Hey all, it’s been a few weeks since my last update. Let me explain.

I had this inner awakening a month ago: no one would get me back to the vita contemplativa, the life of ideas, unless I took action for myself. So I revived this dormant Substack account and started writing ideas and memories—some that I’ve had for years, some that have just started to form. It was therapeutic yelling into the void.

Unexpectedly, the void whispered back. I picked up a few new subscribers (hello!) including two paid ones (hello?!) A few likes on my comments here and there. Enough to make me dare to think…maybe I should take this seriously. Which is to say, maybe I should take myself seriously, because after all these years in tech, I still see myself as a writer, a scholar, and a storyteller.

So I started working on a big idea about AI context vs. AI souls—more on that hopefully next week. And I started seeking community. Calling old friends, coworkers, fellow entrepreneurs, professional philosophers. A coalition of the contemplative, assembling a salon. And above all: finding likeminded spirits through reading.

I started with ’s list of summer reads for 2025, then I worked with Claude to add authors relevant to my current research. A friend or two recommended some more works. Finally, I added some books the good ol’ fashioned way: whatever popped up in Amazon’s “you may also like” suggestions.

Here are the books that I’ve started reading along with my current thoughts:

Red Plenty by Francis Spufford (2010)

Some guys fall asleep to the soothing sights and sounds of World War 2 documentaries. My preferred hypnotic is YouTube videos about the inner workings of the Soviet economy.

This book—half fictional novel, half documentary history—is a tour de force explanation of what went wrong with the Soviet economic system, specifically the idea of reducing an entire economy to a rational, scientific, and planned basis. It has plenty of obscure details, almost pornographic in how much I salivate over them, about the mechanics of how individual individual sovnarkhoz would negotiate with state enterprises and Gosplan, the Union-level economic planning agency, to set prices and production quotas each year. But the book’s real genius is to reveal the gossamer linkages that bound those at the top, like Nikita Khruschev, to those nameless cogs, like the fictional tolkach Chekuskin, who made the system run.

AI has made it possible to reimagine the resurrection of the old Soviet cyberneticists’ dream, to control every aspect of an economy using technology to achieve perfect efficiency. Red Plenty’s overall message is that no matter how much technical progress we make, the success or failure of social engineering ultimately depends on what happens when ideals meet incentives and human nature.

The Strange Order of Things by Antonio Damasio (2018)

This one was recommended by Claude because it thinks that my nascent ideas about AI souls are very similar to what Damasio, who’s a prominent neurobiologist at USC, has to say about consciousness.

Damasio’s main argument is that consciousness arises from the creation of internal images within our brains, a development that’s strictly tied to the evolution of central nervous systems. The original evolutionary impulse for consciousness, he argues, is homeostasis, the self-regulating and up-regulating process that pushes organisms to higher levels of wellness.

I’m about a third of the way through and it’s a tough read. Damasio writes like a deconstructionist French philosopher, making assertions without any proof first then following with evidence rather than using evidence to build toward cautious suppositions. I really dislike this style of writing, it’s like constantly being taken prisoner and then released, over and over.

But I think the ideas are valuable and pose an interesting question: if consciousness is inherently embodied, can AI ever reach human levels of intelligence? Or at least, that’s the way the AGI/ASI question is usually framed. It’s clear to most observers that we’re moving past the notion that “human” intelligence is the yardstick for AI, but only because most presume that AI’s potential is so infinitely beyond human capabilities.

In contrast, the work of Damasio and others such as Peter Godfrey-Smith in Metazoa suggests that the true difference between human and AI intelligence may be in quality instead of quantity. The real question becomes, if disembodied but highly intelligent AI can never achieve consciousness, are we willing to let go of consciousness as a hallmark of intelligence?

Other than those books, I’ve been working on planning a wedding (did I mention that I got engaged?), paying the bills with my Mercor gig, and tax season.

My personal goal is to keep reading and writing, and I’d like to start building some AI systems based on some of my own ideas and research. Working at a startup like Workshop Labs would be a dream, but most of these smaller startups are in the Bay Area and I’m out here in lonely Tucson. OpenAI and the other big labs have started posting remote jobs, I’ve been thinking about applying at least as a web developer, and I’ve been working on my portfolio.

But I’m also want to work on something more than just UIs—I want to get into the real marrow and pith of AI, to take my humanistic and engineering training and create something wildly beautiful. I’m looking for an intellectual home. If you know of one that could benefit from my experience, please let me know—I’m open to new experiences.

A big thank you to my new paid subscribers who have helped directly contribute to that dream. Your trust and interest mean everything to me.

Talk to you real soon!

I’m fired up by ’s call for philosopher-builders, technologists who build with philosophical rigor and intentionality. But do they know how to make philosophically-grounded tech projects work in the real world? Does anyone?

It’s a question that has plagued me for over 12 years, one that’s especially relevant to those of us philosopher-builders who dropped out of the system and went rogue. The type, like me, who quit successful early-stage startups that later IPO’d or left the Bay Area for backwaters because we saw that the emperor had no clothes—and we were brave enough to venture into the unknown.

We rogues have been thirsting for the Cosmos Institute’s clarion call, but we’ve been burned too many times by companies whose philosophical stances turned out to be recruiting pitches, or intellectually rigorous projects that floundered on the rocks of market rigor. So we’ve retreated from ourselves, successfully building others’ startups, saving others’ companies, and training others’ teams, making half-hearted attempts at our own projects that usually end in disillusionment.

The surface problem is motivation. All startup founders must believe in their mission, but philosopher-builder founders must additionally be convinced of their mission’s intellectual rigor, or what we might call its rootedness. “Save the world” does nothing for me. A specific theory of how to save the world also doesn’t move the needle. But if you explain what “saving” means and can explain the complex web of how you would define a “world” over dinner, then I might start getting interested.

Rooting your views purely within yourself is almost impossible: philosophy and engineering are social sports that need a community. If she doesn’t have anyone to talk to about philosophy, let alone build with, then the philosopher-builder sits at home sad. I think this is what the Cosmos Institute is solving for, but for rogues I believe that the problem is deeper than that.

Why rogues need a school of thought

I have a close friend, let’s call him Frank, who’s one of the most talented operators in the tech industry and who’s philosophically-minded like myself. Frank and I have deep conversations about everything from AI to Descartes, from Daoism to the origins of time multiple times per week. I cherish our conversations, and you could say that he and I form a small community of philosopher-builders who are having some of the conversations that the Cosmos Institute fosters.

At the same time, there’s an unbridgeable philosophical gap between us. Frank’s a die-hard techno-optimist who believes in Progress and that, even if the short-term is rocky, technical progress eventually leads to objectively better outcomes for humanity. I’m no pessimist, but something about Progress with a capital “P” makes me wince. Neither position is right or wrong; it’s a question of temperament and how our minds work.

In the long run, the world needs both perspectives. The issue is that Frank has institutional support for his views whereas rogues like myself don’t. Frank truly believes in the industry and what he’s doing, so he throws himself into his work, and he’s appropriately rewarded. A rogue philosopher-builder who rejects the industry’s worldview not only lacks its support, they lack a school of thought that motivates them.

A school of thought is a shared pattern of meta-thought informed by the same structure and response to the same structural problems. Having community is a necessary but insufficient condition. I wrote about what a school of thought looks like and how I rediscovered mine while reading my former teacher’s dissertation.

In my case, it was my teacher, Tom Dechand, who formed the school that I come from through a deeply influential two-year undergraduate seminar that he taught at Johns Hopkins in the 2000s. When I read his work—or the work of Zachary Gartenberg, who sat in the same seminar with me as an undergrad and who wrote a brilliant piece on learning for the Cosmos Institute—I see my own mind at work.

At the top level, I see my, Tom’s, and Zach’s shared interest in the structure of thought, our emphasis on close textual readings, and our deep care for humanity. On closer examination, I also hear echoes of the same conversations and methods we explored almost 20 years ago. Their projects and their goals are legible to me—meaning, I can see how they connect to my own intellectual roots. I can integrate, interact with, and build on them. I not only understand their arguments, I grok where they’re coming from and, to a certain extent, who they are.

Rogue philosopher-builders desperately need the legibility that a school of thought can provide because we’ve rejected the tech industry’s narratives. It’s not about rejecting ideas that are foreign to one’s tradition, it’s about having a coherent base for integration.

Frank gets the base for free—the tech industry provides it for him, and he integrates external ideas into his worldview. It works because the external ideas don’t threaten the market viability of his core views, nor does holding the tech industry’s views violate his own integrity. He’s a true believer.

Rogues are in the opposite position. By default, our core views don’t fit into the market. We face the uphill battle of integrating the industry’s incentive structure into ours while also building successful startups and keeping our self-respect. We know how to build—we do it all the time—but doing it for ourselves requires meeting a bar that no one can meet without some external support. Finding a school of thought that can lead rogues to market validation through philosophical inquiry is the way to make ourselves whole and come in from the cold.

How to create a school of thought

Organizations like the Cosmos Institute have an opportunity. But how do you make a school of thought beyond a community? This is a hard goal and I won’t pretend to have the whole answer.

There’s an element of pull that resists explanation. Somehow, all of us who attended Tom’s seminar came with similar interests and observations from different fields. Call it resonance, shared temperament, or luck—Tom created the conditions of something worth running into. Once we were in the door, he amplified that resonance by welcoming our interests, giving them space during discussions, and even incorporating them into the syllabus.

Part of it is being opinionated. Tom really believed in his research, and even when we disagreed with him, we still thought a lot about his ideas, which influenced our own. For the Cosmos Institute and any similar organizations, I don’t know if “being opinionated” would be a feature or a bug.

A lot of it is just putting in the personal work. Taking those walks with your colleagues. Eating dinner and breaking bread together. Tom did all those things. So did teachers like my former undergraduate advisor , one of the most passionate and involved professors I ever had, who literally broke bread with me when he was the director of the American Academy in Rome. No doubt a couple of my tech colleagues would have, if only I ever met them in person (the estrangement of remote work is yet another reason why rogues need a school of thought).

Why rogues matter

The stakes are high. If, despite their best intentions, Cosmos Institute and others only bring together philosophically-minded operators like Frank—perhaps because their social proof is so legible—then they will propagate the same values and thinking that repelled us rogues in the first place. It's the same worldview they're arguably seeking alternatives to. But if Cosmos and others can foster a school of thought similar to how Tom did—truthfully, they may be doing this already—then those of us who went rogue on philosophical grounds will find a home. Their mission, which I’m enamored with and wholly support, is too important and too worthy not to.

Rogue philosopher-builders aren’t operators at FAANG enterprises or elite academic institutions; but we could’ve been. We don’t have well-funded departments where we can A/B test our hypotheses; instead, we deal with the nitty-gritty of small clients and uncertain distribution. If we start an enterprise and it fails, we don’t get reassigned, we work for a year without income and end up $60k in personal debt; but we try anyway.

We represent putting our money where our mouth is, and we’ve been waiting. The philosophical tech revolution needs us as much as we need it.

Last night, I looked up my former teacher’s dissertation.

My teacher, Tom Dechand, was a teaching assistant for an undergraduate humanities seminar that I attended 2008-2009. The seminar was held by the late and legendary Dr. Richard Macksey. You don’t know his name but your world has been shaped by this man.

In 1966, Dr. Macksey organized a symposium at his Baltimore mansion near Johns Hopkins University’s campus where he taught comparative literature and writing for close to 60 years. He invited a few French names that, if you’ve studied postmodernism, you might find familiar: Jacques Derrida, Jacques Lacan, Roland Barthes, and Gilles Deleuze, among others.

Dr. Macksey’s symposium is why we know these names in the Anglophone world; he was the midwife of postmodern thought. All of the past 60 years of deconstructionism and critical theory, its triumphs and its abuses, initially passed through this gate:

What you’re looking at is Dr. Macksey’s home library, or more accurately maybe about 20% of it. I know that it’s hard to believe, but trust me, the man read every single book you see. This room is where he hosted his 1966 symposium and nearly every class he ever taught. It’s this room where I found myself in as an undergraduate, overawed by the book collection and the towering intellect of the man who owned it.

I remember one time talking with Dr. Macksey after class about an obscure thirteenth-century astronomical text that was the focus of my thesis when he asked me to wait a minute. He left and returned with a copy of the same text. It just happened to be in his collection, replete with notes scribbled in the margins from many years ago. He opened to the same page that I had been referencing and started reading from it, fluently and casually in medieval Latin, as if it were nothing. Nearly every one of his students has a similar story of his intelligence and the generosity that matched it.

By the time I met Dr. Macksey, he was close to retirement and visibly in poor health. He nominally taught the seminar, but actually his graduate student Tom ran the show.

In that seminar, us undergraduates were willing guinea pigs for Tom’s dissertation research, which was on the English poet Samuel Taylor Coleridge, a name that evokes pipe smoke-filled parlors like Dr. Macksey’s. But Coleridge was much more than a stuffy old poet, he was one of the leading intellects of 18th century Europe, a tour de force whose influence on science, politics, religion, critical theory, and many other fields can be felt even today.

At the highest level, Coleridge believed that ideas had a life of their own and that two seemingly unrelated ideas—say, from different fields—may actually share the same structure.

Take, for instance, biological immunity and Karl Popper’s paradox of tolerance in liberal democracies. They seem unrelated at first, but both are about systems (living beings and societies) that are open (ingest food and tolerate free speech) and yet must defend against intrusion (antibodies and political/security control). The threat in such selectively open systems is that the defense mechanism will attack its host (autoimmune disorders and authoritarianism), yet host needs the defense to thrive.

One isn’t an analogy for the other; understanding biological immunity as “like” liberal democratic tolerance or vice versa doesn’t help you understand either idea better. It’s deeper than that: both are basically variants of the same meta-idea. Coleridge called this tautegorical thinking, and it’s a way of understanding the history of every idea, from poetry to penance, from economics to eclairs.

Behind Coleridge is an even more potent idea, that ideas are formed by hidden structures that can be understood, a concept that underpins deconstructionist discourse. But unlike our media—whether it’s critical theory, social media, or the news—that dissects every idea into race, class, gender, or whatever category du jour, Tom and Dr. Macksey taught us the radical thought that we should seek what unites them.

That unifying spirit came naturally to all of us. For instance, while I was studying thirteenth-century Latin manuscripts for my thesis, I was also writing code and studying operating system architecture. People marvel that, after quitting my medieval history PhD, I switched into software engineering. But no one in that seminar would find that unusual; Tom switched from physics to the humanities, and Dr. Macksey originally studied medicine. As the Roman playwright Terence once wrote, homo sum, humani nil a me alienum puto: I am human, nothing that is human is alien to me.

Now imagine yourself sitting in that seminar at 20 years old taking this all in. Tom leads with Socratic dialogue through heady waters. Dr. Macksey occasionally takes the pipe out of his mouth to pierce the air with anecdotes about Picasso or biomechanics—or sometimes South Park (he was obsessed). The rest of us, hormonal undergrads huddled under towers of ancient tomes like a scene out of Harry Potter, were at the feet of the masters, deep in thought.

We weren’t just learning. We were being initiated.

Chekhov wrote a short story called The Student about a young seminarian who encounters two villagers, a mother and daughter, while walking back home on a cold Russian winter evening. He stops by their fire and recounts the Biblical story of how Peter denied knowing Jesus. The mother and daughter are both visibly shaken but don’t say why.

The seminarian bids them adieu, and as the villagers recede into the distance, contemplating their mysterious reaction, he realizes:

“The past…is linked with the present by an unbroken chain of events flowing one out of another.” And it seemed to him that he had just seen both ends of that chain; that when he touched one end the other quivered.

To be in Dr. Macksey and Tom’s seminar was to touch an unbroken chain of ideas and see it quiver. We held the chain. We even aspired to add a link.

Yes, we met inside of a fancy Hogwarts room, but it was never about the library or the pipe smoke. It was about the community of moths who were attracted to the same flame. We were shared a love of the idea behind Idea, whatever its name—Coleridge’s “tautegorical thinking,” Chekhov’s “unbroken chain.” Dare I suggest it: perhaps humanity itself.

It’s easy to get caught up in Coleridge, Chekhov, and Dr. Macksey. They’re larger than life figures.

Quietly, though, Tom was the one who went on long walks with us to discuss our research and share his innermost thoughts. He would take us to the restaurant on campus where his wife was a chef experimenting with locally grown ingredients. He invited his former undergraduate advisor to Baltimore, who later became one of his dissertation supervisors, to speak with us and share his work. Tom spoke with equal passion about the Colorado Rockies as he did about Plato’s Phaedrus, his impatience with the shallow takes of authors we read and his delight in reading texts closely.

I picked up Tom’s dissertation last night trying to remember those ideas that quivered years ago. What struck me more than the ideas was how he brought everyone around him into the intimate recesses of his intellectual life with curiosity and without judgement.1 It was his love of ideas and people, like Dr. Macksey’s, that wove us into a school of thought. Without that love, the unbroken chain that connects us all lies motionless.

Is AI intelligent? The question that everyone's obsessed with obscures a more important one: how do we create novel things with AI?

LLMs are information compression machines. They’re remarkably bad at novel thought, and their “reasoning” capabilities are more about introducing reflection points to enhance compression accuracy and depth. They’re terrible at operating from first principles and their responses, even when not openly sycophantic (“great insight!”), are dominated by whatever context you provide them in a prompt. But they’re incredible editors, encyclopedists, auditors, and so forth—they answer any question that has a right answer (usually correctly) and they easily identify gaps in well-known patterns.

LLMs also combine what’s known in ways that dazzle us, like a kaleidoscope. Websites that took weeks or months for a senior developer like myself to create can be spun up in less than a day. We’re so amazed that we don’t realize that this ability, too, is information compression, leveraging well-known patterns to create things that are new but not novel.

What makes something novel? I think it’s almost mystical. The scientist who discovers a new law in the natural world or the entrepreneur who creates an iPhone cross a threshold that’s hard to define but that we all recognize. They spend years compressing information in school and at work, learning through study and experience just to approach the threshold where they can ask a single novel question, one that might change the world.

In the age of AI, when information compression becomes free, instant, and reliable, the most vital skill will be knowing how to ask novel framing questions. I’m not talking about prompt engineering, or the important skill of knowing which questions to ask the compression machine. The bigger skill is knowing how to frame novel work.

What if expertise were free?

Imagine a fashion designer who wants to design a new nuclear reactor. Right now, they would need to go back to school for nuclear engineering, spend years working for others, and get licensed, all so that they could maybe get the chance to create a new design.

Once the fashion designer gets that chance, there’s some set of novel questions they or anyone designing a new nuclear reactor has to ask in order to create a truly groundbreaking design. Not everyone who’s in a position to create something truly new is good at it. In particular, those who are good at asking novel questions are usually the ones who eventually create novel things.

I posit that this ability to ask novel questions can be independent of your occupation (more on this later). The fashion designer might be just as good at asking novel questions as a nuclear engineer. No matter how skilled you are at asking them, however, you have to know a lot about a field just to get to the point where you can ask a meaningfully novel question to advance it. But if AI can quickly, cheaply, and perfectly compress all the info we need to get started, we can begin to see how the fashion designer whose real skill is novel question-asking could one day skip ahead and start designing a nuclear reactor right away.

The compression and novelty thresholds

Easier said than done. LLMs’ compression abilities are far from perfect right now. They can compress the lower layers of a given field’s knowledge pyramid—laws of physics, basic material properties—almost perfectly, but as you move up, the compression gets lossier and lossier.

For every field, there’s a compression threshold, which roughly marks the point at which LLMs stop effectively reproducing what’s already known, and the novelty threshold, which indicates where we cross over into the unknown. The shrinking gap between these two thresholds is where human information compression is needed—we call this “expertise.”

In some fields, like web development, the gap is almost non-existent. Almost any non-expert can create a high-quality and novel app or website today using Lovable or its competitors.

In other areas, however, the gap is still significant. When I tell my handyman that I’m tired of cockroaches bothering guests in my casita, he’s able to stand in the backyard and rattle off a vision for an entire hour for an integrated approach involving roofing, horticulture and architecture that’s infinitely more creative than anything an LLM can come up with. In law, medicine, and countless other fields—including nuclear engineering—the gap is shrinking but still meaningful. So our proverbial fashion designer probably couldn’t design a novel nuclear reactor today—eventually, however, I bet that they would.

“But information compression is part of novel thought. You can’t separate the two,” one might object. For instance, when I was in grad school for history, we memorized dates of events and the lengths of ancient Roman roads not for trivia’s sake but because knowing them spontaneously triggered new connections and made us better historians. Information compression and novel thought support one another inside of the brain.

But therein lies the skill, no? Somewhere out there, there are people who can do something that most of us would consider unnatural, to externalize that compression and still ask good questions.

Socrates, famously illiterate, decried the invention of writing because he thought that people would become stupid if they could outsource their memories. He saw intelligence and internal information compression as one and the same. He didn’t foresee that literacy would open up new modes of thought, something to which anyone who reads a book can attest.

The rise of the questioners

Who are those people, the ones who are good at combining AI information compression with novel questions?

They’re out there today, quietly working their jobs, tinkering with AI, already doing what I’ve described perhaps without knowing it. Consider the executive assistant who had to step in for a CFO who suddenly quit and is doing a weirdly good job, or a ChatGPT-wielding nurse who’s maybe overstepping their authority a little but has improved outcomes. Maybe they’re afraid of AI—they shouldn’t be. They should be impatiently waiting for AI to catch up to them.

Why should skilled questioners be restricted to one trade? If the cost of reaching a field’s novelty threshold is effectively zero, you could jump from field to field, synthesizing your learnings and experience to keep asking even better questions. Occupations that require credentials and licenses will be off-limits for a while—another reason why nuclear engineering is out of reach—but everything else is fair game.

Why aren’t people trying to become better at asking novel questions? Inertia is a big reason. The overwhelming majority of intellectual labor is information compression. We’re bad at it, yet it’s also comfortable; and that combination is what has historically provided the pleasant economic moat that we call middle class life. Our instinct is to cling to that moat rather than storm over and out of it.

Novelty is also hard. It makes our squishy brains hurt and there are few external incentives. Our education system at best rewards critical thinking, which improves compression outcomes. Our workplaces are a similar story, and commercial success depends more on execution over originality. We’re rarely rewarded in life for true, genuine creativity.

Crossing the threshold over and over

I’m cautiously optimistic that we could create a new economy based on the broad swathe of humanity asking novel questions. As long as there’s a novelty threshold to be crossed—and I don’t see any reason why there wouldn’t be—there will be demand for agents—currently humans—to cross it. The more we discover, the longer the threshold becomes with more opportunities to cross it again and again.

The idea that we might all become novel-question-askers only sounds insane today because of how painfully expensive it’s been to reach the novelty threshold. We’ve organized our entire identities around the friction of gaining enough knowledge to become a master plumber or a partner at a law firm. But as the compression threshold catches up and the gap between AI compression and novelty shrinks, that ego grip will have to loosen.

I recognize that I’m not as good at asking novel questions as I was at being a web development expert, but I think I’m getting better, and I’d especially like to learn from those who are doing it well. I’d like to create wonderful and weird things together.

I thought about titling this “What have I been up to this year,” but who can resist a bit of clickbait?

Since January 2025, I’ve been working with Mercor (pardon the shameless referral link) as a human data trainer and reviewer:

• As a trainer, I use my subject matter expertise in software engineering to help AI get better.

• As a reviewer, I use my empathy and my obsessiveness over fine details to help other trainers get better at making AI get better.

I wrote a Medium post about working in human data if you’re curious about what’s involved, but it’s a trending topic and I wanted to give my followers a more personal vibe for what it’s like.

The whole thing started as a side gig that I didn’t anticipate doing for long, a way of building up cash after I ran out working on my last startup. But now I’ve been wondering whether this might be a third (or fourth) career for me.

The most visible AI jobs are positions out of reach for mere mortals: the researchers in San Francisco and the AI engineers who are implementing the hard tech. I think that “human data trainer/reviewer” might be one of the first AI-native jobs that’s accessible to the rest of us all around the world.

I work on projects that involve large groups of people from around the world, all of us contributing our expertise. I don’t know of many gigs where you regularly interact with smart people from every industry all at once. Some projects are more narrowly tailored, in which case I’m more closely working with software engineers, but still with people from everywhere. Everyone’s smart and the work is super engaging.

The pay so far has been phenomenal. I’ve posted screenshots of my payouts on LinkedIn and people still don’t believe me.

Not everyone is being paid that much, sure, but it’s still top dollar for your experience level and industry. The flexibility is unmatched. I have dozens of people asking me every week on LinkedIn how to apply and whether it’s real. Many people think it’s a scam. (It’s not.)

What makes a career? I think there has to be an abundance of work, and it certainly doesn’t appear like Mercor or other human data companies are going out of business any time soon.

There also needs to be depth and progression, and here’s where it gets interesting. The more projects that I do with Mercor, the better I get at being a trainer and a reviewer. There’s a skill set involved that you hone over time, and folks gradually develop different work philosophies.

For example, I think that the cornerstone of my approach as a reviewer is empathy. I spend a lot of time trying to understand where the trainers whom I work with are coming from and their backgrounds in order to improve their contributions. I encounter so many people who are brand new to the world of machine learning and struggle with translating their considerable expertise and experience to the task of improving AI. I often play the role of an AI sherpa, bridging the gap. There’s a ton of soft skills involved.

Often I feel like I’m being paid to be an AI philosopher. I can’t go into details, but doing this job well on a daily basis means thinking very deeply about what AI is and what purpose it serves. It feels like I’m building a house but I don’t get to see what I’m building, I only get to see the final result along with everyone else. I have to have a strong sense of what’s the end goal to get there.

I did a cheesy little promo video that makes me look like a bit of a corporate shill or a wide-eyed naive AI booster, which is fine. (I’m neither of those things.)

I did it because I wanted to meet some of the people I’ve worked with in person. It’s weird, you’d think that working with a huge mass of fellow independent contractors everyday would be a deeply estranging thing, yet I’ve felt some of the most intense camaraderie doing human data work.

Probably the most frustrating aspect of what I do is that I’m sitting at the epicenter of so many important developments in AI, yet my NDA means that I can’t comment on very much.

One of those developments is the rise of human data providers themselves. Mercor is probably the most well-known and is certainly the highest-quality of the bunch, but there are others. I have an ambition to write more about this emerging field, but I’m not sure what to write or how to write about it.

That doesn’t mean that I won’t try. In the coming weeks, I’d like to write more about human data and AI. Human data as a niche inside of AI fascinates me because I believe that it points to the future of knowledge work, which will involve a lot of training and evaluating AI to make it do our bidding and eliminate repetitive work. I predict that we’ll all be doing aspects of human data work in our jobs before long.

That was my partner last night. We were lying awake in bed talking about Ecclesiastes, which has become my new favorite book in the Old Testament. It’s a text about a narrator discovering emptiness, something that I relate to.

“It’s like everything the author wrote could have come from your mouth, word for word, and they didn’t even know it,” she continued.

I asked, “Are you so sure they didn’t know?”

When I read an old text, sometimes I realize that I’m not alone. Someone else so perfectly captured what I think and feel that the line between myself and the other, which even in the best of times has always been arbitrary, looks blurry. If only for one transcendent moment, I sense that the author is literally here with me.

The eerie, supernatural thing is that author knows it, too. If I can feel their presence by reading an old text, don’t you think that they were able to feel the same thing when they read an old text? And if they could feel it looking backward, don’t you think that they could feel it looking into the future, knowing that one day someone like me would read them and feel just as they do? Just as one day, someone will read this blog post and feel my presence, and they will think of me just as I’m thinking of them. I will have known them before they ever knew themselves.

To read Ecclesiastes, or any other old text that deeply resonates, is to transcend time. When I read, I sense that we share the same consciousness. I see the divine chain that binds us all.

What could be more sacred than maintaining that chain? That was my worldview when I started my history PhD in 2010.

I saw what we were doing in the academy as holy, as the most important job that could possibly be done. We were the inheritors to the priests of ancient Egypt, keepers of transcendent knowledge and all that was good and important.

My colleagues didn’t see it that way.

I hesitate to recall my brief and painful stint in academia with any judgement of others lest I be judged. There are many things I remember doing and saying in my early 20s that I still cringe at almost 20 years later and individuals to whom, if I had the chance, I would apologize for some of the obnoxious and occasionally misguided things I said back then. I was rough character who needed a lot of therapy and didn’t know it yet. But my own sins, real or perceived, don’t cancel out the deep disappointment I felt going through the PhD.

Humanities departments in the 2010s were hotbeds of identitarian politics (for all I know, they still are). I remember a seminar in my first year in which everyone in my cohort went around and introduced themselves and why they were here. I explained, in all my naive earnestness, that I was here to seek the Truth with everyone else and that what we were doing was the most important job in the world. Everyone laughed at me, including the professor. They thought I was joking, but that’s how deeply I believed in our mission.

It wasn’t until some of my other colleagues spoke that I understood not everyone was here for such highfalutin reasons. A few mentioned that they were the first in their families to attend and graduate from college, let alone to start a PhD program. Fricking amazing stuff, bully for them. Genuinely, from the bottom of my heart.

But why did they choose to be here? More than once, I heard: to represent my people. To correct injustice. To be an activist.

Being an agent of progressive political change is a noble aim that I sympathize with, and part of being a good historian is to understand that there is no history without politics. But usually we would consider the politicization of history as a moral hazard of the profession, not as its end goal. I wanted to support my colleagues and their pursuit of justice, but how could justice or any political goal, no matter how laudable, be the supreme end—as I hoped was the case for all of us in a doctoral program—of someone dedicated to the fearless pursuit of truth without agenda?

A few of these same colleagues made outrageous accusations with impunity. One time, I remember being publicly called an exemplar of white privilege just for saying that I felt that the aim of studying history should be to maintain and perpetuate civilization. This was before I understood that “civilization,” an important concept in the anti-imperialist Turkish political vernacular that I was raised in, is a shibboleth on the American left for white supremacy. Not that anyone bothered to ask what my priors were before lobbing such attacks; that kind of compassion was a privilege only reserved for those who towed the party line.

Then there were the courses. We were commonly expected to read three to four thick academic tomes, each usually around 250 pages, per seminar per week, typically while taking four classes at a time. So about 4,000 or so pages per week of dense technical historiography—or about 133 hours per week at 30 pages per hour, which would be a generous pace—on top of teaching and researching.

There’s simply no way to ingest that much text that quickly. If we were really serious about our work, we would have been looking up every footnote, tracking down every primary source, and cross-referencing every argument. Each book would have taken an entire year to get through, let alone cramming 12-16 every week.

Predictably, no one actually did the readings, sometimes including the professors. We skimmed, caught a few catchphrases and trendy references to Foucault, and regurgitated them in seminar to sound intelligent. More than one colleague confided in me that they smoked weed before class just to get through it.

Outside the classroom, there was a pervasive bitterness that everyone felt about being a marginalized underclass, a self-identification as an intellectual lumpenproletariat. People were angry with the conservative—or fascist, depending on the day—politicians and administrators who kept cutting our funding. I was angry, too.

So when I showed up to a union meeting, I said let’s fight back. I proposed pooling our savings to hire a lobbyist and create a marketing campaign for the humanities, creating an outreach campaign, hiring billboards, grabbing TV spots, infiltrating the schools, pressuring politicians. If the system plays dirty, let’s cover ourselves in filth. Let’s dominate the system. Let’s develop our own political machine as humanities graduate students and make them pay.

My proposal was met with confused stares, and the meeting quickly moved on to discuss yet more injustices and grievances, how the academy had become subsumed into the capitalist system, yada yada. It took me years to realize that so many of my colleagues were engaged in a tribal status game. The point wasn’t to solve our problems, the point was to identify with them as bitterly as possible.

The nakedly identitarian politics, the unrealistic course load, and the self-identification as an underclass were parts of a broken social system. Instead of getting paid a reasonable wage, we paid ourselves in wages of moral status. The more we claimed to read and the more voluminous our output, the greater our glory, even though some days the only thing we could afford to eat was $0.50 day-old bread from Jimmy Johns. The poorer we got, the more we identified with the Marxian theories of our professors—the ones who showed up to class in a Corvette and earned six figures in corporate taxpayer money—even as we hoped that we, too, would tenure into the same class. We mocked the idea of a young historian pursuing Truth as dangerously naive, as we trashed our predecessors whom we hardly read for destroying truth with their grotesque biases.

This was the factory where history sausage was made. I was disgusted, and after 2.5 years, I quit.

Quitting the PhD tore out a piece of me.

There were some people there whom I’ll never forget. My advisor was a hero and I can’t imagine what stress I put her through in my early 20s. She didn’t deserve that and I hope that she forgives me. I deeply admired her integrity and I always will.

I had another colleague at a different institution who worked on my same research topic and whom I looked up to. When I told him that I was quitting, he was so generous and understanding. I’ve tried reaching out to him a couple of times in the past decade but I never got a response. Is he too busy—or did I hurt him in some way?

There were also a handful of professors who, I suspect, really saw me on a deep level, and maybe even saw that I was one of them—a dying breed who thought of education as a higher purpose unto itself. They were so kind and gentle toward me. I’m sorry if I disappointed them by quitting. And there were some fellow students with whom I still feel the embers of camaraderie even though we haven’t talked for years.

But the deepest rupture of all was the faith that I lost in what I was doing.

For years after I quit, I couldn’t bring myself to read any history. I joined the tech industry and I’ve been disturbed ever since at what I’ve seen: a culture that fawns over the intellectual prowess of leaders whose philosophies wouldn’t last a day in the rigor of a graduate seminar, who spout corporate slogans that would be deconstructed to shreds within 0.2 seconds.

Yet I couldn’t go back. Having my faith destroyed disabused me. There’s nothing wrong with history, but in the end, isn’t it just as empty as tech?

My younger self wanted nothing more than to be a high priest in history’s temple, a caretaker passing down the flame. I wanted to perform this sacred duty and enter communion with the dead. I wanted them to see me as I saw them, so that they would know that they weren’t forgotten, and I would know that I wasn’t alone.

As I look at the Bible in front of my keyboard, open to Ecclesiastes, I wonder: maybe I still could.

Part of it’s my decision to move to Tucson. Whenever I step outside of my house, I fall in love with this place all over again, but it took me years to appreciate that—at least when it comes to tech—I’m a big fish in a small pond here. I’ve tried for years to create big things in this town, and time and again I’ve been frustrated and disappointed, and I’m reminded: the people who leave are the ones who try to make a place fit to their dreams, and the people who stay are the ones who fit their dreams to the place.

Part of it’s age. I’m slower and heavier. I own a house. My testosterone levels are dropping. I have trouble finding the right buttons on the Instagram app. I live a comfortable petit bourgeois life that I never expected and that my younger self would have been ashamed of. I’m just not that hungry anymore.

But the real issue that’s plagued me for 12 years since I left grad school—and with it, my deep faith in my work—is what I call the Ecclesiastes problem:

“Emptiness, emptiness, says the Speaker, emptiness, all is empty…all things are wearisome; no man can speak of them all…I applied my mind to study and explore all that is done under heaven…I have seen all the deeds that are done here under the sun; they are all emptiness and chasing the wind.” —Ecclesiastes 1:2-14 NEB

In Ecclesiastes, wisdom is empty. Foolishness is empty. Even gifts from God are empty. To understand the message, you have to take in its texture as a whole: the breathless pace of a narrator searching hither and thither for answers, only to find that every answer is empty.

That’s because the world is empty.

Our culture associates emptiness with cynicism, nihilism, and depression. For me, accepting emptiness means freedom. I feel more liberated from the illusions of this world that keep us from true perception. As Zen tells us, just because something’s empty doesn’t mean it’s fake. The love and pain I feel for and from everyone, myself included, is very real. But to hold my feelings—or my dreams and ambitions—is like scooping up waves from the sea. As soon as they’re in my hand, they’re gone.

The plan since I left grad school has been to get rich in tech so that I could start my real life, which for years I imagined meant becoming an independent scholar. But I could never fully commit to tech culture’s empty narratives. Here’s a sampling that I’ve tried on, awkwardly, throughout my career:

• Technology is Progress, and Progress is Good

• Build what people want

• People don’t know what they want until you build it

• Make yourself useful and others will recognize your value

• Capitalism, with all its warts, is the best vehicle we have for making change

• Stay foolish

• Do it for the community

• You’ll find the truth along the way

• It’s all empty, yes, but what other game is there in town? This is how the world works, and if you don’t believe in something and push yourself forward, then you won’t feed yourself, and the world can’t feed itself, and we would be stuck dying from dysentery in caves, and surely you don’t think that’s better, do you? Someone has to do this work, so why not you, and why not try to enjoy it while you’re at it? Have fun with it. It’s fun!

None of these stories are wrong. To call them wrong is to become invested in them, which is the same mistake as judging them right. You could deconstruct and reconstruct the history of each word in each phrase a thousand times, coming up with millions of interpretations. You could find infinite examples to prove or disprove every maxim. But you won’t ever be able to pin the truth down, because there’s no “truth” there.

I always knew that if I was going to make it big, I had to have a deep faith in these stories, yet I could never muster that conviction. I still believe in the things that I build, but never for long. I consider that a blessing, because I feel freer from the illusions that so many tech people seem caught up in. But having my third eye open doesn’t help me achieve great things.

Nor do I feel at peace sitting still. I can’t seem to accept that I would spend this one precious life I have not shining as bright as I possibly can, not for the money or the status, but just to feel what Teddy Roosevelt called the “strenuous life,” to enjoy it juice, pulp, and pith.

I don’t know if there’s a career for me beyond tech’s golden handcuffs. It’s funny, no one ever went out of their way to compliment me on my engineering prowess, but I’ve had many people tell me that I have a gift for words and for teaching. I feel called to write, but the things that well up in my chest sometimes feel too controversial and vulnerable. I’m afraid of being hurt.

Whatever this liminal space I’m in might be, one thing gives me comfort. I refuse to ignore the emptiness any longer.

“Give them tasks that you know they’ll fail.”

I asked, what kind of tasks do you have in mind?

“Nuances with TypeScript. Type unions. Funky closures. Race conditions. Things that would stump a CS grad with a 4.0 GPA.”

I didn’t go to school for computer science so I don’t know what would stump a new grad. But I would imagine that it’s not anything technical.

The biggest thing that junior SWEs lack is knowing why to build anything.

Why do programmers build anything? Usually, because it’s a feature in a product that makes money. Our code turns electrical energy into money literally at the speed of light.

The bottleneck is rarely performance. It’s that we don’t know what makes money.

We can learn what makes money, but that costs money, and your org only has so much of it before it runs out. So the real question isn’t, “how fast does this code run?” Usually, the real question is, how fast can your organization learn how to make money?

Organizational learning velocity is the number one “metric” that most junior SWEs should index on and get good at improving. I put metric in air quotes because I’m skeptical of attempts to measure such a thing. I’m also skeptical that most orgs have this level of foresight. I think that it’s the right goal nevertheless.

The first way to help your org get faster at learning is to write as little code as possible. The more code you write, the more time you will spend maintaining and explaining it.

Sometimes writing less code later means writing more code now, like in the case where you replicate a little utility function to avoid adding a utility library as a dependency. Junior SWEs don’t know when to go with which.

The second way is to build like IKEA: write quality code, but only with as much quality as is required. Any level of effort spent improving performance beyond what is tolerable is wasteful and usually counterproductive.

If you’re refactoring loops with parallelization, remember that some poor schmuck six months from now is going to have to understand that parallelization. Some for caches, microservice orchestration layers, and any number of clevernesses. How long will it take that person to understand your clever thing? And will it distract them from building something else that could help the org learn how to make money?

Every once in a great while, if you’re lucky, technical performance really is the bottleneck. I say lucky because this usually means that you’ve grown to a certain point, which is not the norm. (Failure is the norm—of organizations and of feature releases—and you should optimize for that in your career and your code.)

The problem is that you can’t prematurely optimize for this unlikely bottleneck. But at the same time, you can’t completely ignore that it might happen, because if it does and you’re not prepared, then you’re really cooked. (Ever try sharding a Postgres db in production with millions of users once you’ve maxed out your cloud provider’s db capacity?)

So what do you do? You build in a way that’s easy to undo: modules, service boundaries, tests, APIs, etc. The easier it is to undo, the easier it will be to scale when and if you really need it.

I have a hard time going to coffeeshops where they know me by name because I end up wanting to party with the baristas, invite them to my wedding one day, and never give them an order ever again.

One time, I racked up tens of thousands of dollars in debt one time training a junior engineer fresh out of coding boot camp because I really believed in her even when I was fighting to pay my rent.

I once gave the shoes off of my feet on the BART to a homeless guy and walked into work barefoot in the rain. He had painful pustules all over his toes and he was crying aloud, begging that someone give him shoes on a crowded San Francisco workday train at 8:30 AM. Some other guy with Italian leather loafers who was tired of the homeless dude’s cries told him “no one’s going to give you their shoes, so just shut up!” How some people can be so fricking heartless, I’ll never understand.

Not only do I love people so much, I’m tired of pretending that I don’t. So I’m announcing what I actually, truly want:

I want a cofounder. Someone with as big a heart as myself. I want to go into battle with them and fight to the death for them. I want the glory, the power, and most of all the love.

I want an insanely ambitious team of metaphysical misfits, voices in the desert. You don’t have to be in the Bay Area or dress goth or whatever. Show me that you have a big heart, attention to detail, that you’re a little weird, and that you want to fight.

I want engineering managers who, in another life, might have been therapists. I want HR departments that aren’t just CYA factories but that actually work to help and improve people and who remove bad actors, even when they’re economically important.

I want investors who can be a real, genuine friend in a founder’s darkest times. I want influencers and dev rels who talk about alleviating the brutal frustration that is sometimes coding. I want ICs who feel as comfortable writing an AST parser as they do opening up about their fears and aspirations.

I want a hometown, Tucson, that supports me and my ambitions, and that I can pour all my love back into. I want to have kids here. I want to raise people’s voices and help people join this amazing industry called tech for which I’m so grateful.

Some of what I’m describing goes directly against what’s considered corporate best practice. I really don’t give a f— anymore.

I’ve been working on Scrappy Cash for six months like a hermit, slowly dying inside, convinced that I needed to have a measure of success before anyone would follow me into battle. Now I see that it’s the opposite: build the team first, and then the product. And how can one build a team other than by being exactly oneself?

I’ll leave you with the wise words of Creed, who put on an incredible performance last night in Phoenix. (Man, talk about a misunderstood band.) Never forget that the greatest prison is the one in our own minds.

I went to RAPID + TCT, the world's biggest 3D printing conference, and told everyone about my idea. A few were supportive of me as a newcomer to the space, but no one liked my idea. I was ridiculed and some folks were even aggressive. They told me that 3D printing had just gotten out of B2C and was finally making 🤑🤑🤑 in B2B and enterprise—why would I want to drag the industry back?

Well, because I believe in giving power to the people. That's my dream. But I let myself be discouraged.

I packed up my prototype and my workshop into some dusty boxes and I gave up.

While I was asleep, a new startup called Bambu Lab came in and made exactly what I was working toward. Their sub-$1500 printers beat the crap out of much more expensive machines. They're fast, beautifully designed, and they "just work."

So many of the things Bambu has incorporated were exactly what I was working on: CoreXY geometry. AI detection of bad prints. A heated enclosure with built-in air filtration and filament drying. A slick integration between their software and hardware. I had the core intuition three years ago that all the tech we need to make a 3D printing home appliance was ready, we just lacked integration. Apparently, so did Bambu.

Instead of having created the Apple of 3D printers, I'm about to order one. How very bittersweet. Kudos to the Bambu team. Creating this machine was one of the great dreams of my life.

Paul Graham was right: don’t give up.

I will not repeat this mistake.

I’ve spent the past month doing a deep dive into LLMs. I went from being deeply afraid of generative AI to becoming a true believer.

And let me tell you, developing LLM-based apps sucks.

There are lots of great startups working on making different parts of the process suck less. For example, the fine folks at Brev.dev, who just launched their v2, are solving for finding GPU compute and provisioning.

In my opinion, however, the greatest pain point is that we haven’t created effective idioms and patterns for AI-based app development, which I wrote about before. Nor am I the only one:

We need a dead simple way for developers to make AI apps, one that recognizes that we’re in a new, post-classical programming world.

We need Next.js for LLM apps.

Why OpenAI and Replicate aren’t enough

Currently, if you want an easy-ish experience, you use OpenAI or Replicate. They provide easy-to-use APIs that don’t require you to know anything about AI. You can use in-context learning or fine-tuning to get inferences that work for your app.

In the process, you pay an arm and a leg for access to a black box. There’s no guarantee that the black box will keep working. You’re not allowed to send any personally-identifiable information.

Most importantly, you can’t build proprietary equity. Models are becoming the new data, the secret sauce that makes your company valuable to investors and differentiates your product from the competition. When you rely on someone else’s AI PaaS, your locking yourself into their proprietary system without any easy way to transition if poo hits the fan.

Why Hugging Face and Llama aren’t enough

Many developers, recognizing the above, turn to Llama and Hugging Face to fine-tune models and create products. The problem with this approach is that it’s far too low-level.

Seriously. Look at this API for fine-tuning from Hugging Face’s TRL:

dataset = load_dataset("imdb", split="train")

peft_config = LoraConfig(
    r=16,
    lora_alpha=32,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM",
)

trainer = SFTTrainer(
    "EleutherAI/gpt-neo-125m",
    train_dataset=dataset,
    dataset_text_field="text",
    peft_config=peft_config
)

trainer.train()

To even begin to approach understanding what is going on in the above code, you need to learn what supervised fine-tuning, LoRA, and PEFT are, never mind how to “un-PEFT” your model from checkpoints once you’re ready to run inferences. And even then, we’re nowhere even close to making this code production-ready.

Adding LLMs to your app shouldn’t involve learning AI research. It should just work, and it should be exposed to a reasonably experienced developer in a way that they could tweak their model expressively without any more effort than learning React or Svelte.

The vision

I want to take this crazy stack:

and simplify it into this stack:

Key ideas:

• An open source framework like Next.js for building LLM apps. This means building React for LLM apps first: a framework that encapsulates opinionated best practices for handling data flows and deriving user-visible state from them. It introduces a vocabulary of concepts—like “hooks,” “update lifecycle”—that devs can use to expressively reason about LLMs, only exposing underlying ML theory when it delivers business value. Then we build a meta-framework, similar to Next.js, to allow for creating a complete app.

• Streamline logging and orchestration, making them understandable to devs. The gold standard for logging is currently Weights and Balances, which requires becoming an ML engineer to interpret correctly. Let’s creating logging that allows devs to get useful insights without AI expertise. Let’s also create simple orchestration—more Heroku Elements than LangChain—that enables combining LLM providers through a plug-and-play interface.

• The user owns the model, always. They can always export their model to another provider, and we never use their data to train our own models. Their model is their business, their model is sacrosanct.

What do I need to do this?

More like, what we need to this. I want to do this, but this is a crazy ambitious project that’s going to take many hands.

I need a team! A cofounder who’s passionate about using AI to improve the future and who cares about helping developers in this new, post-classical programming world would be a great start. I can build hella developer experience but I need folks who are great at marketing and/or really know what they’re doing with LLMs.

I need funding and mentorship! I will apply to Y Combinator for a third time. Their deadline for the winter batch is in a few days. Too bad I have scarcely more than an idea, but maybe if I build in the open for the next few days it will attract the right energy from the universe? Perhaps an angel?

I think in the meanwhile that I will try to make a small product with better logging for training models to validate at least part of the concept and charge for the service.

Maybe someone is already working on this. Maybe we can join forces.

If you can help, or if you can use my help, contact me: me@ersinakinci.com.

What do I call it?

Working title is PricklyPear.ai (bought the domain, so don’t even try 🤠). Because Atacama, a desert where llamas live, is apparently already taken.

As of writing, it’s way still way too hard to wrap your head around what is going on with AI—specifically LLMs—if you’re a programmer/developer/software engineer who wants to use AI in your products. In fact, we’re in dire need of paradigms to bridge the gap between high-level APIs like OpenAI and Replicate’s web endpoints for fine tuning models and Hugging Face’s libraries.

Before we can make those paradigms as a community, though, we need better educational materials for app developers to understand what’s going on in the large language models (LLMs) that we use. Personally, I think acclaimed courses like Andrej Karpathy’s YouTube series and fast.ai still require too much time to go through.

I’m going to give it my best shot on this blog, explaining what I’ve learned about how AI works with fellow developers in mind. I assume that you’re already a programmer and that you have some minimal experience with LLMs, even if it’s just having played around with ChatGPT.

By the end of this series, you should have enough of a foundation in current AI systems, particularly LLMs and reinforcement learning (RL), to confidently be able to use higher-level libraries like Hugging Face’s Transformer Reinforcement Learning (TRL) to create and deploy AI-enabled apps.

Key terms will be bolded and italicized.

Note: I’m not an AI researcher, I’m mostly writing this stuff as I learn it, and I’m mostly using ChatGPT to tutor myself. If you find something wrong, please let me know in the comments! Keep in mind that I’m simplifying certain concepts to keep it appealing for developers.

Introduction

What is an AI?

The number one thing that you need to know about any AI system is that it’s a prediction machine.

The entire purpose of AI is to predict what comes next. For example, an AI chess player predicts the next move. More abstractly, we can say that an AI like ChatGPT predicts the next word in a conversation. An image-generating AI like Midjourney predicts what’s the next pixel as it constructs an image.

For clarity, the community uses these technical terms when describing this prediction process:

• The “AI” is called an agent. ChatGPT is an agent.

• The thing being predicted is an action. Agents generate actions, one after the other.

This terminology is intuitive for agents that visibly do things, like play a game of chess: each move is an action. But more abstractly, you can think of any prediction as an action. Each word generated by ChatGPT and every pixel created by Midjourney is a “move.”

Agents always move toward a goal, and this is very important. No matter how human-like an agent appears, the most fundamental distinction between us and it is that every action leads toward a certain goal.

Put another way, the purpose of an agent is to execute tasks that can have a successful end state:

• A chess-playing agent’s task is to play a game of chess that results in checkmate.

• An image-generating agent’s task is to create an image that represents a prompt.

A chatbot’s task is to…hmm, does it have a task? Does it have an end state? There isn’t an actual end state since you can, in theory, have an infinitely long conversation with a chatbot, so we say that the chatbot has a continuous task. But for the purposes of teaching a chatbot how to speak, we feed it conversations of limited length that in fact do end, forming a kind of imaginary end state. More on this later.

The opposite of a continuous task is an episodic task. You can think of a single game of chess as an episode for a chess bot. It has a logical end with a well-defined successful end state.

One last foundational piece: just as task has a successful end state, every agent has a certain state at any given moment. State represents the agent’s environment, position, or general orientation:

• A chess bot’s state in the middle of a game includes the current position of all the pieces on the board.

• An image-generating agent’s state in the middle of creating an image includes all the pixels it’s painted so far.

• A chatbot’s state in the middle of a conversation includes all of the words written so far in the conversation, both by it and the user.

Put in a more rigorous way, state includes all the data an agent needs to make the next action. It represents all progress that an agent has made so far in completing its task. AI agents, like humans, take their next action based on everything that’s happened up until that point.

At this point, we can update our answer to “what is an AI”:

An agent, commonly referred to as an AI, is a program that generates actions in order to complete a task. Specifically, it generates or “predicts” one action at a time with the aim of achieving a particular goal.

If the agent is working on an episodic task, then the task has a successful end state. Reaching that end state means that the agent as achieved its goal.

There are also continuous tasks, meaning tasks with no predefined end (we’re not sure yet how this fits into our mental model of AI).

At any given moment, an agent has a particular state. Each state represents all the info about the agent’s environment and its orientation in that environment—which often includes or implies info about every action taken up to the present—needed to take the next action.

But given a certain state, how does an agent figure out what the next action is?

Policies and rewards

Agents determine the next action using a policy, which means “a strategy for completing a task.”

More formally, we can define a policy as a mapping between states and actions:

• Given a certain state on the chess board where some of the pawns are out, a knight is on a certain square, white’s bishop is threatening black’s queen, etc., the chess bot should move its rook to D2 to get closer to checkmate.

• Given a certain set of pixels with certain colors that forms a fuzzy image that kind of looks like a koala, the next pixel generated by Midjourney should be rgb(233, 242, 79) at x = 27, y = 139 to get closer to the finished koala.

• Given a certain set of words that comprises a conversation up until this point, a chatbot should say the word “fries” next to get closer to…more on this later.

As humans, we intuitively know what to do in a given moment. You could say that we’re constantly updating “policies” in our head and acting on them—not that the brain actually works that way, but we can pretend for the sake of argument. Our reasoning is opaque and ranges from “this is rational” to “this feels right” to “I dunno.”

AI agents in contrast have a concrete measure of why certain actions are better than others given a certain state. That measure is called reward, which is stored as a number (let’s label reward as r in the diagram below):

You can think of reward as points. The agent gets more points if it takes an action that brings it closer to its goal and it gets less points for actions that take it further away.

Policies map states to actions in a one-to-many relationship. You can take any number of actions at any given point, and some are better and some are worse, just like how there are great moves, OK moves, and crappy moves on a chessboard. The better moves have greater rewards associated with them and the worse ones have lesser, even negative, rewards. A policy that always takes the next action that has the greatest reward is called greedy.

In its simplest form, a policy can be stored as a table, which you can generalize as a function: given a certain state, tell me how much reward I get for every next action that I could potentially take. A greedy policy, which is a kind of deterministic policy, then tells the agent to take the action with the highest reward. It’s like a chess player who only knows one way to play.

In contrast, we can also have policies that are more nuanced, perhaps ones that intentionally select actions with lower rewards from time to time. Stochastic policies, for example, choose an action based on a probability. Perhaps the policy is programmed to pick the action with the highest reward only 60% of the time. The main reason you would want to do this is to introduce variety—into the chess bot’s strategy, into the image generator’s painting, and into the chatbot’s conversation.

Regardless of whether our agent uses a deterministic or a stochastic policy, it fundamentally decides on a course of action based on reward. The only difference is how the policy relates to the reward (always go for the highest reward? Sometimes go for the highest reward?)

But how do we know how much reward is associated with each action for every given state? Stay tuned for part 2, where we delve into how AI systems learn rewards.

The year is 2023, but we’re in a 2010ish moment when it comes to AI and app development.

2010 is when Backbone.js, the groundbreaking web framework, was first released. From that moment forward, we saw a rapid evolution of web frameworks but more importantly programming paradigms. Functional reactive programming (e.g., React, Redux) eventually won the day, but we had to go through several alternatives—for example, Angular’s two-way data binding—to get there.

We’re at the very beginning of going through those paradigms today when it comes to building AI-powered apps. It’s clear that AI will change everything about the software engineering and product process, from the interfaces we create to our infrastructure to the way we write code.

However, there’s a yawning chasm right now between AI IaaS providers like OpenAI and Replicate, which offer opaque APIs for inferences and fine-tuning, and fine tuning your own Llama 2 instance with Hugging Face libraries. The former is too expensive at the moment and too much of a black box for many applications. The latter is too close to AI research, requiring devs to learn lots of machine learning fundamentals just to use the library.

AI programming paradigms are the missing middle. What’s going to be the reactive programming equivalent that enables the React of LLMs? We don’t know yet.

The first step to getting there is to give developers a set of new “mental primitives.” We need to acknowledge that we’re transitioning from “classical programming” to “AI programming” and that the two are very different. In other words, we need to develop idioms, shorthands, and a way of reasoning around AI app development that’s distinct from how researchers talk about AI/ML.

This conviction has grown out of my frustration with the standard recommendations for devs who want to learn AI, such as Andrej Karpathy’s YouTube videos or the fast.ai course. The barrier to entry is still too high.

Instead, we need stuff closer to Brev.dev’s straightforward guide on how to fine tune Llama 2. The problem there is that when something goes wrong—like when after fine tuning your model still doesn’t provide the desired output—you have to get your hands dirty with Hugging Face TRL’s APIs, and those APIs are still too low-level for the average developer.

Another analogy is to look at SQL databases. The information theory behind SQL and all the things that make Postgres work is way above my head. Yet as a web dev, I intuitively understand concepts like atomic commits and ACID. More importantly, concepts like ACID enable and inspire higher-level paradigms like CRUD, the bread and butter of many a app/web developer’s career.

I’d like to explore potential paradigms in this blog as I learn more about AI and machine learning. Hopefully others will find them useful. Consider this post my clarion call 📯.

“Sure, San Francisco people are ambitious, but when we recruit locally, we lose sight of how everyone else is.”

That’s the view, roughly speaking, of a San Francsico tech startup founder. Specifically, it’s the view of a founder I became friends with over the past few days as we soft-pitched each other the idea of me relocating from Tucson, my home for the past three years, back to Silicon Valley, where I lived for five.

But why did I, as a software developer and startup entrepreneur, leave the center of the tech universe for this godforsaken desert in the first place? It wasn’t for lack of ambition.

I want to address my new friend, and I want to address other entrepreneurs and engineers who might feel like something is missing in their lives. As much as I love San Francisco, what I learned out here is so much more valuable to me as an entrepreneur—and as a human—than what I could find when I lived in tech mecca.

This is the story of how I ended up nowhere.

It was the summer of 2017 and my life was falling apart.

I was just another sexually-repressed, overpaid 20-something drunkenly roaming around the streets of San Francisco, like so many other software engineers I knew.

I made six figures as a Senior Software Engineer at Instacart, a company that I quit before I could vest, so their recent IPO means jack squat for me. Today, that company has a halo: recruitment-happy founders’ eyes go wide when I mention that I worked there. As I look at photos from their recent Nasdaq IPO event, all I can think is, “I used to get shitfaced with these guys in a supply closet.”

It wasn’t until this year that I came to terms with how toxic Instacart’s culture at the time was for me. There, I met some of the best people in my life, and they thrived. Yet my anchor memory from my time there was spending months launching a major new feature and being screwed over by the product manager whom I was paired with. My leaders were helpless to protect me and the psychopaths seemed empowered.

My personal affairs that summer were even worse. I desperately clung onto a marriage in crisis, partly out of love and partly out of fear, immiserating myself in the process. I was also at my deepest point in therapy, working through intergenerational trauma. My relationship with my family ruptured soon after and never recovered, which fills me with melancholy to this day.

On top of all this suffering, there was the disdain I felt for my community of Bay Area tech people. I carried a caricature of them in my head: hapless, delusional, confused, and coddled. Moneyed babies running around with scissors pretending to be deep thinkers with no sense of history. I imagined they would gaze doe-eyed at the mushroom cloud once the world they wrought went boom-boom.

Years later, I realized that my tribe isn’t so bad—in fact, we’re pretty great. Some of them are the humblest, most thoughtful, and most genuine people you’ll ever meet. A few showed me great kindness during that dark time. But I don’t think any measure of kindness could have saved me back then.

I was isolated in a dark hole surrounded by misery at work, at home, and in the street. Somehow, I had to change my life.

Ersin. Ersin. Ersin.

I came up with a brilliant idea that summer: a roadtrip to Austin to clear my head. I went there once as the captain of my middle school robotics team to compete at nationals, and I remembered thinking, “this place would be way cooler as an adult.” Also, many Bay Area tech workers seemed to be moving to Austin for warmer weather and better vibes. Maybe I would discover something special there, too.

San Francisco to Austin is too long for one sitting, so I’d need to break up the trip. I turned to Paul Simon’s Under African Skies:

I said, “Take this child, Lord
From Tucson, Arizona
Give her the wings to fly through harmony
And she won’t bother you no more”

Tucson. 14 hours away, the perfect first stop. I didn’t need anything but a childhood song to hop in the car and disappear.

The first Tucsonan that I met was my Airbnb host, Jenny, who also became my first friend in the city. She had a parakeet named Nacho who took gentle poops when perched on your finger and she played harp in a Celtic punk band. Jenny’s husband, an opera singer with a resonant baritone voice (who turned out to be an excellent human being when I met him years later), was away on tour. I was worried that Jenny might feel put off by this single male straggler sleeping down the hall in her home.

Nothing could have been further from the truth. Even though we had just met, Jenny made me feel more welcome than almost anyone back in San Francisco. We stayed up into the small hours that first night drinking beer from Borderlands Brewing and talking about everything: politics, love, money, video games, the desert, finding friends in one’s 30s.

That’s when I started to unravel. Suddenly, I became aware of just how exhausted I was, followed by what felt like my soul exhaling a sigh of grief.

I opened up to my host with an epiphany:

“I compromise on everything in my life,” I confessed. “I compromise on my job, I compromise on my woman, I compromise on where I live…

“I even compromise on my own name. I introduced myself to you as UR-sin. That’s not how you say my name.”

For the first time in public, I introduced myself using my name in its original Turkish pronunciation: “It’s Ersin. Ersin. Ersin.”

Those kinds of realizations happen a lot in Tucson. This place is an authentic conversation with a new friend while her parakeet gently poops in your hand, reminding you that you’re alive. Here, you can let go of the six-figure salary, the billionaires exhorting legions of programmers to “change the world” through $9/mo. SaaS subscriptions, and the warehouse rave revelers claiming enlightenment, lost in the dark. No one here cares about any of that. Neither do I.

I care about losing myself with companions among the scorpions and saguaros on a moonlit hike through Sabino Canyon. I care about embracing my best friend when he comes to visit from Berkeley as we scream into a monsoon on the top of Mt. Lemmon. I care about the 24 year old video game devs working 9 PM to 5 AM, the ones who pop over the backyard fence when I’m throwing a rager. I care about my sexy oddball girlfriend as we belt Bohemian Rhapsody in the morning and dance all night.

Tucson is the first place where I gave myself permission to build a life around these moments.

Coming to Tucson

Tucson was meant to be a pit stop on the highway, but I loved it so much that in 2020 I moved to the city, which is where I’ve lived ever since.

As much as I love it here, leaving San Francisco for southern Arizona as a tech entrepreneur was, how shall I put this—a quixotic choice. Some people would call it career suicide. Silicon Valley has all the talent, finance, and culture. Tucson has lots of…cacti. I found personal fulfillment here, but what about work?

In my time here, I’ve discovered that the most fundamental thing as an entrepreneur is to know why you’re working on something. The “why” boils down to what you value, and you don’t get to pick what you value because it’s simply who you are. To be a truly effective entrepreneur, you must embrace who you are and build your business around it. And to embrace who you are, first you must recover yourself.

Recovering ourselves is what “coming to Tucson” is about. Trust me, it isn’t easy work. My life came apart, not once but a few times. It took many missteps before I could claim the love that I deserved from friends, community, and a romantic partner. And it took several failures to finally claim a vision for what business means to me: taking care of the people I love, and building the most exquisite, most joyful, most wonderful little products you’ve ever used.

Let me be explicit. I don’t care if I never scale to a zillion MAUs. If you work with or for me and you feel genuinely cared for; if you pull up one of my apps and feel the love and care; if you can’t stop fiddling with it, like a beautiful pen or a perfectly springy pair of pliers in your hands: then I’ll know that I’ve done my job.

But my values might not be the same as yours.

I heard that when Dustin Moskovitz launched Asana, his goal was to make everyone in the world 1% more productive. Good on him. Sounds like his Tucson is actually San Francisco. For others still, their Tucson isn’t even a place, it’s a mindset.

The point is that the most important work we can do as entrepreneurs is the inner work to become whole. Then the business isn’t about business anymore, it becomes about living. Which is exactly what we’re here to do.

It was January 2013 and I had just dropped out of my PhD program.

It was a shit start to a shit year. I was 25, a bit of a jackass with no friends, and had about $400 to my name. I moved in with my future ex-wife into a postage stamp-sized apartment in the Chicago suburbs which, after years of grad student living, seemed excessively palatial, even though we could barely squeeze through the bedroom door. And I didn’t know it then, but I was about to develop a recurring caffeine-induced pain in my left ear that feels like your friendly neighborhood ballista-sharpener installing a tooth implant.

The only thing kept me warm those winter nights was a dream: to become so rich that one day, I could become an independent scholar free of academia’s politics. Unshackled from my colleagues’ small ideas, I would revolutionize the world! I would become the next Karl Marx! (But rich!)

It wasn’t until much later, when I had deconstructed my entire identity in therapy, that I saw that I was suffering from intergenerational trauma-induced delusions of grandeur, commonly treated with alcohol. It took even longer for me to be diagnosed with ADHD by a dodgy online nurse practitioner. To be fair, it wasn’t that she herself was dodgy (she was also from Arizona, not that that made her less dodgy), rather it was the business model of the app for which she worked that was dodgy and that probably incentivized her to diagnose me with ADHD. I should rephrase: despite the questionable circumstances of my diagnosis, I definitely have ADHD, and I definitely should not have the meds that are helping me write this article taken away from me. As the meandering nature of this article—or blog post, or naked ploy to get you to subscribe for $7/mo.—should attest, I likely have ADHD and, by the same token, am ill-suited to a scholarly life spent patiently scrutinizing ancient manuscripts in the Vatican Secret Archives (which I did one summer, but I’ll write about that later).

Where was I? Right. Alcohol.

You see, it was January 2013 in Chicago. I had just quit grad school, and although I was drinking myself under the La Salle St. bridge, I had big dreams. I would become independently wealthy by somehow combining my PhD dissertation research on Ethiopian travelers to medieval Europe with programming, which I learned when I was younger.

Naturally, I set my sights on the most lucrative form of programming there is: indy video games. I would make a game about Ethiopia’s untold critical role in the development of modern science.

The idea didn’t seem far-fetched. I grew up on hardcore historical simulators like Europa Universalis and Hearts of Iron, but they were for PC. It was 2013 and no one had made such a game for mobile yet. What if I cornered the market? And if not me, this odd mashup of humanist scholar and software engineer, who else?

This is how I took my first farcical steps toward becoming the middling tech entrepreneur that I am today.

The logo

My first step was to do some market research and make a smoke test. It’s important when writing software—when building any product, really—to figure out whether anyone wants your product before you actually build it. I started reaching out to key influencers, working my way through Steam Communities and gaming subreddits…

Just kidding. Obviously, the first thing I did was to pay my long-lost Austrian half-cousin $200 for a logo. You can have an album without any music, but you can’t have a band without a name and a logo.

My new company’s name was Pendulum Gameworks, named after the proverbial pendulum of history. A naive American freshman who takes a US history course in college and learns about their country’s crimes for the first time turns into a little Malcom X—swoosh, that’s the first swing of the pendulum. But as they learn even more history, they learn nuance. They learn to deconstruct the deconstruction. They learn the biases of those uncovering bias. Swoosh—that’s the sound of the pendulum swinging back. Learning history, I figured, was a process of swinging one’s worldview back and forth until the pendulum settled somewhere in the center: Buddha’s middle path. My educational games would be a road to enlightenment.

I had my half-(or step-?)cousin, a talented designer by the name Günther, make the logo. At least, I think it was Günther. That may be his DJ name. Or maybe his brother designed the logo, which looked like this:

I’ve never met Günther or his brother. The only reason I have distant Austrian relatives is because my grandfather, a Turkish Air Force pilot, had an affair with a US Air Force colonel’s wife while he was stationed in Biloxi, Mississippi in the 1950s. When his posting was up and he was recalled back to Turkey, the poor woman apparently wanted to come back with him. To which he said, “touch grass!”

Fast forward a pregnancy, a couple generations and some transcontinental sojourns by one very mystical half-(or step-?)aunt, and I’m Paypal’ing her kid Günther $200 USD for a logo that will only ever be seen by my parents, my ex-wife, a few hundred Chicago-based Ethiopians, and now you.

Unfortunately, I don’t have the emails that I sent Günther anymore. Below is the most faithful reconstruction that I could patch together from memory:

Hi Günther,

Thank you for agreeing to make this logo. Here’s some background that might help you understand what I’m looking for:

1. History is but a pendulum that we all succumb to. It swings hither and thither, and with it, the very nature of our minds change. I want to capture the essence of this change in the logo.

2. It’s not just that history enlightens us by helping us see wiser (or more correct? ha!) positions to guide us through life. I want to emphasize that it is the very nature of the positions themselves that changes through the process of learning history. My gamers will go through so many life changes as they play these games, and this logo will be the entrepôt to this new lifestyle.

3. Obviously, I trust in your judgement as an artist above all else. Please feel free to modify or disregard any of the above or below points if you feel that it contradicts your aesthetic spirit. I want you to feel empowered as an artist, and I want Pendulumers to feel empowered by your spirit.

It went on like this for several more points, as I recall. Günther was a total pro and replied “OK no problem, give me two weeks.”

The logo wasn’t anything close to what I wanted, but honestly it’s a really solid logo and better than what I deserved.

The research

As I waited for Günther to finish the logo, I started sketching out a game based on my dissertation research, which was on Ethiopian visitors to medieval Europe from 1200 to 1500. It boiled down to this:

1. Some ancient and medieval European theories and models suggested that the world outside of Europe was uninhabitable and undiscoverable.

2. One of these theories, the torrid zone, said that the equatorial belt was too hot for human life even though Europeans traded in and eventually colonized those regions. Yet textbooks and treatises as late as the 1600s still talked about “the torrid zone” as a thing.

3. Not only did Europeans go to the torrid zone, but torrid zone dwellers came to Europe—for example, Ethiopian ambassadors to European courts, well before the European colonization of the tropics.

4. How could Europeans have ignored the fact that they lived in and knew people from the torrid zone? And what does that say for “scientific progress” if we’re able to collectively ignore super obvious facts right under our noses for hundreds of years?

I never did answer my questions in grad school, although my colleague Nicolás Wey-Gomez pretty much nailed it. Instead, I decided to turn this jumble of thoughts into a video game.

I had a nice thought: what if instead of some random white-passing Turkish kid making a game about Ethiopia, a country I only knew from academic research, I got some buy-in and thoughts from Ethiopians instead? Maybe I’d even find a co-founder.

One Google search later, I found the Ethiopian Community Association of Chicago (ECAC) . I called them up and spoke with one of ECAC’s leaders; let’s call him Abel. I told him that I was making a game about Ethiopia and I wanted see whether the community would be interested in supporting the project.

Abel’s interest was piqued. He invited me to come talk about it in person.

Take me to church

I showed up to my meeting with Abel totally unprepared. You’d think that I’d have thought of something on that interminably long train ride from my apartment to Rogers Park, the remote Chicago neighborhood where ECAC is located.

Here’s what I thought about instead. When you live in Oak Park, a leafy suburb just 15 minutes from downtown, every insufferable hipster technically within city limits will cut you down to size if you claim to live in Chicago, too. Even though they might live 50 minutes away in some godforsaken studio out in Rogers Park, they will make you suffer.

I got over my pissy mood and put on my game face when I arrived at ECAC. I was here to discuss business. Maybe Abel could connect me with other scholars or programmers, someone who could grow the business. Or maybe—just maybe—we might discuss financing for this worthy cultural project. The word “Kickstarter” was ringing in my ears. I was ready to hustle.

But Abel, who was a gentle middle aged man with kind eyes, had a different plan. He invited me to see their church, which was part of ECAC’s building.

We walked from his office on the second floor down a staircase to the basement. He opened a door and I followed him through.

Before me lay a cavernous warehouse that had been converted into an Ethiopian Orthodox church, hidden in plain sight. Large industrial fluorescent lighting tubes illuminated enormous prayer rugs on the dark gray concrete floor.

“This used to be AT&T. Now we pray here,” he told me with a smile.

This was a place of refuge hewn from America’s rustful past: Lalibela on the prairie. It was an honor and a sign of trust that he let me see this magnificent place. I felt disarmed and a little naked. I felt ashamed of how eager I was to hustle this helpful hand. My host was on a different level, showing me that the best business is never a hustle. The best business is barely even business.

After getting back upstairs to his office, Abel called in a surprise guest: an Ethiopian Orthodox bishop, who greeted me kindly and asked about my research. I felt the room begin to glow softly as I opened up, warmth meeting warmth. I was hungrier for more something deeper than money.

I was hungry for the love that they poured on Donald Crummey.

Dr. Crummey was an illustrious scholar whom I had the honor to have met shortly before his death. He was the doyen of Ethiopian studies, lord of his field, and somehow I had convinced him to write me a letter of recommendation for my dissertation research.

Every Ethiopian I talked to seemed to know and love Dr. Crummey, or Don as he was affectionately known. Invoking his name unlocked doors at the embassy when I applied for a visa for my dissertation research, and it unlocked smiles at ECAC as I mentioned that he had blessed my research.

“Don!” Abel and the bishop delightedly exclaimed.

Growing up, we had our own Donald Crummeys in the house. My father admired American scholars like Cornell Fleischer—or, more unfortunately, the Armenian Genocide-denying neocon Bernard Lewis—for their dedication to Turkish and Ottoman history. These names were talked about with respect at the kitchen table. But I have yet to find an equivalent to Ethiopian culture, which to me seemingly combined a deep and broad reverence for scholars with a first-name affinity for them.

I had quit grad school because I felt misunderstood, unappreciated, and miserable. But maybe, if I did right by the Ethiopians, they might love me as Don Jr.

A stroke of genius

Riding on this high, I started talking about the idea for my game. I wanted to make it for phones so that everyone could play and especially so that young people could learn about Ethiopia’s history.

That’s when I had a stroke of genius:

“You know, I have a demo that I’d love to share with the community,” I blurted.

This was a lie. What I had was a $200 logo by Günther and a business registration in the State of Illinois. But I had the certainty that, if I had to, I could hack together a demo over a weekend.

Abel’s eyes lit up.

“That’s great!” he replied. “We have our quarterly leadership meeting in four weeks. How about you share your demo then? You can hook up your laptop to the projector. There’ll be dinner.”

Hot damn, I thought. Just a few days ago, I was a PhD dropout with Xcode and a dream. And now, I’m an entrepreneur being offered an opportunity to pitch to this community’s leadership at their quarterly meeting! I’m good at this.

I was so good, in fact, that I was annoyed that I had to wait four whole weeks. “No problem,” I replied confidently. With a touch of arrogance that I was convinced they would find charming, I added, “It’s a bit far from now but I’m grateful for the opportunity.”

We shook hands. The deal was sealed.

Real programmers don’t read docs

Any self-respecting entrepreneur has the common sense to build something only if it’s necessary, and any self-respecting programmer has the misplaced confidence that they can build anything in a weekend, if necessary. I believe that successful tech entrepreneurs figure out a way to harmonize these two contradictory impulses, and I am nothing if not a serially unsuccessful tech entrepreneur.

I had promised a demo that I didn’t have and that I needed to whip up on short notice. The obvious move would have been to download Unity, a well-known game engine and editing environment.

I did in fact download Unity right after my meeting with ECAC, and I did open it:

What am I even looking at? Who the fuck knows. I said: “probably easier to write a game engine myself.”

The resulting effort ate up all four weeks of my runway. By the end, I had the crappiest lil’ 2D tile-based rendering engine you ever saw. No sound, no UI, no input, and no textures. Just a bunch of janky green hexagons floating in a dark void that crashed every time you panned too far to the left. I wrote the engine using Objective-C, a loathsome language with a syntax as charming as a sweaty Microsoft executive yelling “Developers, developers, developers!”

I kept working on that stupid game engine until the morning of the ECAC meeting. I had nothing even remotely resembling a game. I was in very deep trouble.

I had no choice but to fire up Unity again. Again, that inscrutable initial screen. Fuck. Fuck. How the fuck does this thing work.

Turns out, it’s not that hard, you just need to read some docs and look up a tutorial on YouTube. I became acquainted with the basics:

There’s the editor screen, which shows you a scene, kinda like a movie. You can put objects in the scene. A camera provides a visual perspective, which could be first-person, birds eye, or whatever. You need a light source to see your objects through your camera, just like in real life. Finally, you add a bit of code so that you can control your objects with inputs like the mouse or keyboard.

Seriously? That’s all? I was kicking myself for having waited this long.

I anxiously looked at the clock. 1:30 PM. The meeting was at 6 PM. Minus the 45 minute train ride and assuming that I skipped showering, which in those days was a regular occurrence, I had 3 hours and 45 minutes.

Fuck. What’s the most basic fucking thing I can put on the screen?

As it turns out, this:

It’s a map of Africa. Well, “map” is a generous word. It’s a silhouette of the African continent floating in a void.

I look at the clock again. 3 PM. What’s the most basic representation of a player that I could put on this map?

It’s a sphere. An untextured, perfectly round, white ball. You can move it around the map like this:

Oops, you moved off the map and have fallen into the void. The good news is that you don’t die because the pendulum of history and time are illusions.

This embarrassment was my demo and would have to make do. By the time that I figured out how to make the ball move on the map, I had just enough left over to shower and head uptown to ECAC. I threw my laptop into my backpack and, with a gnawing pit in my stomach, got on the train.

Demo night

I felt like throwing up the whole ride over. I’ve had always flown by the seat of my pants, asking my teachers extensions for every paper and skimming the assigned readings ten minutes before class. But that was class, and this was the real world. My dreams were on the line, and I was failing.

Still, I tried to maintain my composure and some semblance of bravado. “The leadership team doesn’t know programming,” I told myself. “It’s hard work. I’ll explain that it’s super alpha early days, just toying with concepts. They’ll understand. I’ll be back.”

The first thing I noticed when I got there was the parking situation. Thank God that I took the train because street parking was packed. Must be a concert somewhere, I thought.

I notice lots of people filtering in as I approach the ECAC building. I ask someone if they knew where the leadership meeting would be. She looks at me quizzically and says, in a confused tone, “upstairs.”

The steady stream downstairs has now turned into a throng approaching a pair of doors. As I get close, I see something altogether unexpected.

The “leadership meeting” was actually a general meeting for the entire Ethiopian community of the greater Chicago area. I see hundreds of people seated at tables dozens of feet long inside a massive hall. Massive buffet-style food stations dole out delectable stews as kids run around. At the front, four seated women roast coffee beans over camp stoves, their helpers grinding and brewing behind them. A DJ plays Ethiopian music near the front, where there’s a makeshift stage set up with a mic, a projector, and one lonesome-looking HDMI cable dangling from the side, waiting for some poor sucker’s laptop.

Whatever bluster I had managed to retain sank like a rock inside my guts. These good people had taken me in, put their faith in me, and I was about to make a total ass out of myself with this absurd excuse for a demo.

What about being misled into thinking this would just be a small meeting of ECAC’s leadership? I racked my brain trying to understand where I had been wronged, when it dawned on me: I had made the whole thing up. Abel never said anything about a leadership meeting, he just said that there would be a quarterly meeting. Truly, I was the sole author of this unfolding fiasco.

Baffled and shaking, I made my way to Abel, whom I saw smiling and waving from across the hall.

“Hey, you made it! Welcome,” he greeted me.

“Yeah. You know, just so we’re on the same page, this is gonna be a really simple demo. Like, really simple. I’m still in the research phase, it’s going to be more about my research than the game demo,” I explained.

“OK, that’s no problem. We’re just glad you could make it. Find a seat, grab some food, and I’ll call you up when it’s time,” he replied. Abel was killing me with kindness.

I grabbed a plate of injera with some meat and sat down next to a nice family. We made some small talk as men with massive novelty-sized bottles of Johnny Walker Black Label patrolled the aisles, topping up the tiny white plastic disposable cups that everyone had for drinking whiskey. The more we talked, the drunker I got.

Abel came up to the stage to interrupt this feast and began speaking. He kicked off a series of speeches in a mix of Amharic and English that I didn’t understand.

After what felt like an interminable wait, Abel switched to English and announced that we had a special guest. He’s a young man making a video game about Ethiopian history and that he would be talking about his project and performing a demo. With that, he called me to the stage.

There was some brief applause as people searched around and spotted my pasty face. I apprehensively took the mic, and I started to speak.

And I killed it:

Imagine a broke grad student who goes all the way to the Vatican to uncover the hidden life of Ethiopians, who—he’s convinced—were secretly critical to the development of modern science. This grad student couldn’t afford a hotel or even a hostel, so he camped two hours outside of Rome in the July heat, schlepping forty pounds on his back every day to the city and back, just for a chance to read these important manuscripts. What he found was big—so big, that he decided to go to the source, to mother Ethiopia herself, and learn Geʽez with the blessing of [dramatic pause] Donald Crummey himself.

[Audience gasps.]

But this young man’s story was so explosive that the academy couldn’t contain it. So he quit grad school to create a startup and take this knowledge to all the people of the world who need to know about Ethiopia’s glorious role in the development of modern science. And he spoke the language of the people: video games.

Ladies and gentlemen, I am that broke grad student. Still broke, but not a grad student. Your community is my new teacher and I’m but a student. I can’t tell this story without you. I need you.

I stand before you tonight in humility, ready to give you the tiniest taste of what’s to come.

The crowd was enthralled. Every word I said was true and the people knew it.

That’s when I pulled out the laptop from my backpack and plugged it in. Günther’s $200 logo briefly flashed on screen. The crowd waited in anticipation.

I loaded up Unity, and I pressed play.

The crucial factor that most people overlook is time.

All indications suggest that AI will eventually develop into a generalized super-intelligence, the products of which can meet or exceed our own output as humans.

I doubt that we’ll be able to resist the temptation to let it take over our economy. A tireless mind that needs no food and doesn’t complain about healthcare is very appealing to businesses. When it can offer legal, medical, and engineering services for almost no cost, that will appeal greatly to consumers.

But today’s advanced AIs don’t “think” in the same way that we humans do. How could they? We don’t yet understand how our own minds work, nor have we mastered the basic physiological processes that support our brains, like sleep or nutrition.

Not to mention the can of worms that’s spirituality. So many of the leading engineers and investors in Silicon Valley are active proponents of psychedelics, which has led to a new spiritual awakening in the heart of the AI revolution. Have these humans reconciled the opening of machine minds with the opening of their third eyes (some are trying)?

I think there’s plenty of mystery in humans still left for us to create something altogether different from what AI is able to produce. If we could augment our own output with that from AI, perhaps we could evolve our own intelligence.

How the invention of writing evolved our intelligence

There are precedents. In Plato’s Phaedrus, Socrates decries the invention of writing—a relative novelty in Greece at the time—because he believes it would supplant memory, making us more stupid.

What Socrates couldn’t have seen is how writing changed the very way that our minds work. Our greatest scientific achievements would scarcely be imaginable without mathematical notation or the ability to precisely communicate and record observations. Some of our deepest sentiments and emotions, from schadenfreude to hygge, might never have been felt without the ability to write about them.

Just as writing proved to be much more than a memory-killer, we can direct AI development in a similar direction, augmenting our intelligence and spirituality rather than supplanting it.

To clarify, augmentation isn’t a way to “outsmart” AI—that’s a fool’s errand. It’s about differentiating our intelligence from what we have today and, by extension, what LLMs have been trained to mimic. A new human intelligence would create worlds that we can’t yet imagine, ones less bleak than the zero-sum world in which AI replaces all human creative output.

Timing is everything

All these changes depend on how much time we have before we reach generalized artificial super-intelligence (or AGI).

If LLMs are ready within a year or two to replace all professions that require no physical presence, I’m pretty sure we’re screwed. Government in theory would be the most likely actor to step in to handle the fallout. In reality, our sclerotic politics show no sign of being able to cope with a sudden displacement of millions of jobs. For those replaced by AI, the hit to their sense of purpose and meaning would be even deeper.

But if LLMs still need, say, ten years to replace doctors and lawyers, programmers and marketers, and so on, then maybe we have hope.