When I was a kid I had a ‘Stretch Armstrong’. You could pull that guy in every direction. Arms, legs, neck, you name it. My sister and I would each grab an end and try to rip the thing apart like it owed us money, but no matter how far we stretched him, no matter how distorted he got, the moment you let go he slowly oozed back into the same shape with the same dumb grin. We threw everything we had at him and the only thing we changed was his size.
That is, and I did not realize this until twenty years later or so, exactly what an eigenvector is.
Now, why would you care about something called an eigenvector.
Fair question, my friend, I give you that.
You would care because every system you interact with has one.
The social media feed that keeps dragging you toward clickbait and outrage content even though you never asked for it. . .Yup. That one’s based on an eigenvector. Or take the company that talks about innovation every quarter but somehow always defaults to cost-cutting (ahem), yeah, funny enough that’s also an eigenvector. But what about the hiring pipeline that produces the same kind of candidate no matter how many diversity initiatives get launched . . . you guessed it, eigenvector. Heck, Google’s entire original business model was even built on one and the reason Netflix knows what you want to watch before you do – same thing.
Let me start with a quote:
“Eigenvectors are the hidden directions that every system defaults to when nobody is paying attention”.
Keep this definition in mind and it will open up a whole new universe full of wonders you never thought you’d know.
These tiny math gems decide what gets amplified, starved to zero or what gets delete, and they’re everywhere but perfectly hidden so you won’t have to take out your calculator – you do not even need to know the math to feel its effects. You have been living inside someone else’s eigenvector your whole life.
You just did not have a word for it.
But now you do.
An eigenvector is the direction that survives the stretch.
You can apply force, pressure, distortion, the full transformation, and this one direction refuses to rotate. It might get longer, or it might even get shorter, but it keeps pointing the same way whatever you do to it. The system tried to change it, but the system failed, and the only thing the system could do was make it bigger or smaller.
Stretch Armstrong knew this before any of us took a linear algebra or matrix multiplication class. But before I explain the details behind that rubbery stubbornness, let me tell you what the word itself means. Because the name “eigenvector” is the whole story, and English almost ruined it . . .
More rants after the messages:
- Connect with me on Linkedin 🙏
- Subscribe to TechTonic Shifts to get your daily dose of tech 📰
- Please comment, like or clap the article. Whatever you fancy.
What “Eigen” actually means
The word “eigen” is German. Jawohl. And since I’m Dutch, which is a rebellious offshoot of the original Germanic culture (cause they preferred wet feet over mountains), I use the same word on a daily basis.
You pronounce it like ‘EYE-GUN’. And a British chap called David Hilbert stuck it onto mathematics in 1904 when he used the terms Eigenwert and Eigenfunktion in a paper on integral equations, but let’s not go there. Mister Hilbert was not starting from nothing though because before him the Prussian army surgeon Hermann von Helmholtz (klopp!) had already used something he called Eigentöne to describe the natural resonant frequencies of a room. He was obsessed with how humans perceive sound. But his real project was not rooms – he wasn’t an interior decorator or audicien – he just wanted to understand how the human ear breaks apart complex sounds into individual tones, how we hear a chord and somehow separate it into notes, or how we tell the difference between a violin and a flute playing the same pitch†
His theory, which turned out to be largely correct, was that the inner ear works like a room full of tiny resonators. The cochlea, the spiral-shaped structure in your inner ear, contains thousands of hair cells along a membrane. Each section of that membrane resonates at a different frequency, like strings on a piano and when a sound wave hits your ear, the membrane vibrates most strongly at the point that matches the incoming frequency. That is how you “hear” a specific pitch. Your ear is doing frequency decomposition, splitting a messy wave into its component tones‡
And to prove this, he needed to understand resonance itself. And the simplest example of resonance is a room. We’ve all been into a mostly empty room yelling ‘Echo’, and what you get back is a distorted version of the original sound because every enclosed space has natural frequencies, tones that it amplifies or dampens on its own because of its shape and size and the materials used. Sing the right note in a bathroom and the whole room hums, up to the point your mirror breaks. Now that is an Eigenton. A tone that belongs to that particular room. The room’s own ‘frequency’ so to speak.
But remember he was a physician, and he studied rooms to understand resonance, so he could explain what the ear was doing, so he could build a physical theory of hearing.
The Eigenton is the sounds a space produces on its own, without anyone telling it to. The tones that belong to the room itself.
And that idea of belonging is the entire key to the word. “Eigen” does not translate cleanly into one English word. It sits in a cluster of meanings that English splits across half a dozen terms including ‘Own’ – like self, inherent, characteristic, peculiar, particular, belonging to, intrinsic.
In everyday German, eigen shows up everywhere.
Eigenkapital means equity, literally “self-capital”, the money that is actually yours, not borrowed. Eigenschaft means a property or attribute, something that belongs to a thing by nature – and no, it’s not ‘eigen-shaft’ you dirty, I know you . . . Eigenart means the particular nature of something, its distinctive character. Eigentum means property in the ownership sense. Even Eigenartig, which means strange or peculiar, carries the idea of something being so distinctly itself that it stands apart from everything around it.
The English word “own” actually shares the same ancient root as eigen. They are etymological cousins, but fortunately English never developed a compact prefix that carries the same weight (German and Dutch are such complicated languages, you really don’t want to go there, trust me). So when Hilbert’s math crossed into English, translators tried “proper value” and “characteristic value” for a while, but neither stuck and in the end, everyone decided to keep the German word.
Two syllables beat five. So Eigen won by efficiency, which is fitting for a concept about what survives.
So when you say eigenvector, you are saying something closer to “the vector that belongs to the system”. Its own vector so to speak – the direction that is inherent to the transformation, not imposed from outside. The system’s self-expression. The direction the system produces naturally, the same way a room produces its own resonant tone.
And when you say eigenvalue, you are saying “the system’s own value for that direction”. How much the system amplifies or diminishes the thing that already belongs to it. The system’s honest opinion about its own nature, expressed as a number.
This is why the German is better than any English translation.
“Characteristic vector” sounds clinical, “proper vector” sounds like a dress code to me, but Eigenvector carries the meaning that this direction is the system’s own. It belongs to the system the way a fingerprint belongs to a hand which is a thing you did not assign to it, you did not choose it. You discovered it since it has always been there.
The whole idea lives inside that one German prefix. Your own direction. Your own value. The thing that is characteristic of you and nobody else. The math just gave it a formula.
Now, before I throw some numbers and strange glyphs at you, let me show you what this looks like when you can actually see it.
† If you’re interested, Thursday will air a new post about a new segmentation model from Meta, which basically solves this problem for both audio and visual elements.
‡ If you’re into electronics or math, you’ve probably heard of Fourier Transformation. Fourier is the guy that built the math that allows for any complex function (like an overlap of different sounds) to be broken down into a sum of simple waves.
Eigenvectors as a movie you can watch
Forget the formulas for a minute. Let us first think about movement like characters in a scene from a movie. Every mathematical concept here has a role to play, and the easiest way to understand the plot is to watch it happen.
Let’s start by picturing a rubber sheet that is lying flat on a table with a simple line drawing on it. A bunch of straight lines radiating out from the center, pointing in every direction like a starburst. Simple, clean, geometric.
Now grab the edges and pull. Stretch the sheet diagonally, unevenly, the way real forces work in real systems and watch what happens to those lines. Most of them will twist, they curve and warp and bend into shapes they never agreed to. The rubber moved, and it took almost everything with it.
Look again at the image, almost everything warped and in all that chaos, but the red lines are still perfectly straight, still pointing in exactly the same direction as before – they got longer, sure – the stretch pulled them out, but they did not bend. Every other line on the sheet lost its identity, but these two kept theirs.
The system applied its full force and these directions simply refused to change. Everything around them moved. They, however, did not.
And those two lines are the eigenvectors.
These are the directions the transformation could not rotate, the only thing the system could do was make them longer or shorter, that is it. That is the whole Eigenvector concept, visible on a piece of rubber. A system full of force and distortion, and the stubborn directions that survive it unchanged.
You can stop reading now, because you basically know all you need to know. The rest is only about how these things present itself in real life. About things that have an ‘internal compass’ so to speak.
The Eigenvector is the stubborn one that refuses to bend
In this paragraph, I’ll explain the concept of the Eigenvector and the Eigenvalue in more detail, so you know how it is applied in real life.
When I talk about the Eigenvector, I call it the ‘inner’ direction that refuses to compromise. Everything around it may bend its knee and deforms, but this particular arrow will not change where it points. It stubbornly stays on its original line. It might get longer or shorter, but it will not turn. The Dutch word “eigenwijze” fits perfectly here, someone who insists on their own way.
The eigenvector is the eigenwijze direction. It heard the system’s instructions and decided they did not apply to them.
Think about a spinning wheel as a concept.
Every spoke is constantly changing position as the wheel turns, and the spokes even stretch out a bit – though you won’t see it because it is so minute – but the axle in the center keeps pointing the same way no matter how fast the wheel spins. The axle is the eigenvector. Everything else moves, but the axle stays.
And the Eigenvalue is the volume knob. Once you have found the direction that refuses to turn, the eigenvalue tells you what happened to its size.
If the eigenvalue is 2, the eigenvector got twice as long. The system likes that direction and amplified it and when the eigenvalue is 0.5, the eigenvector got cut in half – the system is slowly starving it or if the eigenvalue is -1, the eigenvector flipped to point the opposite way, but it is still on the same line. The system reversed it without rotating it. That is how it works, same track, different direction and it’s mathematically tidy, but emotionally confusing.
This is the moment for the formula that ties these two, the direction and it’s value together
The Matrix is the Director. The force that causes the transformation, and the rule that makes “everything pointing this way now change like that”. In our rubber sheet example, the matrix is the person pulling the edges, and in the spinning wheel, the matrix is the motor. It is not a character in our scene, but the one calling the shots, deciding how reality gets deformed.
Think of space, or more precisely spacetime, as a clean non-physical example of a matrix. It behaves like a matrix, but not like a spreadsheet you can sort by column B. It is a rule system that tells matter and energy how to move and interact, but not something you can point at. A matrix is not an object you hold. It is a rule set that states that if something enters its domain with this position and this motion, this is how it will leave.
In Einstein’s universe, spacetime is a structure that tells matter how to move, and matter tells spacetime how to bend and once those rules exist, everything else is forced to comply. Light bends, and time slows down and planetary orbits curve. Not because they want to themselves but because the system applies its transformation relentlessly.
Mathematically, physicists describe this using tensors and metrics, which are essentially generalized matrices. They encode how distances, directions, and time intervals change depending on where you are and how fast you move.
So when an object moves through spacetime, spacetime applies its rules to that motion. Again. And again. And again.
That is matrix behavior.
And yes, this means spacetime has its own “eigen-directions” like stable orbits and geodesics – paths that remain consistent under transformation.
Now look at space as a machine. And every once in a while, something goes through the machine and comes out looking basically the same – same shape and direction. Maybe it got bigger, or perhaps smaller, but in the end it is still pointing the same way.
The machine tried to change it and failed.
That thing is the eigenvector. It is the direction the machine cannot mess with.
And the amount the machine made it bigger or smaller is called the eigenvalue. And it has a score. A high score means the machine likes that direction and makes it stronger, or a low score means the machine weakens it and a score of zero means the machine kills it completely.
That is it.
That is the core idea. It is a direction that survives a machine, and a number that tells you how much the machine cares about it.
And the formula is Av = λv, where A is the matrix representing the system itself – all the rules, incentives, constraints, and transformations baked into one mathematical object that takes inputs and spits out outputs according to its internal logic. The v is the eigenvector, the stubborn direction that refuses to get twisted or rotated when the system gets applied to it, the trajectory that survives intact while everything else gets scrambled into unrecognizable mush. The λ (lambda, because mathematicians hate using normal letters when Greek ones are available) is the eigenvalue, the scaling factor that tells you whether the system amplifies that direction into dominance or suppresses it toward extinction or maybe leaves it roughly where it started. This value is the single number that captures how much the system cares about this particular direction compared to all the others it’s busy crushing.
So Av means “apply the system to this direction” and when that equals λv – when applying all the rules just stretches or shrinks the direction without changing where it points – you’ve found something the system cannot corrupt, only scale, and that’s your eigenvector staring back at you with the smug satisfaction of having survived the gauntlet.
The whole Eigenvector concept is so strong because it tells you that his direction is so aligned with how the system works that the system can’t do anything except make it bigger or smaller”, which is why eigenvectors end up running the show whether you planned for them or not.
Av = λv. Clean, short, and brutal.
What is a matrix really?
Let’s strip it to bedrock.
A matrix like spacetime is a Rube Goldberg machine† for applying the same transformation repeatedly.
That is it.
It does not think and it certainly does not care. A matrix simply takes an input and produces an output according to it’s fixed rules.
You’ve probably bumped into matrices today without realizing it, because they’re sneakier than Zucky’s terms-of-service updates. AI uses them to turn your typo-riddled prompt into suspiciously coherent text where it is multiplying matrices like it’s a possessed accountant until patterns emerge from the numerical chaos. And video games use them to spin camera angles and make lighting look moody enough that you forget you’re staring at simple triangles. And what about the recommendation algorithms we’re bombarded with at Facebook and Netflix, they use matrices to figure out that because you watched one video about alpacas, you clearly want your entire feed to become a camelid petting zoo. Image compression, voice recognition, physics simulations, GPS navigation – matrices are literally the duct tape holding the digital universe together, and they don’t even get dental.
Ha!
Which brings me to GPUs.
Yup, those overpriced space heaters that we’re buying for our computers. They are matrix-multiplication sweatshops full of Oompa Loompas and these things are designed to slam numbers together billions of times per second because turning “rotate this dragon 47 degrees while light bounces off its scales during a particle explosion” into pixels requires more math than your CPU wants to deal with before it had its morning coffee. Every time the camera swooshes around your character’s brooding face, yup, you got it, that’s matrices getting bodyslammed together at near lightspeed velocities. The GPU doesn’t care if it is rendering explosions or training an AI to write marketing copy, it just multiplies ‘excels’ until someone tells it to stop or (like in my case), the electric bill arrives.
AI does exactly the same thing, except instead of rendering cool explosions, it is calculating weights and pushing through a neural network to render mediocre poetry and confidently wrong medical advice. It’s simply data getting shoved through a meat grinder of matrix multiplications, layer after excruciating layer, while “learning” amounts to turning the volume knob (the weights) slightly left or right until the machine stumbles ass-backwards into something that looks like intelligence. Some directions get amplified into genius, but the rest mostly get squashed into oblivion, and we call this “deep learning” because it sounds better than “statistical brute force with good marketing from Jensen Huang”.
Gaming and AI. Different industries, but identical math. Graphics programmers call them transformation matrices and AI researchers call them weight matrices, but in the end, they’re both lying – it’s the same damned mathematical thing, just one group works for Nvidia and the other works for OpenAI.
Now, the power of matrices comes from three properties.
First, there’s consistency → A matrix applies the same rules every time. That is why repetition in AI matters. A one-off transformation is noise, but repetition reveals structure.
Then there’s directionality → Matrices treat different directions differently. Some directions get stretched while others get squashed.
And last, there’s accumulation → When you apply a matrix many times, small effects stack and things like weak biases disappear and strong directions dominate.
That is why eigenvectors exist at all. They are the directions that survive accumulation.
† A Rube-Goldberg machine 👇
What else behaves like a matrix?
You get it by now that matrices are rule systems that crush inputs into predictable outputs, and the thing is that once you see that, you can’t unsee it no more, and congratulations, you’ve ruined brunch conversations forever. But don’t fear, your talks will not only be about AI nor gaming because there are other concepts that behave like matrices and yes, the tool is used for those too . . .
For instance, organizations are matrices.
There you undergo things like policies, incentives, performance reviews, budgets – you feed a desired ‘behavior’ into the corporate meat grinder and check back in six months to see what limps out the other side. The thing, however, is that the behaviors that survive aren’t the ones in the mission statement, they are the eigenvectors of the organization, the directions that don’t get squashed by middle management’s “strategic realignment”. Your company’s identity, the Freudian ID so to speak, and not it’s Ego – the latter is the domain of branding and PR specialist – the ‘pretend’ people. Oh, want to know what your company actually values? Just ignore the posters in the break room or the statements on the intranet and watch which actions get promoted versus which get you a thoughtful conversation about “cultural fit”.
Markets are matrices as well.
Have you ever considered price signals, regulations, supply chains and investor panic attacks at 3 AM to be part of a matrix? Simply throw your brilliant contrarian idea into that blender and see how long you last swimming against the current. The market only cares about dominant directions, and if you’re not aligned with them, you get to learn about “liquidity events” the hard way.
And most AI algorithms are matrices.
Recommendation engines, ranking systems, ad auctions – the horrible math that’s deciding what you see, when you see it, and how guilty you should feel about it. You input actions and the system outputs reinforcement so you repeat that behavior, rinse and repeat until you’ve watched 47 videos about competitive cheese rolling and can’t remember how you got here. This my smart friend, is matrices all the way down, except instead of numbers it’s your attention span getting multiplied into ad revenue.
Data scientists use eigenvectors to simplify complicated information. Say you have data about a thousand customers with fifty different measurements each. That is a mess. Nobody can see patterns in fifty dimensions. So you ask this question, “if I had to boil this data down to just two or three important directions, which directions capture the most information?”. The answer is the eigenvectors. They are the skeleton of the data underneath all the noise. This technique is called Principal Component Analysis, and all it really does is find the directions the data fights hardest to keep when you squeeze everything else away.
Funny enough, you can consider cultures to be matrices.
Take norms, taboos, rewards, punishments, that awkward little thing everyone does but nobody talks about. Certain behaviors get amplified until they’re “just how things are done around here” and others simply vanish like my problematic opinions at our startup after someone introduced the group chat.
You don’t design a culture, but you create boundary conditions and let the dominant eigenvectors emerge, you just “see what survives when nobody’s looking”.
And yes, your habits are matrices too.
Daily routines apply passive-aggressive rules to your time and energy like a personal trainer you can’t fire. When you’re into something new, after some time a direction emerges – not the one you swore you’d stick to on January 1st, but the one that survived contact with reality – like your fundamental inability to wake up before 9 AM. My friend, you are the output of your own matrix multiplication! And that explains so much about why you’re reading this instead of going to the gym.
In all these cases, the math is optional, a nice-to-have for people who enjoy Greek letters, but the end-behavior is not. The crushing, repetitive, direction-selecting machinery grinds on whether you understand linear algebra or think eigenvectors are a German car brand. You’re in the matrix either way, you just don’t get to see the code.
Why eigenvectors only make sense when matrices exist
No, I’m not finished yet, because now the pretentious metaphor collapses into something uncomfortably useful, and you’re stuck here with me now.
An eigenvector isn’t some personality trait floating around in the ether waiting to be discovered like your Enneagram type or your Hogwarts house. It is a direction that emerges specifically because a system exists that cannot twist it, only amplify it into dominance or suppress it into irrelevance, and without that system applying pressure there’s nothing for the direction to resist against, no reason for it to remain stubbornly pointed in one direction while everything else gets scrambled.
A lengthy sentence, but it makes sense when I read it back.
Eigenvectors are like diamonds who are born from pressure, they are what survives when the system crushes everything else into compliance, which is why they’re honest in a way that aspirational vision statements and strategic plans will never be. They don’t describe what you hope happens or what the McKinsey consultant promised in slide 47, but they describe what the actual rules of the actual system allow to persist.
Scrollbreaker time!
The uncomfortable implication here, the thing that makes this more than just a funny math excursion, is that if spacetime itself behaves like a matrix – and yes, physics suggests it does, with its conservation laws and speed limits and annoying insistence that entropy only flows one direction – then structure is not some optional add-on you can decline during checkout.
It is fundamental to reality itself.
Freedom exists, absolutely, but it exists inside constraints the same way a river is free to flow but somehow never manages to flow uphill no matter how inspiring you find that Rumi quote. Movement exists, but only along paths that the underlying rules permit, which is why unfortunately you can’t teleport to work no matter how late you are, and stability exists but only in configurations where the forces and incentives and physical laws reinforce each other rather than tearing the whole thing apart like a poorly designed Jenga tower.
If you want different outcomes, you don’t waste time arguing with eigenvectors or wishing they were more cooperative or writing a manifesto about how things should work in a just universe, you redesign the matrix itself and you change the rules and incentives and constraints until different directions become the ones that survive. That’s how countries lurch from kingdom to dictatorship via plutocracy to democracy or communism and then slouch back through plutocracy to dictatorship again, because they’re redesigning the matrix over and over, tweaking which behaviors get amplified and which elites get to skim off the top.
👉 And that, friends and enemies and people who clicked here by accident, is why this whole discussion I’m having with you, keeps drifting back to sovereignty like my Weiner gravitates to his favorite toy, because whoever controls the matrix and whoever sets the rules and tweaks the incentives, controls which directions survive and which get mulched into irrelevance.
You can have all the values statements and core principles and inspiring all-hands speeches you want, but if the underlying matrix keeps amplifying you things you claim to hate, well, you have discovered what your system actually values, and it’s not what’s on the poster in the break room.
This brings me to Eigenvectors and the topic of Sovereignty and AI.
What do Eigenvectors and Sovereignty have in common?
At this point, you have spent close to twenty minutes reading my stuff about a simple piece of math, and now you realize that you have been fooled into reading an advertorial. I know you’re pissed off because you feel betrayed, but before you decide to call it a day and give me a thumbs down for this bait-and-switch, there’s actually a nugget in this piece of sponsored content that actually matters. It has to do with this thing we keep hearing everywhere at this moment, this word that politicians and tech executives throw around like confetti at a parade they don’t fully understand, and that word is “sovereignty.”
Yes, we built a company we named ‘Eigenvector’, and people ask us constantly why that name? What does some obscure concept from linear algebra have to do with sovereignty, and the answer is . . . everything. It has everything to do with it. And if you’ve made it this far through an article about stubborn mathematical directions and rubber sheets and Stretch Armstrong, you’re about to understand the following way better than 99,999% people who use the word AI in their slide decks.
Let me start with what sovereignty actually means.
Sovereignty in this context is not the flag-waving anthem-playing military-parade version that gets trotted out during election cycles. What I do mean is whether you control your own destiny or you’re just a variable in someone else’s spreadsheet.
Sovereignty basically is about a system that operates according to its own logic and its own values or direction, without being forced into someone else’s transformation, without getting twisted and bent and rotated to serve priorities that were never yours to begin with.
A sovereign nation sets its own laws instead of having them dictated by treaty obligations written in someone else’s capital. A sovereign individual makes their own choices instead of having them shaped by algorithmic nudges they never consented to, and a sovereign system runs on its own terms instead of optimizing for metrics chosen by whoever built the underlying infrastructure you’re renting from them.
Now back to the eigenvector.
Remember me calling it “that stubborn mathematical direction” that we’ve been circling for the last few thousand words?
An eigenvector is a direction that passes through a transformation and comes out unchanged in orientation – the outside system applies all its force, all its rules and pressure, but that direction holds like Stretch Armstrong refusing to tear no matter how hard you pull. It does not get rotated into something else and it certainly does not get bent to serve another system’s priorities. Whatever pressure it gets, it stays itself even when everything around it gets scrambled into chaos. The only thing the transformation can do is to scale it or try to make it bigger or smaller, or suppress it into irrelevance, but the direction itself, the identity, the orientation, that is non-negotiable.
That is sovereignty expressed as mathematics, and it’s not a metaphor I’m stretching to make my company name sound clever, but it is the actual structural principle that determines whether a system is self-directed or whether it’s just an amplifier for someone else’s agenda.
The current AI landscape is the opposite of this.
Big Tech has given us a massive infrastructure built on rented sovereignty where most AI systems are not sovereign at all but are eigenvectors of someone else’s matrix. They were built inside a specific corporate transformation, based in countries that don’t adhere to your world-view, and that were trained on priorities set by a specific set of incentives, optimized for metrics chosen by people or a business model that has nothing to do with whether the AI actually helps the people using it and everything to do with whether it helps the political goals or the quarterly earnings of the company that built it.
The direction was decided before the AI was born, but it got locked in during architecture meetings and objective function debates and boardroom conversations and oval offices, about who gets the right to what tool, but also things like addressable market size, and the AI just amplifies whatever direction the company and it’s overarching government was already pointing.
For instance, when a model is trained to maximize engagement, the eigenvector of that system is addiction – not because anyone sat down and said “let’s build something addictive”, but because the matrix rewards time-on-platform and the eigenvector that survives is the one that keeps people scrolling until their eyes bleed. And then you get abhorrent initiatives like Character dot AI that simply refuse to die, even after countless court cases.
If I tried to catalog everything currently wrong with how Big Tech handles AI – the concentration of power, the plutocratic capture of the American society and the cozy relationship it has with El Presidente that makes Gilded Age robber barons look like amateur hour – it would require a book-length manifesto that I’d get too angry with to write coherently and no one would read anyway. Instead, I’ve been ranting in installments.
The last year, I have created a scattered archive of blog posts in which I documented all the ways the current setup is fundamentally broken, who controls the data and how they monetize your digital exhaust, who owns the compute infrastructure and what that ownership means for everyone renting access, who trains the models and whose priorities get baked into the weights before you ever see the API documentation.
I have listed the links to a lot of them below this article, if you care to read on.
This has to stop.
But not because I’m an idealist who thinks we can build utopia with better prompt engineering, but because the trajectory we’re on ends with a handful of companies controlling the infrastructure layer for every decision that matters while everyone else becomes a tenant in their matrix, optimizing for their eigenvectors while pretending it’s autonomy.
And yes, writing about it mobilizes people, gets them angry, maybe shifts the conversation a few degrees, but it does not create alternatives, and at some point you have to stop complaining about the restaurant and open your own kitchen.
So that’s why we created Eigenvector.eu – the website is still embarrassingly barebones and looks like it was designed by someone who learned HTML in 1997 (me), so maybe don’t click until we are able to hire an actual designer, but we are here to make a difference rather than just document the decline. We are building sovereign AI because it is time, past time really, to take control of our own algorithmic destinies instead of outsourcing them to whoever has the biggest GPU cluster and the best relationship with venture capital.
The math says the dominant directions will emerge whether you plan for them or not, so we might as well plan for them.
Every AI system has eigenvectors. Every platform converges toward its dominant directions. The question is not whether your AI has a direction. It does, but whether that direction is yours.
That is why we build sovereign AI. That is why our company is called Eigenvector.
The math has always been our mission statement.
We just had to read it in the original German.
Signing off,
Marco
P.s. there’s an inconsistency in our Eigenvector branding. We did it on purpose. After reading this blog you should be able to spot the ‘flaw’. Drop me a line if you know what’s wrong.
I build AI by day and warn about it by night. I call it job security. Big Tech keeps inflating its promises, and I just bring the pins and clean up the mess.
👉 Think a friend would enjoy this too? Share the newsletter and let them join the conversation. LinkedIn, Google and the AI engines appreciates your likes by making my articles available to more readers.
The blog posts I talked about in the article👇
👉 So you think you own your AI? | LinkedIn
👉 These greedy-ass AI chatbots want all your data | LinkedIn
👉 The algorithm from hell we asked for (and now can’t stop) | LinkedIn
👉 LinkedIn is not helping you get hired. It’s helping itself get rich. | LinkedIn
👉 Seize the means of Artificial Intelligence – before they enslave us with it! | LinkedIn
👉 Trump’s new America. The bro-ligarchy | LinkedIn
👉 The $95 million apology for Siri’s secret recordings | LinkedIn
👉 AIs dirty little secret. The human cost of ‘automated’ systems | LinkedIn
👉 The $95 million apology for Siri’s secret recordings | LinkedIn
👉 Prediction: OpenAI will go public, and here comes the greedy shitshow | LinkedIn
👉 School surveillance systems are causing a staggering number of arrests | LinkedIn
👉 OpenAI’s teen safety theater is just that – theater | LinkedIn
👉 AI is built on human burnout and broken promises | LinkedIn
👉 TikTok employees spill the beans: ‘Oopsy, we made it too addictive’ | LinkedIn
👉 AI perverts make millions undressing your daughters for fun and profit | LinkedIn
👉 Buy for me, but ask no one | LinkedIn
👉 Oh great, another year of botshit piled on botshit, thanks Google | LinkedIn
👉 Blood-red algorithms | LinkedIn
👉 Palantir and ICE built an algorithmic deportation machine | LinkedIn
👉 America’s Tech is f∩cked. Here’s my plan for an un-Trumpable stack | LinkedIn
👉 Grieving mother takes on Character AI | LinkedIn
👉 AI search. The biggest con since snake oil, now with a subscription fee | LinkedIn
👉 How foreign bad peepss are weaponising AI | LinkedIn
👉 I’m sick of these parasitic Tech Bros and their “Creator Economy” scam | LinkedIn
👉 Musk’s $38 billion taxpayer money grab | LinkedIn
👉 Why Big Tech devours startups and leaves nothing but bones | LinkedIn
👉 MIT pushes decentralized AI to break Big Tech’s hold | LinkedIn
👉 Seize the means of Artificial Intelligence – before they enslave us with it! | LinkedIn
👉 Tech elites promise ‘public AI.’ Translation: Rent’s due. | LinkedIn
I didn’t lie, right?
Leave a Reply