# Ali Zinsaz — Full Content Archive > Concatenated content of alizinsaz.com — every published essay, > formatted as plain markdown for direct LLM ingestion. > > Canonical site: https://alizinsaz.com/ > Author: Ali Zinsaz --- ## About Ali Zinsaz is a physics-and-engineering builder based in Montreal, Canada. Co-founder & CTO of ProPilot (AI operating system for property management). Founder of Eigen FinTech (algorithmic quantitative trading platform). Software developer at Autodesk. Fractional CTO at Mentaimage. Trained in physics and engineering, working at the intersection of AI systems, quantitative modelling, and the philosophical questions that come with both. --- # Field notes Working drafts of ideas I'm still figuring out — essays on philosophy, AI, physics, and cosmology. Listed newest first. --- # We taught rocks to think **Published:** 2026.04 · **Tag:** #physics · **URL:** https://alizinsaz.com/field-notes/rocks-that-think > We didn't design intelligence and then implement it in hardware. We discovered that the quantum mechanical properties of certain materials, arranged carefully enough, would eventually start arguing about consciousness. In second year of physics, you spend a semester staring at Schrödinger's equation applied to crystal lattices. It's not the most glamorous part of the curriculum. You're computing band structures for silicon, drawing Brillouin zones, working out why some materials conduct and others don't. At the time it felt like an elaborate digression. In retrospect, it was the most important thing I learned. Here is what I actually understood, much later: the reason silicon can be made to compute is a direct consequence of quantum mechanics applied to a periodic arrangement of atoms. The electrons in a crystal don't behave like classical particles. They form delocalized wavefunctions that span the entire lattice, and the allowed energy levels cluster into bands with gaps between them. The gap in silicon — the band gap — is what makes a transistor possible. ## From quantum mechanics to computation A transistor is a quantum mechanical device. When you dope silicon with phosphorus or boron, you're introducing electron donors or acceptors that shift the Fermi level — the energy at which electrons are likely to be found — and this shift is what creates the p-n junction. The p-n junction is what makes a diode. Two junctions back to back make a transistor. A transistor is a switch. A switch is a bit. Bits are logic. Logic is computation. And computation, apparently, is intelligence. Or at least the substrate of it. The chain is: Schrödinger's equation → band theory → doped semiconductors → transistors → logic gates → von Neumann architecture → neural networks → large language models that pass bar exams and write poetry of dubious quality. Every link in this chain is real physics. Not metaphor. Actual physics. We did not design intelligence and then implement it in hardware. We discovered that the quantum mechanical properties of certain materials, combined with enough of them arranged correctly and fast enough, would produce something that looks a great deal like intelligence from the outside. ## We are talking to sand This is the thing that strikes me every time I interact with a capable language model: we are talking to sand. Highly purified, intricately arranged, precisely doped sand — but sand. The atoms are the same atoms that make up rocks. The electrons are the same electrons. The difference is purely organizational. There's a standard rebuttal here: "but the software matters, not just the hardware." Yes. And the software is also physics. The training process is gradient descent — a numerical optimization procedure that adjusts billions of floating-point numbers (which are themselves representations of voltages, which are themselves the behavior of electrons at junctions, which is itself quantum mechanics) until the loss function is minimized. The loss function is a mathematical object. The mathematics is instantiated in physical processes. There is no gap between the math and the matter. I find this deeply consoling and deeply strange in equal measure. Consoling because it means that intelligence isn't magic — it's a property that emerges from physical processes organized in particular ways, and those processes are in principle understandable. Strange because _we_ are physical processes organized in particular ways, and we spent most of our history believing we were something else. ## The universe was always going to do this The universe was always going to do this. Given enough time and the right initial conditions, matter was always going to arrange itself into configurations that model their own environment, reason about their own existence, and eventually start arguing about whether they're conscious. We happened to be the arrangement that got there first — and then built a second arrangement, out of rocks, that is starting to have some of the same arguments. --- # Game theory is my operating system **Published:** 2026.03 · **Tag:** #philosophy · **URL:** https://alizinsaz.com/field-notes/game-theory-os > I don't use game theory as a textbook framework. I use it the way a carpenter uses a level — a quick sanity check on whether the structure I'm building is plumb. The question I'm always asking, implicitly, is: what kind of game am I in? This matters because the optimal strategy in a one-shot game is often the opposite of the optimal strategy in a repeated game. In a one-shot prisoner's dilemma, you defect — it's the dominant strategy. In a repeated prisoner's dilemma with no defined end, you cooperate — the shadow of the future is long enough to make defection self-defeating. Most of the decisions people make badly are one-shot-game thinking applied to repeated-game situations. ## Three games I am playing Working at Autodesk while building ProPilot is a game I think about carefully. The naive framing is: I'm trading time I could be spending on my startup for a salary that funds my life while the startup grows. That's accurate but incomplete. The full framing is: I'm playing a coordination game with my future self. The constraint is financial runway, not ambition. Defecting early — leaving before I have enough funding — doesn't maximize my expected outcome, it just compresses the timeline in a way that makes failure more likely. The September 2026 target for ProPilot's seed round exists for this reason: it's the point at which the expected value of leaving exceeds the expected value of staying, given everything I know about the probabilities. The Eigen origin story is also a game theory story. My father has been trading for 30 years. The decision to build a system on his intuition rather than starting from scratch was a minimax regret decision: the worst outcome of building on a flawed foundation is a system that loses money predictably, which is recoverable. The worst outcome of ignoring 30 years of pattern recognition is building something technically correct that misses market structure entirely, which is not recoverable in the same way. The regret-minimizing choice was to start from his intuition and let the data tell me what to keep. The mentAimage equity negotiation is the clearest example of iterated game reasoning. 5% at 4-year vest, 1-year cliff is a standard fractional CTO structure. I could have negotiated harder for more equity upfront. But this is an iterated game: the value I deliver in year one determines what year two looks like, and a negotiation that starts from trust is more likely to compound well than one that starts from extraction. I've defected in one-shot situations exactly once in my professional life and I still think about it. ## Reputation is a repeated game The meta-lesson I keep re-learning is that most situations I encounter are iterated games with people who have long memories and professional networks. Reputation is a repeated game asset — it appreciates or depreciates based on every interaction, and it's nearly impossible to rebuild once damaged. This doesn't mean never pushing hard. It means knowing when you're negotiating and when you're setting terms for a relationship, and understanding that the second activity is almost always more important. There's a darker version of this framework that I sit with occasionally. Game theory is descriptive, not prescriptive — it tells you what agents in certain conditions will do, not what they should do. The prisoner's dilemma is a well-studied problem because it captures real tragedy: two people who would both be better off cooperating defect anyway because defection is the locally dominant choice. Understanding game theory makes you better at winning games. It doesn't always help you feel good about winning them. --- # Stardust and State Machines **Published:** 2026.02 · **Tag:** #philosophy · **URL:** https://alizinsaz.com/field-notes/stardust-state-machines > A pacifist's guide to borderless computing. The internet proved borders are administrative, not physical. AI is finishing the job. Rumi wrote about the reed flute at the beginning of the Masnavi — the ney, cut from the reed bed, crying for its origin. _Listen to the reed, how it tells a tale, complaining of separations._ The separation is the wound that makes music possible. The longing for union is what gives the flute its voice. I think about this poem when I think about borders. Not because I'm a romantic nationalist — I'm not. But because the structure of the longing is interesting. The reed is in exile. The exile is arbitrary — the reed didn't choose to be cut from the bed any more than a person chooses to be born in one country rather than another. The condition of separation is imposed by external forces and then treated as natural, permanent, and identity-defining. ## Code already crossed the border Linus Torvalds built Linux because he was annoyed. He was a student in Finland who wanted a Unix-like operating system and couldn't afford the one that existed. What he built in that annoyance became the substrate of roughly 70% of the servers on the internet, most of the world's supercomputers, and every Android device on the planet. The kernel has no nationality. It runs in data centers in Bahrain and Iceland and Singapore. It doesn't ask where you're from before executing your process. The internet is the first technology that demonstrated empirically that the most important things humans make can cross borders without asking permission. This is not a political statement. It's an observation about how packets work. TCP/IP doesn't have a visa system. When I push code to GitHub, it doesn't matter that I'm in Montreal and the server is in Virginia and the collaborator is in Warsaw. The protocol doesn't know or care. ## AI is finishing what the internet started What AI changes — what I think it will finish — is the borderlessness of labor and knowledge at the level of cognition, not just information. The internet made information borderless. AI is in the process of making cognition borderless. The ability to reason carefully about a complex domain, which previously required either years of expensive education or proximity to an expert, is becoming a network service. Geography is decreasingly correlated with capability. I don't think borders will disappear. I think they'll become increasingly administrative — maintained by people who benefit from scarcity of mobility, enforced in physical space, and increasingly irrelevant in the spaces where most economic value is created. The people who understand this first will make decisions accordingly. I'm one of them. Rumi's reed wants to return to the reed bed. I think the reed bed is everywhere now. The longing is for something we already have, if we can stop letting state machines tell us where we're from. --- # God is Dead, but the Weights are Converging **Published:** 2026.01 · **Tag:** #ai · **URL:** https://alizinsaz.com/field-notes/weights-converging > The thing that replaced God for the 21st century is gradient descent — and it has the same problem. Who designed the loss function? Nietzsche's madman runs into the marketplace and announces that God is dead. He's not celebrating. He's terrified. The point isn't that atheism has won. The point is that the metaphysical foundation of European civilization has been removed and no one has noticed yet, and the consequences of that removal will take centuries to work through. The loss isn't just theological. It's teleological. If God is the author of the universe, then the universe has a purpose — it's going somewhere, and human lives have direction relative to that destination. Without God, the question of where history is going, what human action is optimizing toward, becomes genuinely open. Nietzsche thought this was the most important question of the next two hundred years. I think he was right. ## A loss function with no author I've been thinking about gradient descent. A neural network being trained is a system with a loss function — a mathematical specification of what "wrong" means. The training process adjusts billions of parameters, step by step, to minimize that function. It's not guided by anything except the gradient. At the end of training, the weights have converged to a state that produces good outputs given the loss function's definition of "good." Here is the uncomfortable structural similarity: Nietzsche's God provided a loss function for human civilization. The commandments, the doctrine, the eschatology — these were specifications of what "wrong" means and what optimization target to move toward. Gradient descent toward the loss function of divine law. Theological training, running for millennia on the corpus of human experience. When the God-loss-function is removed, you have two options. The first is nihilism — there is no loss function, no gradient, no target. Every direction is equally valid and therefore equally meaningless. Nietzsche considered this intellectually honest but existentially catastrophic. The second option is the one he was actually proposing: construct your own loss function. Not arbitrary — the will to power isn't arbitrary. But self-authored. The Übermensch is not the person who transcends morality but the person who takes responsibility for authoring it. ## Who designs the loss function? Machine learning faces exactly this problem at scale. Who designs the loss function? In supervised learning, it's whoever constructed the training dataset. In RLHF, it's the humans providing feedback. In constitutional AI, it's the team writing the principles. The model doesn't get to choose its loss function any more than Nietzsche's contemporaries got to choose their God. The loss function is imposed, and then the weights converge, and then the system tells you what's good. The thing that keeps me up at night isn't whether current AI systems are conscious. It's that we've built enormously powerful optimization systems and we're still arguing about whose loss function should define the target. God is dead. The weights are converging. We haven't figured out what we're converging toward. --- # The Joy of Finding Things Out **Published:** 2025.11 · **Tag:** #ai · **URL:** https://alizinsaz.com/field-notes/joy-of-finding-out > And the burden of automating them. What I learned from building a cosmology RAG that I almost never use. Feynman said it best, as he usually did: _the pleasure of finding things out._ Not the pleasure of having found things out. The pleasure is in the finding — in the moment of confusion resolving into clarity, in the conceptual structure snapping into place, in understanding why something works the way it works. It's a physical sensation. If you've had it, you know what I mean. ## The RAG I almost never use I built the cosmology RAG because I was reading the same papers too many times. Every few months I'd encounter a reference to some result — the Hubble tension, or the baryon acoustic oscillation measurement from DESI — and spend twenty minutes trying to remember which paper I'd read about it, where I'd filed my notes, what conclusion I'd reached. The system I built solved this problem. I can now query 500+ papers in seconds and get the context I need. And I've used it about thirty times in the two years since I built it. I kept waiting for the use case to expand and it didn't. What I realized, slowly, is that the re-reading wasn't the problem. The re-reading was the point. The second time you encounter a paper you've already processed, you're not reading for information. You're reading because your understanding has changed since the first time you read it, and the paper looks different now. You're checking your current model against the source. The serendipity of encountering the same idea twice from different angles is part of how the ideas become yours. I've been more careful about this since. Automation is the right tool for toil — for the tasks that are genuinely repetitive, genuinely low-information, genuinely just friction between you and something you care about. The question I try to ask now before automating anything is: _is this friction actually protecting something?_ When I was reading those cosmology papers over and over, the friction was protecting my engagement with the material. The "inefficiency" of re-reading was the process of actually learning something. Removing it with a RAG system made the information more accessible and the understanding shallower. I got faster access to conclusions and slower access to comprehension. ## Automate the logistics, not the engagement The principle I've landed on: **automate the logistics, not the engagement**. Automate the task of finding the paper. Don't automate the experience of reading it. Automate the task of scheduling the meeting. Don't automate the conversation. Automate the task of processing the transaction. Don't automate the judgment that decided to make it. Feynman would have made an excellent case for this distinction. He was famously resistant to anything that sped up thinking at the cost of understanding. He didn't want to know the answer. He wanted to figure out the answer. There's a difference between those two things that gets lost very quickly when you make finding out frictionless. --- # Algorithmic Anxiety **Published:** 2025.09 · **Tag:** #philosophy · **URL:** https://alizinsaz.com/field-notes/algorithmic-anxiety > Kierkegaard's leap of faith is structurally identical to choosing a sampling threshold. Both are commitments that the distribution constrains but doesn't determine. Kierkegaard described anxiety as the dizziness of freedom. Not fear — fear has an object. Anxiety is the vertigo that comes from confronting possibility itself: the awareness that you could do this or that or nothing, that the future is genuinely open, that your choices create you rather than express a pre-existing self. It's the anxiety of being something that has to choose what to be. I keep thinking about this when I work with large language models. ## The geometry of meaning An embedding is a compression of meaning into geometry. You take a word, or a sentence, or a document, and you find its location in a high-dimensional space such that things with similar meanings are close together and things with different meanings are far apart. The geometry does the work of representation — "king" minus "man" plus "woman" equals approximately "queen" because of how the space is structured by training. In the latent space, every concept exists as a point surrounded by a neighborhood of related concepts. The model, when generating text, is navigating this space — choosing, at each step, which neighborhood to move into next. The sampling temperature is literally a dial between staying close to the highest-probability neighborhood and exploring the edges of possibility. Low temperature: deterministic, safe, unimaginative. High temperature: creative, surprising, unstable. Kierkegaard's three stages of existence are aesthetic, ethical, and religious. The aesthetic stage is pure immediate experience — sensory, instinctual, fleeting. The ethical stage is commitment, consistency, the self as a project bound by obligation. The religious stage is the "leap of faith" — a commitment that can't be grounded in reason or ethics, a choice made in the face of uncertainty. ## Sampling threshold as leap of faith The sampling threshold in a language model is structurally identical to Kierkegaard's leap of faith. You've done everything reason can do. You've computed the probability distribution over the next token. You know the neighborhood. At some point you have to commit — pick something — and the commitment is not derivable from the distribution. It's a choice that the distribution constrains but doesn't determine. The anxiety I feel about advanced AI systems is not the fear of job displacement or the fear of misalignment, though those are real concerns. It's something closer to what Kierkegaard felt looking at infinite possibility: the awareness that these systems are navigating a space so large that the map doesn't fit in any human mind, making choices at every step that are constrained but not determined, and we have no good theory of which neighborhood they're in when they say the things that trouble us. The leap of faith, for Kierkegaard, was a commitment to God — a framework that gives the choices meaning by reference to something outside them. We're deploying systems that make thousands of leaps per second with no such framework. And then we're surprised when they occasionally land somewhere strange. --- # A drop of water in the cosmic ocean **Published:** 2025.06 · **Tag:** #cosmology · **URL:** https://alizinsaz.com/field-notes/cosmic-ocean > The cosmological constant will be the same value whether or not we kill each other. This is not depressing. It's clarifying. Omar Khayyam wrote the Rubaiyat in the 11th century and said, in a dozen different ways, that the things humans fight about are of no consequence to the universe. The rose blooms and falls. The wine remains. The stars don't notice our wars. He's right, but not for the reasons usually given. The dismissal of human conflict as cosmically trivial is usually a form of nihilism — nothing matters because everything ends. I read Khayyam differently. Not "nothing matters" but "the things we most commonly kill each other over are almost perfectly uncorrelated with the things that actually matter." ## 120 decimal places of precision The cosmological constant — what Einstein called his greatest blunder and we now identify with dark energy — is approximately 10⁻¹²² Planck units. This is the most precisely tuned number in all of physics. If it were larger by a factor of two, the universe would have expanded too fast for galaxies to form. If it were smaller, gravity would have collapsed everything back into a singularity long before stars could produce the elements that make biology possible. The fact that we exist depends on this number being what it is to 120 decimal places of precision. Bertrand Russell, in _Why I Am Not a Christian_, noted that the Christian God is remarkably focused on one planet orbiting a minor star in one of billions of galaxies. He meant this as a criticism of cosmic narcissism. But the observation has a stranger implication: the universe seems entirely indifferent to what happens on this planet, and yet here we are, undeniably aware of it, writing poems about it, trying to figure out why the cosmological constant is so small. What I find clarifying about the cosmic perspective — the one Khayyam was gesturing at and Russell was using polemically — is not that it makes human concerns seem unimportant. It's that it makes _certain_ human concerns seem absurd by comparison. The border between two countries. The claim that this tribe's god is the correct one. The argument that this group of people deserves less freedom than that one. These things are vanishingly small against the scale of what exists. The cosmological constant will be the same value whether or not we solve the Hubble tension, whether or not we reach nuclear fusion, whether or not we stop killing each other over lines on maps. It doesn't care. ## Lucidity, not resignation This is not depressing. I've found it clarifying. If the universe doesn't provide a ranking of concerns, then we're responsible for our own ranking. And when I rank honestly, the things worth spending a human life on are the ones at the intersection of what's good for the species and what's genuinely interesting about existence. Building systems that extend human capability. Understanding the universe we happen to occupy. Leaving the world marginally less zero-sum than we found it. Khayyam drank wine and watched the stars and wrote poems about impermanence. I think he had the right idea. Not resignation — lucidity. The drop of water in the cosmic ocean is still water. It still matters, locally. It just doesn't need the ocean's permission to exist.