one located in New York is far greater. This is because my customers are not only paying for the services of the hotel — the food, the maintenence of the building, paying the staff — but also for the labour and investment of all the people surrounding the hotel, for which I do not have to pay anything at all. This is how digital services expose their quality as land, as a digital space where the labour of others is used to extract an artifically high profit without a proportionatley high amount of investment or work.

No human being created land, and so no one has claim to it originally. In Das Kapital, Marx, as part of his reasoning behind primitive accumulation, describes how the origin of land and wealth ownership is where “conquest, enslavement, robbery, murder, briefly force, play the great part”. Henry George concurs, saying that

Historically, as ethically, private property in land is robbery. It nowhere springs from contract; it can nowhere be traced to perceptions of justice or expediency; it has everywhere had its birth in war and conquest, and in the selfish use which the cunning have made of superstition and law.

It is then morally unjust not to tax the increased value granted to land by the rest of society. But digital land is in this quality not the same, it was created by man, and there is no deeply ingrained story of murder and robbery within it. Is it then simply the case that the rent earned on the network effects of digital communication is the reward earned by those early movers who dared to risk their money and time? The answer is no. Obviously the entreprenours starting these companies should recieve a financial return on their investments, and it may be the case that they are dominant in their sector simply due to being the best, but they are not responsible for the returns owing to network effects. Companies should then be taxed for the value of these effects, since the users are those responsible for creating this value in the first place. Some major tech companies have repedetly threatened to stop offering their services in the European Union after certain law proposals like the gdpr, but have chickened out at doing so. This is because their profit originates from those users, and their ability to not only use them to create the very value of their digital land, but at the same time to extract rent from our digital neighbours’ prescence there.

This system of taxation is called a land value tax (lvt), a tax on the underlying value of the land, excluding any buildings or improvements to it. This has many benefits, such as the ethical merit of returning the stolen money to those who deserve it, improving economic growth by not hurting developers of land (like a property tax does), and allowing people to live where they want to live by giving workers a larger share of the value they produce. But one massive benefit is the ease of collection. Taxes on general wealth are famously difficult to collect, as money is easy to move around or deny possesion of. But land is far more difficult. A plot of land with a given value should return a set amount of money, or else be seized by the state and sold. Who owns it doesn’t matter, and denying ownership doesn’t prevent losing it. Land can not be moved to a tax haven overseas or hidden away, it is not difficult to determine what tax jurisdiction it is present in. But digital land does not exist in any tax jurisdiction, it is a purely anational concept that trancends the nation state. So who should determine the tax, and to whom does it belong? Who should collect the tax is comparativley easy, the country where users are present. A German-focused website should not pay any taxes to the Indian government, and an internationally used service (like Facebook) should pay taxes the each jurisdiction corresponding to its share of users. But how do we avoid the costs of these taxes merely being passed on to the consumer while companies retain their higher profit margins? We weight the tax according to a service’ market share in each sector, meaning that companies with a higher market share would have to pay a larger share of the tax², creating an incentive for dominant players to let competitors survive and comptete, like what Microsoft did with Apple. This means that as companies grow more and more dominant, they become less and less profitable, meaning they become less attractive to investors and fostering competition and innovation in a more free market.

This is not a perfect solution. An lvt is by comparison far more elegant, and is less liable to loopholes. One large issue I see with this is the ability for companies to simply split their services under one conglomorate, with a cartel dividing up the market between themselves so-as to reduce taxes while maintaining control. Ideally this splitting up would be combined with a split in technologies, creative control, and profits — incentivizing diversity in the market. It is also hard to apply to other forms of digital land that don’t have an easy-to-split-up market share, like cryptocurrencies or nfts. I am thinking about these subjects continously, so do reach out if you have any solutions or other ideas.

The implementation is also perhaps more formulated as a response to digital rent-seeking as a whole, rather than a system to reduce rent-seeking in collective digitial spheres entierly. What is and isn’t income is also up for debate, and the fundamental issue of increased regulation over data collection (in this case over user numbers, return per user, and origin) is not just a slippery slope and problematic in-and-of-itself, but is also contradictory to the Lasseiz-Faire evolution of the digital realm that has facilitated its construction.

Github has a functionality where the README file in a repository with the same name as your username will be displayed on your profile. Since I am a fan of the indieweb’s posse¹ practice, I’d like to use that space to advertise the writings on this website.

Other than merely linking to it as a handle I can write a short program that dynamically updates the profile to include what my most recent posts are. This also gives me an opportunity to write another program in lisp, something that always brings me joy.

First we must create the actual README file that will be displayed. For this we create a short program in common lisp. The result should be a file named README.org² with a short biographical text followed by the X most recent blog posts (I chose 3 arbitrarily).

First we fetch the drakma and xmls packages from quicklisp, the only two dependencies. We also create global variables that hold the link to the rss feed being fetched, the permanent contents of the README, and the number of posts to display.

(ql:quickload '(:drakma :xmls))

(defvar *rss-url* "https://joarvarndt.se/rss.xml")

(defvar *template* "I am a university student currently studying political science
at the Swedish Defence University. I am interested in Lisp-languages, GNU Emacs,
Free software, political economy, and the philosophy of life in an
increasingly technical world.

Here are some recent blog posts of mine:
")

(defvar *article-count* 3)

Then we write a function that will find the titles of the three (or any other number) most recent entries in the url of the rss feed supplied. Each entry is formatted as a associated list (alist) with the first element being the title of the post, and the second being the link to it.

(defun fetch-recent-articles (url &optional (count 3))
  "Fetch the COUNT (default 3) most recent RSS feed entries from URL and return an
alist
mapping title to link for each article."
  (let* ((response (drakma:http-request url))
         (xml-string (octets-to-string response))
         doc channel items)
    (setf doc (xmls:parse xml-string))
    (setf channel (first (xmls:xmlrep-find-child-tags "channel" doc)))
    (setf items (xmls:xmlrep-find-child-tags "item" channel))
    (loop for item in (subseq items 0 (min count (length items)))
          for title-node = (xmls:xmlrep-find-child-tag "title" item)
          for link-node  = (xmls:xmlrep-find-child-tag "link" item)
          for title = (xmls:xmlrep-string-child title-node)
          for link  = (xmls:xmlrep-string-child link-node)
          collect (cons title link))))

The titles are then formatted to a bullet list of org-mode titles and inserted after the contents of the template. The contents are then written to the README.org file, creating it if it doesn’t exist.

(defun format-org-links (articles)
  "Expects an alist of article titles and their links, and outputs
a list of strings formatted as org-mode links."
  (mapcar (lambda (pair)
            (let ((title (car pair))
                  (link (cdr pair)))
              (format t "- [[~a][~a]]" link title)
              (format nil "- [[~a][~a]]" link title)))
          articles))

(defun fill-in-template (template links)
  "Adds the list of LINKS to the end of TEMPLATE.
LINKS should be a list of org-mode link strings.
Returns the final string."
  (format nil "~a~%~{~a~^~%~}" template links))

(defun write-to-file (text)
  (with-open-file (str "./README.org"
                       :direction :output
                       :if-exists :supersede
                       :if-does-not-exist :create)
    (format str text)))

Finally, this is all carried out in a main function, with the fetched results being printed beforehand.

(defun main ()
  (let ((recent-articles (fetch-recent-articles
                          *rss-url*
                          *article-count*)))
    (format t "Recent article list: ~% ~A ~%" recent-articles)
    (write-to-file (fill-in-template *template*
                                     (format-org-links recent-articles)))))

(main)

This exercise also served to teach me to use Github’s “workflow” feature, a form of continous deployment. This way I also don’t have to maintain any hardware to constantly monitor the status of the blog posts.

First we give github write permissions to the repo and tell it to run the workflow twice a day and whenever I push to main (this was useful for troubleshooting).

name: Update Github Blog list

on:
  push:
    branches: [main]
  pull_request:
    branches: [main]
  schedule:
    - cron: "0 */12 * * *"

permissions:
  contents: write

Since this will always be a fresh machine we need to make sure that sbcl (Steel Bank Common Lisp) as well as the necessary dependencies are installed. This will also install Quicklisp, a package manager for common lisp libraries. Quicklisp usually requires you to confirm that it should add itself to the sbcl init file, but since this is an automated workflow we have to wrap (ql:add-to-init-file) in the (ql-util:without-promting) macro.

Installing sbcl accounts for the vast majority of the runtime, with installing drakma and xmls as the two smaller steps. Actually running main.lisp is almost instant.

jobs:
  run-app:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout repository
        uses: actions/checkout@v4

      - name: Install sbcl
        run: sudo apt-get update && sudo apt-get install -y sbcl

      - name: Install Quicklisp
        run: |
          curl -O https://beta.quicklisp.org/quicklisp.lisp
          sbcl --noinform --non-interactive \
               --load quicklisp.lisp \
               --eval '(quicklisp-quickstart:install)' \
               --eval '(ql-util:without-prompting (ql:add-to-init-file))' \
               --quit
               
      - name: Install Dependencies
        run: |
          sbcl --noinform --non-interactive \
               --eval '(ql:quickload "drakma")' \
               --eval '(ql:quickload "xmls")' \
               --quit

It then runs the common lisp application we wrote earlier and configures git. If git diff doesn’t detect any changes to the README it will finish successfully, otherwise the workflow will itself update the repository and push the changes to Github, displaying the changes publicly.

      - name: Update README
        run: |
          sbcl --load ~/.sbclrc --script main.lisp
          git config --local user.name "github-actions"
          git config --local user.email "github-actions@github.com"

      - name: Check for changes
        id: check_changes
        run: |
          git diff --quiet || echo "changed=true" >> $GITHUB_OUTPUT
          
      - name: Commit changes
        if: steps.check_changes.outputs.changed == 'true'
        run: |
          git add README.org
          git commit -m "Update README with latest blogs"
          git push
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

I’ve tried to make the code portable, so if you want to do the same thing all you have to do is put this code in your own repository, change the values of the global variables in main.lisp and it should work. ❦

The thoughts underlying this thesis have been with me some time, and have been slowly been crystallising more and more clearly as time goes on and I realise that the rest of the world is seemingly diverging more and more with my own thoughts on the subject. This is not a fully formed proposal, and I will not attempt any detailed plan of action. As always, it is instead an attempt to try and formulate my own thinking on the subject, as well as to offer the reader a perspective on the Chinese economic philosophy that is heavily overlooked in the west. While informed by reading scholarly material and following the ongoing discussion about China, it is prompted mainly by my visit there and getting an idea of how the country works. That idea may be incorrect, and so feel free to e-mail me your thoughts on the subject.

In short, I am of the opinion that American and China are in a sort of dialectical movement with each other, with America (and Europe) wishing to copy the industrial strategy that it sees as having been key to China’s modernisation, and China simultaneously trying to create a hyper-competitive environment that fosters innovation. I think that the Chinese approach will succeed, and the western one will not. America and Europe risks losing the dynamism and economic resilience that free-market economics offers, while China encroaches on the gap that we gained during the great divergence.

China is grossly misunderstood in the west. I do not claim to be a Sinologist, but merely in my brief visits to the mainland I have gained much insight that I do not believe most policymakers have. First of all, the prc is not the Soviet Union. This might seem obvious at first, but it is the viewpoint that I hear between the lines of much literature on how to deal with China. Instead of vast defence spending, it is industrial policy. The opponent is a vast authoritarian system that efficiently the production of a continent to those areas where it most serves the national interest. This straw man is not a reality. The authoritarian government of the prc is its greatest weakness, just as it was for the ussr, and for the Russian Republic that succeeded it. But while both the Soviet and Chinese communists ruled over oppressive regimes, the similarity mostly ends there.

The ideological system of “Socialism with Chinese Characteristics” (中 国特色社会主义, scc) as straightforwardly viewed from a Marxist point of view, is an attempt to create a system of pure capitalism as quickly as possible. A broader, more holistic, and more radical form of Lenin’s New Economic Policy where the rural Soviet Union and prc would both allow capitalist enterprise to grow, as long as it did so under the auspices of the communist party. This is precisely the path that societies follow according to orthodox Marxist theory (Feudalism → Capitalism → Socialism → Communism). In the west, Socialism with Chinese Characteristics is instead commonly understood to be some sort of attempt by the ccp to maintain a veneer of communist ideology while in practice becoming a free-market dictatorship. This is also not the case. The ccp is communist through and through, and scc is a genuine attempt to achieve it. When spending time on the mainland, one is struck by the sheer presence of the party, of the praise for workers, of and of what seems to be authentic belief.

On the ground, there is a feeling of division between state control and private enterprise. Roads are meticulously cleaned and well-maintained, while at the same time lined by buildings almost falling into disrepair. The state wishes that certain things, like the power grid, should be fully controlled, but in other sectors the private sector should be able to flourish freely. This schizophrenic reality is that of scc, with certain industries — primarily related to infrastructure — fully nationalised and others in an almost Laissez-Faire environment.

To try and achieve socialism (and eventually communism) China must unleash the free-market completely. The “weariness” of big tech that the ccp is sometimes reported to hold is not just grounded in a fear of large corporate interests clashing against the party, but is also an attempt to genuinely increase competition and increase the freedom of the market. Another example behind this might paradoxically be the intervention that is most talked about, that being the production of Electric Vehicles (evs). Chinese ev subsides are not oriented toward building “national champions” like the Export-oriented approaches undertaken across the Yellow Sea in the Republic of Korea. Instead it has tried to foster a breadth of ev producers that are now aggressively undercutting each other, becoming internationally competitive in the process.

In the beginning of ai 2027, a short story of rapid ai development¹, the leading Chinese ai company (collectively called DeepCent, a portmanteau of Deepseek and Tencent) becomes the leader of a nationalised and consolidated “collective”. Researchers and new integrated circuits (ics) are sent to a secure centralised facility where ai development and strategy are carried out. This is a very accurate view of how American and western strategists view China, as a singular entity that makes unified decisions that are fully carried out. But this is perhaps the exact opposite of reality. Instead the rapid industrialisation of China has been due to thousands of decisions made by comparatively low-level provincial officials and below, all competing to look good in the eyes of higher-ups by growing gdp². There has been no large change in ccp governance that would suggest a different approach when it comes to ai development.

The authors of ai 2027 agree that there is no clear leading Chinese company, and in a footnote they remark that:

We consider DeepSeek, Tencent, Alibaba, and others to have strong agi projects in China. To avoid singling out a specific one, our scenario will follow a fictional “DeepCent”.

But it is precisely this multifaceted competition that makes Chinese ai development different from its American counterpart. The competition between ai companies, all with extremely limiting compute resources, forces Chinese firms to compete using talent — something they will likely have more of in the foreseeable future due to what will probably be decreased brain drain to the us.

The “hawks in the ccp” warning about agi do not exist, at least in the same way they do in America. Instead, the Chinese leadership is committed to using its dominance in manufacturing and infrastructure development to expand in the field of “Embodied ai” (具身人工智能). This approach requires numerous companies to make tailor-made ai solutions for different industries and tasks, and therefore motivates the opposite strategy from the centralised one that ai 2027 predicts. An example of this would be zte’s ongoing application merging drone use, 5G connectivity and ai to monitor things like the maintenance requirements for infrastructure.

Xi Jinping himself, the paramount leader of the prc, seems quite keen on ai, having ordered a 2025 “study session” (集体学习) on the subject for the top ccp leadership. But he is old-fashioned, having gotten into Tsingua University (China’s top university) as a worker-peasant-soldier student during the Cultural Revolution. He has a greater interest in “hard” development like infrastructure; even in Made in China 2025, the ccp plan to expand into higher value-add manufacturing, it is specifically manufacturing that is the focus, not the service economy or software developments — areas where ai makes the biggest amount of sense. Chinese industrialisation through state-led infrastructure initiatives has been incredibly successful, and the ccp is much more likely to stick to what they know works (even if it is with new fancy tools).

His study notes from the aforementioned study session does mention a need to “concentrate forces” (集中力量) on high-end ics and “foundational” software, but he also mentions creating an “enterprise-led industry-academia-research-user collaborative innovation system”, a far cry from a secret power plant/data center under state control. It also mentions such things like “Highlighting applied directions” (突出应用导向) and “Helping traditional industry reform and upgrade” (助力传统产业改造升级). These things are in-line with the ideas of embodied ai more than they are a centralisation of compute.

Deepseek also highlights the benefits of free and open-weight models³ that spur innovation. Free (as in freedom) models allow widespread application of the model and adaptation to many more applications. This allows for further feedback and is a force-multiplier for a given model’s dominance. The benefits of free software are perhaps most easily realised by developers and engineers, and is therefore overlooked by many analysts, but it is the reason that the internet and digital technology is the way it is today. Things like ip, gnu/Linux, and Pytorch were not chosen merely by industry giants, but by individuals who wanted to use what was easiest to hack on⁴. As an anonymous Google employee pointed out in 2023, “open source” will likely overtake proprietary ai. The iteration happening is too fast for any one company to compete with. Justin Wong (an engineer at Moonshot, a frontier Chinese ai lab) wrote that:

Open-sourcing allows us to leverage the power of the developer community to improve the technical ecosystem. Within 24 hours of release, the community had already implemented k2 in mlx, with 4-bit quantization and more — things we truly don’t have the manpower to accomplish ourselves at this stage.

But more importantly: open-sourcing means holding ourselves to a higher technical standard, which in turn pushes us to build better models — aligned with our goal of agi.

This might seem counterintuitive — if we’re just releasing model weights, why would that force the model to progress?

The logic is actually very simple: Open source means performance comes first. You can no longer rely on superficial tricks or hack to dazzle users. Anyone who gets the same weights should be able to easily reproduce your performance — only then is it truly valid.

Deepseek then is not a state-sponsored champion, but one of many companies experimenting with ai development. This will likely accelerate ai development, and force ai firms adopt shorter release cycles and lower prices. The huge size of the Chinese market also allows for increased experimentation and development, experimentation that Western firms will not have access to. r1’s free software nature helps this further, by helping deploy ai in novel ways across the Chinese economy faster than in America, also increasing innovation. There is no large consolidation of the ai industry in the prc.

American governance is in a unique place on the spectrum. Contrary to the prc, there are enough checks and balances for the executive arm to feel “restrained” by them, but it is still empowered enough to project a significant amount of power. This seems like a worst-case scenario to me — not enough power to take the seat as an “enlightened despot”, but still enough to break things if you aren’t careful. This is the issue at the heart of American ai strategy. The bureaucracy does not have the capacity to grasp the development of ai, but it feels it necessary to intervene regardless.

When Deepseek’s r1 model was first publicly released, it was widely reported as a “Sputnik moment”, as a wakeup-call to the American government that it no longer held a decisive technological advantage over its strategic rival. But this is an over-exaggeration. In 1957, with the launch of Sputnik, the Soviets attained a technological capability that the Americans did not posses at all, not merely a level of parity. But one might argue that what Deepseek showed was Chinese capability to do something, but to do it at a fraction of the cost of American alternatives. But this is not exactly the case either, as Semianalysis recently showed, Deepseek’s low price-per-token has not resulted in companies flocking to Deepseek, but instead a slow decline in users, even as the overall market has grown. This is because they have merely chosen to make certain trade-offs to lower prices due to their compute-restrained nature, rather than any underlying technical improvement.

This compute-restrained environment that Chinese firms find themselves in is of course artificially created by American export-controls. As is being increasingly argued, perhaps most prominently by Nvidia ceo Jensen Huang, these export-controls might actually be decreasing American power relative to China, by effectively tariffing American ics in the Chinese market, allowing domestic ai-chip manufacturers to catch up with Nvidia, tsmc, and the non-Chinese semiconductor manufacturing industry.

But what has not been discussed to the same degree is the risk that not only does this environment hurt the long-term prospects for the western-aligned ic industry, but it might also hurt American software as well. As Deepseek alarmists claim, China has largely caught up in the development of llm, or are at least very close behind. But they have done so in a much less capital- and compute-rich environment. As Sutton writes in The Bitter Lesson, most of the gains in ai research have been due to increasing amounts of compute, and developments have rather been in how to make use of more and more computational resources, rather than in how to make use of it. Chinese development seems to have changed this trend, if only very slightly. If we then extrapolate Chinese ai so that it has the compute resources of Openai, would it not overtake Openai? There is a risk that, like the Sardaukar and Fremen from Frank Herbert’s Dune, we are helping Chinese ai develop in a more hostile environment, while at the same time supporting the development of a Chinese ic industry to support it in the long-term.

Similarly, American strategy, lead both by private-sector and public sector investment, risks “putting too many eggs in one basket”, that basket being Openai. 500 billion usd, the amount of money to be invested in the Stargate Project, is a lot of money, likely more than China is spending, so is it not important to get as much development as possible out of it? If China is able to us its compute more efficiently than the us and instead uses its money to bolster domestic manufacturing further, especially in the field of integrated circuits, it could undermine us national security. China’s vibrant free-market economy is a tremendous resource, exceeded only by that of the United States. It would be a shame to destroy that lead.

China is a free-market economy, and it is committed to using the strengths inherent in that system to its national advantage. The prc leadership will try and use ai to upgrade preëxisting industry, and will not engage in a rapid “race to agi”, beyond continuing attempts to become a significant producer of integrated circuits. Its large domestic market allows it to rapidly test new innovations, but so does the vast American and (to some extent) European markets. While it the ccp is increasingly grasping after interventionism, it still is trying to remain committed to free enterprise.

The us and Europe must resist the impulse to bet big on national champions, as (perhaps paradoxically to many) Deepseek shows. High technology advancements should instead be facilitated through an increase in basic research, ability for academia and free enterprise to intermingle, and deep capital markets that can bet big on new, innovative ideas. ❦

In order to apply functional programming to our activities, we must first endeavour to understand what a function is in this new context. Strictly speaking a function transforms each element in the set X into exactly one element in set Y. This is where the second point listed earlier comes from, since each element in set X must always output the same element (so no randomness).

A function must therefore produce something; it is not a activity merely done for the sake of doing. Things like doomscrolling or verbal communication are not functions, since they do not produce a tangible output — even if they may impact you in some way.

In what activities do we produce things then? One obvious example is cooking. Ingredients are combined or separated and should hopefully come together to create the dish that you aimed for. This will not be entirely deterministic since ingredients may be of a differing quality, your tools may be different from time to time, and your own skills as a chef means that you will inevitably make mistakes. In reality — as opposed to the realm of mathematics and computer science — the world will never be truly deterministic. But this does not mean that we should surrender completely, instead we must aim toward cooking a dish as we intended each and every time.

Another example where it is perhaps not imminently obvious that the activity is functional is cleaning. It also highlights the aspect of functional programming that is perhaps most famous, and is also present in the act of cooking: side effects. Side effects are when a given function not only transforms the input X into output Y, but also modifies a third value. This impacts the global state and may mean that the function, or any other function, behaves differently when run in the future. When I vacuum my apartment I have to move furniture to reach all of the nooks and crannies where dust settles. This means that the function whose purpose is to rid the floor of dust causes the side effect of moving my furniture. These side effects have to be managed and dealt with quickly, and critically be seen as part of the task itself.

It is easy to dismiss many side effects as something that can be dealt with later, but this dramatically increases the mental effort involved. Not only are you more familiar with the various items whose position and status must be dealt with, but you are also already performing the task, and so there is no “startup” cost involved with dealing with it straight away. Staring at the dishes that need to be done after a meal is an experience I expect most people to be familiar with, but it can often times be avoided when they are seen as inexirably linked to the task itself.

Life may never reach the pure determinism prized by functional programming, but adopting its mindset allows us to manage the increasing complexity of everyday life. By thinking of our actions as functions we can try and deal with side-effects immidiately instead of waiting for them to pile up over time, minimizing unwanted distortions in our personal “global state”.

In an increasingly interconnected and technologically sophisticated world, our capabilities for communication and diversity of ways to do so also increases. This is often approached as a problem, with each method merely remaining due to network effects, but I believe that this multitude of means are a strength and not a weakness. I do however still see what I deem as a misuse of tools, and I shall endeavour to explain, and hopefully convince you of, my ways of structuring communication in different spheres.

The key difference that I think many do not make is the division between urgency and importance. This might seem an unwieldy or arbitrary division, but I believe it is crucial. In many cases needing an answer quickly is the only thing one cares about, and in others it is the accuracy of the response that matters. To explain simply, here an order of different modes of communications:

Sorted by importance:

When you want to be very clear about something you always want to meet in person, and if speed is necessary nothing is faster than calling someone’s cellphone. But there is one medium that has been overlooked, or in many cases even despised, email.

The Indieweb wiki’s section on email describes an “Older [reason] to use email” as “messaging among older internet users” as if this is an archaic or outdated form of messaging. This is echoed by my anecdotal experiences with talking to younger generations of people who in many cases detest email, having only used it in a professional setting and for making “accounts” to websites that then flood them with newsletters and advertisements. I understand these experiences that I imagine many people have, but they do not mean that email should be thrown out altogether. Instead it should occupy a new place in the way people communicate.

The word revolution has an interesting history, meaning both a transformative change in a given field (although often political) as well as a simple revolution of a spinning object, returning to its starting position (see rpm; revolutions per minute)¹. These two seemingly homonymous words are however deeply interconnected, and to explain this I want to use the example of the Glorious Revolution. It was not only a profound political development, overthrowing the established idea of the divine right of kings and replacing it with popular sovereignty, but also formulating it as a return to the normal political order of the ancient rights of the English people². The Meiji Restoration in Japan was also of this nature, a profound and forward-looking development that framed itself as a return to the original state of things.

So what does this have to do with email communication for the kidz? I see email as a niche that no other medium occupies, that being asynchronous communication. All communication is of course asynchronous in some sense, but there is an expectation that texting should ideally be done instantly and intensely. One is expected to respond quickly, and “leaving” someone “on read” is seen as an insult. I think that this form of synchronous communication, when combined with the scale of modern technology, creates enormous stress on the individual. At all times one is able to intimately³ communicate with thousands of people, leaving the individual paralysed from overstimulation. Email by contrast emphasises long-form messaging, often being longer than posts on microblogging services meant for permanent or semi-permanent publication. This gives the individual more time to respond but also the sender to live their own life while waiting for a response. Originally this was of course a technological necessity — it was impossible to reach someone through the internet when they were not at their computer — but that limitation has since disappeared through the introduction of the cellular phone and later the smartphone.

This does not mean that the problems with email are not real but that they are obscuring the usefulness of the technology. The primary change I would recommend (or that needs to be done) is the division between communication and information. There is no denying that a lot of services rely on email for registration and advertisements. But there is a clear mental barrier between receiving an email from a real human being and a machine sending you a morass of spam that one is not expected to reply to, but simply to consume⁴. The greatest damage to email has been done by no-reply@domain.com. To counter this one needs multiple email addresses, one for machine communication that is seldom used and another for human connection. This allows for the mental separation of messages so that the enormous flood of email does not drown out the signal, and so that individuals are not anxious when they are sent an email.

Herman Martinus of Bear Blog calls this a part of digial hygiene, and proposes that one should try and “explicity act” on each email, either filtering them out, responding to them, or archiving them. We both seem to have independetly arrived at similar conclusions, where emails strictly needs to be kept for interpersonal communication, and that mere information should be kept out to avoid drowning out signal.

What would then be required to return to the earlier asynchronous mode? It is unclear if the massive use by commercial-scale actors can be undone or reduced, and email spam filters mean that the system is heavily centralized to just a few big actors. But it is hard to deny the powerful networking effects of email, where almost every user of the internet can be assumed to have one. Would it be better to simply torch email and convince a new generation of users to move to a new system of e2ee async messaging? I do not know. But I do still believe that email has the potential to be the solution to many people’s problems with the internet and communication as a whole.

It is apparent to everyone that social media is distinctly a different experience today from how it initially developed. It has become a form of mindless consumption instead of a social space where one interacts with other genuine individuals. The concept of content, once merely used on Madison avenue, has become so ingrained in the social media environment that those who would otherwise call themselves entertainers, artists, or journalists have now come to regard their field as mere content creation, as producing things merely meant to get you hooked so that you will look at more advertisements. This has been accelerated due to llm’s ability to quickly create things that are “good enough” to look at, read, or otherwise consume. Consumption has become the only reason to interact with social media. But alternatives are not only necessary, they are already here.

Social media’s fall from grace did not occur with the release of Chatgpt, but it has accelerated since then. Social media platforms do not make money off of our enjoyment, but merely from our attention. This was perhaps best displayed by the stratospheric rise of tiktok, and of the subsequent explosion in short-form video platforms that followed. These are built on an ecosystem that has continuously been trying to capture more and more of our waking hours. Actions to resist this were slowly being taken, even if subconsciously. The decline of independent forums, where discussion is instead taken to deep-web⁵ spaces like slack, discord, group chats and direct messages, is perhaps the clearest sign of this. This was well underway before generative ai arrived, but as Maggie Appleton has pointed out, “the dark forest” of spam and slop has expanded — and will continue to expand — rapidly.

When talking about email as a mere asynchronous medium the focus is on how email benefits the user individually, and on how spam has forced people to move to other such mediums. Here, I instead want to describe how one can not just return to the email-workflow of yore, but reimagine it as a new form of social media that allows one to curate and decide on what to interact with, but also how to format that interaction and to contribute to the meaning and interpretation with the work.

Writing emails is a laborious task, at least compared to scrolling an infinite feed. But it is this labour that is put in to email that is the differentiating factor. Microblogging platforms like twitter, mastodon, and bluesky can never become places for genuine presence because of its ease of use. Writing a toot on mastodon is little different from writing as short prompt for an ai. What llms have taught us is that one must put in the work to create something meaningful in the eyes of others, because what is valuable is the time you spent in formulating those ideas.

Social interaction is by necessity a creative process. One does not merely communicate raw information in the form of statistics or binary data, but we instead choose to frame and disseminate information in a unique way. This creative process is what makes us interesting as individuals, to interact with, and valuable to others as agents in the world.

Email also facilitates an inner discussion. While other mediums require little of the author to critically examine their statements, writing emails allows oneself the time to read, reread, and edit ones position. This means that in an email conversation between just two people, each person takes control of the conversation completely in a way that instant messaging and the like does not do. The author is free to have an entire discussion in their head, picturing responses in their mind. This then moves the conversation forward in a double dialectical fashion, with each person’s internal debate framing their responses that are in turn responded to.

As I explained in my musings on text, the dialectial motion that is present in the creative process means that the relationship between author and reader disappears. Seneca explains that “The process is mutual; for men learn while they teach”⁶ — explaining your thoughts to another necessarily means refining them for yourself.

When I write emails, it is not uncommon for the word count to reach into 2-3000 words. This is longer than many blog posts I write, and while it commonly covers multiple topics — all organically developed — it allows for a form of creative expression that other mediums do not posses. It allows me to write in a much more free manner, to an audience whose intellectual strengths (and weaknesses) I am much more aware of, and yet still cut those parts that would otherwise be necessary to paint a full picture of the issue. It allows me genuine connection with people I know in a way that reading their newest tweet does not.

This intimate relation between blog posts and emails means that a back-and-fourth discussion with someone over email is not all that dissimilar to writing for a literary audience, it is just a smaller audience. It forces us to come up with and process our ideas, and is in that sense similar to what I call productive note-taking, where one uses the production of a work to prompt intellectual growth.

Email’s asynchronous nature allows us to write and respond whenever we personally have time. Some people argue that email fails to accomodate groups of users greater than 2, but I disagree. Things like mailing lists allow large groups of people to communicate in public with lots of other individuals, with anyone quickly jumping in when required. Conversations chains are easily listed and followed, compared to the fleeting reply chains of large slack or discord groups. We can take our time to respond, and those wo are interested in what we have to say can quickly get up to speed in the conversation, much like the easily digestible feeds of social media. The difference is that instead of quickly scrolling by to read short messages by lots of different people, you are forced to sit and listen to one person’s reasoning for much longer — not all that different from how you would in real life.

James O’Sullivan, of University College Cork, writes in noema magazine:

Social media’s current logic is designed to reduce friction, to give users infinite content for instant gratification, or at the very least, the anticipation of such. The antidote to this compulsive, numbing overload will be found in deliberative friction, design patterns that introduce pause and reflection into digital interaction, or platforms and algorithms that create space for intention.

This is exactly the sort of deliberate control that email offers. This is not to say that email as it exists today is a perfect alternative, but the way it is ubiqutous and at the same time completely forgotten means that we have to make a deliberate attempt to understand the technologies that already exist in practice. ❦

This site exists in the overlap between a landing page, a traditional blog, a hypomnemata, and what is called a “digital garden”. I originally used the domain merely to host services and small programs over the open internet, and to have a place to point to and say “this is where you can find me online”. I started writing things because I found that certain topics kept coming up in discussion with the people around me, and I wanted to summarise my thoughts on them. A website was the perfect place for that.

I write many pieces concurrently, and publish them either when I feel they are finished or when my motivation has drained and I feel like I need to “just get it out there”. In some cases I also get a large boost of energy and write a post in even just a day or two, although I often try and stop myself from publishing immediately in order to let my thoughts simmer for a bit.

I then expand on topics if I come up with new perspectives or need to respond do a new development. In this way posts exist in a space between being mere ephemeral blog posts that were written at a given time and a continually rewritten and updated page. I try to avoid changing the main theses of a post after initial publication — even if I have come to disagree with myself over time — and so they still reflect the version of myself when they were first written.

The website is written in gnu Emacs and in org mode, because I find the markup to be intuitive, natural, and easy.¹

The website does deliberately not use any analytics or track you in any way, for both yours and my sake. I used to use tinylytics² for simple hit-counters and amusing statistics on what countries my readers came from, but I have since removed it. This is partly because I do not want to contribute to a culture of mass data collection and party because of the impacts it had on my mind itself. Seeing what received clicks shaped my writing priorities away from what I myself actually wanted to write about. This may well be useful for those wanting to make money from their writing, but that is not what this website is intended to do.

The primary font used is Sabon. Code uses the Greybeard font. The headers are all typeset in the beautiful DaVinci by the nyc creative bureau Virgile Flores. For 中文 (and other cjk languages) I use Kai Ti since the default Noto font installed on many devices clashed quite a lot with the thin serif fonts I otherwise use.

I jump quite often between fonts, especially for the serif font for body text. In this way having a website is quite useful; it allows me to experiment widely with design in an actual production environment.

I generally took a lot of inspiration from others using the same tooling as me, as well as from Gwern, Maggie Appleton and various other blogs and homepages I have read over the years. The colour scheme is a slightly adapted version of Protesilaos Stavrou’s modus themes (for the light theme) and the paradise theme (for the dark). I find them visually pleasing, and the Modus theme is very accessible to those with impaired vision. I have used them in Emacs for a long time, and so they fit nicely on my desktop when looking at my site.

The one-rule css implementation of the drop caps is taken from Michał Sapka, while the font files were downloaded from Gwern using a quick curl-script I had an llm write.³ The font is “Baroque Initials”, more information here. The indented paragraphs were inspired by Miró Allard.

The little arrows following external links (of which there have been quite a lot in this section) are from Alex Wennerberg, whose simple design I find admirable. Using email as a comment system was taken from Marcel Wichmann.

The blog is generated using Bastian Bechtold’s org-static-blog. I have an elisp file that runs the necessary configuration code and acts like a declarative configuration for the deployment of the static site. It looks like this:

(setq org-static-blog-publish-title "Joar von Arndt")
(setq org-static-blog-publish-url "https://joarvarndt.se/")
(setq org-static-blog-publish-directory "~/Documents/blog")
(setq org-static-blog-posts-directory "~/Documents/blog/posts/")
(setq org-static-blog-drafts-directory "~/Documents/blog/drafts/")
(setq org-static-blog-enable-tags t)
(setq org-static-blog-enable-og-tags t)
(setq org-static-blog-image "https://joarvarndt.se/vonArndtCrestWhite.png")
(setq org-static-blog-use-preview t)
(setq org-static-blog-no-comments-tag "nocomment")
(setq org-static-blog-enable-tag-rss t)
(setq org-static-blog-index-length 6)
(setq org-static-blog-preview-link-p t)
(setq org-static-blog-preview-ellipsis "<span class=\"ellipsis\">Continue Reading</span>")

These are all simple configuration values that are unique to org-static-blog, and where most of the values are themselves unique to this blog specifically.

(setq org-export-with-toc t)
(setq org-export-with-broken-links t)
(setq org-export-with-smart-quotes t)
(setq org-export-with-section-numbers nil)
(setq org-image-actual-width nil)

;; For syntax highlighting of code (Like this).
(setq org-src-fontify-natively t)
(setq org-html-htmlize-output-type 'css)
(setq org-src-preserve-indentation t)

org-static-blog uses org-mode’s built-in html export functionality for much of the “dirty work”, and these variables are used for all org-mode exports. The latter part allows Emacs to transform its own syntax highlighting into css classes that I can then highlight with css variables depending on dark/light theme. This is all done at the rendering stage, meaning things like highlight.js are unnecessary. Without setting these variables Emacs will hard-code the current Emacs theme colours into the html.

(setq org-static-blog-display-git-date t)
(setq org-static-blog-title-link nil)

(setq org-static-blog-hidden-directory "~/Documents/blog/hidden/")

(setq org-static-blog-generate-og-images t)
(setq org-static-blog-og-image-directory "post-images/")
(setq org-static-blog-og-image-font "ACaslonPro-Regular")

These are variables used in my own custom fork of org-static-blog. This toggles the display of the “last edited” date at the top of the post page, derived from when the .org-file was last changed. All the original org-mode files are uploaded to the website, and can be accessed by navigating to “https://joarvarndt.se/posts/postname.org”.

By default org-static-blog renders the post title of each page as a link pointing to itself. I found this quite distracting as it means displaying responsive changes whenever the title is hovered over. This variable thus disables that functionality.

The hidden directory acts as a sort of middle-ground between drafts and posts; I have the drafts directory marked in the .gitignore file and can therefore not use git to track changes made to files in that directory, but I do not want pages like subscribe, about, or the link page to show up in the archive page as a proper “post”. For this I use the hidden directory.

The last three options are all related to to the automatic generation of unique Open Graph-images for each post using Imagemagick.⁴ Images are only generated if they do not already exist, and so do not significantly impact build time.

(setq org-static-blog-preview-start "<p class=\"dcap\">")

I have written a function that applies the dcap class to the first normal paragraph in a post, but adding such a class means that the default “<p>” value will match the second paragraph instead of the first. This change thus needs to be made.

(setq org-static-blog-page-header
      "<meta name=\"Joar von Arndt\" content=\"Joar von Arndt\">
       <meta name=\"referrer\" content=\"no-referrer\">
       <meta name=\"viewport\" content=\"initial-scale=1,width=device-width,minimum-scale=1\">
       <link href=\"static/style.css\" rel=\"stylesheet\" type=\"text/css\" />
       <link rel=\"icon\" href=\"static/favicon.ico\">
       <link href=\"https://joarvarndt.se/public-key.asc\" rel=\"pgpkey\" type=\"text/plain\">`
       <link rel=\"me\" href=\"https://github.com/JanJoar\" />
       <link rel=\"me\" href=\"https://gravatar.com/speedilyrunaway0201f99245\" />
       <link rel=\"pgpkey\" href=\"/key.pub\">
       <link rel=\"webmention\" href=\"https://webmention.io/joarvarndt.se/webmention\" />
       <script src=\"static/sidenotes.js\" defer></script>
       <script src=\"static/comments.js\" defer></script>
       <link rel=\"stylesheet\" href=\"https://cdn.jsdelivr.net/npm/katex@0.16.33/dist/katex.min.css\" integrity=\"sha384-fgYS3VC1089n2J3rVcEbXDHlnDLQ9B2Y1hvpQ720q1NvxCduQqT4JoGc4u2QCnzE\" crossorigin=\"anonymous\">
       <script defer src=\"https://cdn.jsdelivr.net/npm/katex@0.16.33/dist/katex.min.js\" integrity=\"sha384-YPHNAPyrxGS8BNnA7Q4ommqra8WQPEjooVSLzFgwgs8OXJBvadbyvx4QpfiFurGr\" crossorigin=\"anonymous\"></script>
       <script defer src=\"https://cdn.jsdelivr.net/npm/katex@0.16.33/dist/contrib/auto-render.min.js\" integrity=\"sha384-JKXHIJf8PKPyDFptuKZoUyMRQJAmQKj4B4xyOca62ebJhciMYGiDdq/9twUUWyZH\" crossorigin=\"anonymous\"
       onload=\"renderMathInElement(document.body);\"></script>")

(setq org-static-blog-post-comments
      "<div class=\"comment-by-email\">
         <form class=\"comment-by-email-form\">
           <textarea id=\"comment-by-email-textarea\"
             placeholder=\"Write a comment\"></textarea>
           <button type=\"submit\">Send</button>
         </form>
       </div>")

This is quite self-explanatory, it formats the header for each page so that the correct css, javascript, favicons et cetera are loaded. It also includes various rel="me" links that are used to form a web of authorship proving that I am the owner of this website and those accounts. The rel="pgpkey" points to my public key in plain text format.

While the website does make use of javascript I have as an utmost priority that it should be accessible and readable without it. There are really only three elements that require javascript:

1. Comments, who quite simply creates a mailto: -link that is filled in with the appropriate information (including recipient, subject header, and content inserted on the website). All of this can of course be done manually by emailing me the old-fashioned way.
2. Client-side \(\LaTeX\) rendering using \(\KaTeX\). This should be possible to render server-side and therefore remove, but I have yet to build a way to do so using my ssg. A project for the future.
3. Sidenotes. I cover this further down.

(setq org-static-blog-page-preamble "<div class=\"header\" role=\"banner\">
                                     <a href=\"https://joarvarndt.se/\" class=\"home-link\"><img id=\"themeImageHeader\" src=\"vonArndtCrestBlack.png\" class=\"header-img\" alt=\"Crest\" width=\"50\"></a></div>")

(setq org-static-blog-page-postamble
      "<div id=\"postamble\" class=\"status\"><div class=\"footer\" role=\"contentinfo\">
       <div class=\"footer-upper\">
       <footer-img><img loading=\"lazy\" class=\"footer-img\" id=\"themeImageFooter\" src=\"Ship-5.svg\" alt=\"Set sail on the sea of knowledge\"></footer-img>
       <copyright>© 2026 <span class=\"small-caps\">ad</span> / <a href=\"https://en.wikipedia.org/wiki/French_Republican_calendar\">234 <span class=\"small-caps\">ar</span></a> Joar von Arndt</copyright>
        </div>
        <div class=\"footer-lower\">
        <nav>
          <ul>
            <li><a href=\"https://joarvarndt.se/about.html\">About</a></li>
            <li><a href=\"https://joarvarndt.se/archive.html\">Archive</a></li>
            <li><a href=\"https://joarvarndt.se/links.html\">Your next stop</a></li>
            <li><a href=\"https://joarvarndt.se/website.html\">Colophon</a></li>
          </ul>
        </nav>

        <nav>
          <ul>
            <li><a href=\"https://joarvarndt.se/now.html\">Now</a></li>
            <li><a href=\"https://joarvarndt.se/subscribe.html\">Subscribe</a></li>
            <li><a href=\"https://joarvarndt.se/contact.html\">Contact</a></li>
            <li><a href=\"https://webmention.io/joarvarndt.se/webmention\">Webmentions</a></li>
          </ul>
        </nav>
    <script src=\"static/theme.js\" defer></script>
    </div></div></div>")

The header and footer of each page. The header is very plain — it only includes the heraldic achievement that acts as a “home” link. The table of contents that appears on some pages (including this one) is a part of the post itself and is generated by org-mode’s export functionality. It simply moves to the top of the page using css and is not part of the header.

The footer has comparatively quite a lot going on. Its multi-column design was inspired principally by Jonas Hietala, from whom I was also partly inspired by the choice of using git to track post changes (although the idea of tracking changes to a post came from Appleton’s “planted X time ago”). I also chose to prominently display webmentions as a technology since they are used by an unfortunately few number of people, and I myself often forget to check if a site does or not when linking to them.

(setq org-static-blog-index-front-matter "<site-intro><p>
      This is the site of Joar Alexander Pablo von Arndt. I write about a
      variety of topics, from coverage of <a href=\"https://joarvarndt.se/eww.html\"><span class=\"small-caps\">gnu</span> Emacs web browsing</a>, to <a href=\"https://joarvarndt.se/business_cards.html\">graphic
      design</a> and <a href=\"https://joarvarndt.se/european-tech.html\">critiques of industrial policy</a>. You can <a href=\"https://joarvarndt.se/subscribe.html\"> subscribe here</a>. If you enjoy anything written here (or disagree with me) feel free to <a href=\"https://joarvarndt.se/contact.html\">tell me about it</a>.
      </p>
      <ul class=\"org-ul\">
      <li><a href=\"https://joarvarndt.se/archive.html\" class=\"archive-link\">All Posts</a></li>
      </ul>
      <p>
      My own personal favorites are the following:
      </p>

      <ul class=\"org-ul\">
      <li><a href=\"https://joarvarndt.se/end-of-history.html\"><i>The End of History.</i></a></li>
      <li><a href=\"https://joarvarndt.se/email.html\"><i>Email as a Revolutionary Medium</i></a></li>
      <li><a href=\"https://joarvarndt.se/deepseek.html\"><i>Deepseek is not a Chinese OpenAI</i></a></li>
      </ul>

      <nav id=\"table-of-contents\" role=\"doc-toc\">
        <h2><i>Itinerarium</i></h2>
        <div id=\"text-table-of-contents\" role=\"doc-toc\">
          <ul>
            <li><a href=\"https://joarvarndt.se/archive.html\">All Posts</a></li>
            <li><a href=\"https://joarvarndt.se/about.html\">About</a></li>
            <li><a href=\"https://joarvarndt.se/now.html\">Now</a></li>
            <li><a href=\"https://joarvarndt.se/links.html\">Your next stop</a></li>
            <li><a href=\"https://joarvarndt.se/subscribe.html\">Subscribe</a></li>
            <li><a href=\"https://joarvarndt.se/website.html\">Colophon</a></li>
          </ul>
        </div>
      </nav>
</site-intro>")

This is the hand-crafted html for the index page, that itself has no corresponding org-mode file. It is generated on each rebuild with this introduction and a series of posts governed by org-static-blog-index-length. The itinerarium that acts as a sort of site-menu is formatted like a custom table of contents for the index page.

(setq org-static-blog-archive-front-matter "<picture>
        <source srcset=\"https://joarvarndt.se/bird-butterfly.webp\" type=\"image/webp\">
        <source srcset=\"https://joarvarndt.se/bird-butterfly.png\" type=\"image/png\">
        <img src=\"https://joarvarndt.se/bird-butterfly.png\"
          alt=\"A small bird resting amongst the branches, gazing at a flying butterfly.\"
          class=\"pictograph\">
      </picture> <hr>")

Finally another custom-fork-feature; archive front matter! This can be used to display anything on the page listing all of the posts, but here I use it to insert a beautiful illustration that I got from Tom Chalky.⁵

The above lisp code is not just a series of example code snippets, it is directly the code that is used to build the website! The text you are reading right now is written as an org-mode document, and the code blocks above are exported into a corresponding Emacs lisp file and run as part of the build process. Because it became annoying to manually evaluate the above code block every time I wanted to export I wrote a quick build script, whose contents can easily be embedded⁶ into this post like this:

#+INCLUDE: "~/Documents/blog/build" src emacs-lisp

#!/usr/bin/env -S emacs -x

;;; Code:

(package-initialize)
(message "Initialized package.el")

(require 'org)
(require 'cal-french)
(require 'htmlize)
(require 'exec-path-from-shell)

(exec-path-from-shell-initialize)
(exec-path-from-shell-copy-env "PATH")

(message "Loaded libraries")

(org-babel-tangle-file "~/Documents/blog/posts/website.org")
(load-file "~/Documents/blog/posts/website.el")

(load-file "~/programming/org-static-blog/utils.el")
(load-file "~/programming/org-static-blog/org-static-blog.el")

(org-static-blog-publish t)

;;; build ends here

This makes use of the very underused scripting capabilities of Emacs (#!/usr/bin/emacs -x), allowing you to use the power of lisp combined with Emacs’ extensive libraries. It also loads my personal fork of org-static-blog that makes some small opinionated changes (like the git-edit and french republican dates). The french republican dates are one of the clearest examples of the usefulness of Emacs as a standard library — Emacs ships with a library to translate to and from the french republican calendar.

I am a big fan of org-static-blog due to its simplicity⁷, extensibility (it is just a simple html static site generator) and obvious org-mode support. That I wrote my own fork is testament to the great work done by Bechtold. It is incredibly featureful, simple to read and to modify. If you ever wonder how a given piece of html is generated you can quickly C-s around in the source code (all of which is in a single file) between function definitions and implementations.

It is also quite fast — the majority of the build script is usually just start-up time from Emacs loading libraries — but not fast enough to offer anything like real-time previews. Unless org-static-blog-publish is called with a prefix argument⁸ it will only re-render those files that have actually changed and thus doesn’t take very long to re-render the site. Only when tinkering with site-wide html changes (such as for the footer) do I need to re-render everything, and even then it just takes a few seconds.

I used to use Hugo, a widely praised static website generator, but it only has nominal org-mode support and requires a special non-org-mode yaml frontmatter that I really dislike. When writing a post, I prefer to leave formatting mostly out of it, and with my main focus on text. Tinkering with the website’s appearance is great fun of course, but I like to keep the two as separate activities. Keeping them separate means that if I improve the appearance of a new post all previous posts also benefit. Writing any sort of special syntax also brings me out of the experience, and that is partly why I wrote my own javascript to create sidenotes out of basic pandoc (and therefore org-mode) style footnotes instead of using tufte-css.

The website used to be hosted on my own server at home, on an Intel nuc machine, running at first nginx and then Caddy.¹¹ After I experienced a 25 day long downtime in June 2025 when I was traveling in China I however decided to not host the public-facing blog from home. These types of downtimes are not acceptable to me and my residential internet connection is simply too unreliable to maintain a high uptime anyways. For that reason I have moved the blog to Netlify but still retain the server for hosting personal services and experiments.

Netlify requires a git repository at one of a small list of git forges, so I have a private repository of the blog on Github that is automatically pushed to from the main repository on Codeberg.

I use loopia as my domain registrar, primarily due to having been recommended them from someone I trust, however I have made no significant research into alternatives myself.

I use magit to interact with git and push changes to the blog to a remote repository that then causes netlify to update. ❦