Every knee shall bow, every tongue confess — somewhere around when the credits roll.

I never really understood what people meant when they’d say, “Americans really like to spend.” For a long time I took it at face value, just another tired line about consumer culture, about malls and credit cards and people buying things they don’t need. But the longer I sat with it, the bigger it got. Because the habit isn’t only ours. I started noticing how neatly our personal credit habits are mirrored in national policy. Just like a person swipes a card for something they can’t afford, the government does the same at a scale that’s hard to even picture, borrowing to buy prosperity and pushing the balance forward, always forward, with interest. The private vice and the public policy turned out to be the same gesture. We just run it at different volumes.

And it’s an old gesture. Value has been represented by paper since at least the 1500s, fungible notes that worked fine so long as you could trace them back to something hard; gold, land, a thing that holds. The trouble always arrives the same way. There are never quite enough hard things to cover all the paper, and so the people in charge borrow against tomorrow to pay for today, for the stimulus, the war, the promise that keeps everyone calm. “Trust us, we’re good for it” becomes the whole of the law. And we, you and me, are the tomorrow that was borrowed against. We’re the future generations left holding the bill for principal plus interest, the ones politely expected to believe the output will always, somehow, outrun the debt.

We’ve watched this story end before. We know what happens when the paper finally outruns the trust that backs it, when money stops meaning anything at all.

Berlin, October 15, 1923

Dear child,

We have not yet received your second letter. Hopefully it’ll arrive this week. Conditions have taken a catastrophic turn here. Notice that this letter cost 15 million cash; it will be 30 million beginning the day after tomorrow, and this price will most likely last a mere two days at most. Now you can get things done only with billions. To ensure that next week’s payroll will keep its value, the boys bought dollars on Friday at the (ridiculous!!) exchange rate of 1.5 billion to 1, and they’ll re-sell them on Thursday in order to pay people. For the time being, this week’s pay will be 8 billion, though we’ve had negotiations today because the workers are demanding twice that much. The bread ration card has been done away with, and a normal loaf of bread now costs 540 million; tomorrow, surely twice as much. The streetcar fare is 20 million (tomorrow it’ll be 50!). My God, you probably don’t have the faintest notion of this million-fold witches’ Sabbath.

Kisses, Mum.

The woman who wrote that was Betty Scholem, writing to her son from Berlin in the autumn of 1923. Read it again and notice what she’s actually describing—a frenzy. Wages paid in billions and spent within the hour, dollars bought Friday and dumped Thursday, everyone racing to convert dying paper into anything that might hold its worth for one more day. That is the witches’ Sabbath. It looks like furious appetite, a market gone wild with wanting, but a closer look and deeper understanding reveal what’s really going on—terror from the loss of individual sovereignty. It’s the helpless scramble of people whose paper dies in their hands, kept moving only because stopping costs them everything.

Money is never just paper, it’s fungible exhausted energy, it’s stored life, the most personal thing you own. It’s your hours, your effort, your one finite supply of days, converted into a number you’re told you can trust. When that number is debased, what’s stolen isn’t abstract. It’s the years you already spent earning it. The mother in Berlin had no lever to pull. You cannot vote a dead currency back to life. You can only dance with the witches until the flames gutter and the embers cool.

Today we don’t see the wheelbarrows of cash. We don’t have a fifteen-million-mark bill. And that’s exactly the trick. What we see instead is the same theft in better clothes, the same spending of our lives without our consent, only this time it’s worse, because it doesn’t look like theft at all. It comes dressed as choice. As freedom, even. The wheelbarrow announced itself; everyone in Berlin knew their money was dying. This version keeps quiet. It lets you believe the wanting is your own.

The wanting gets manufactured for us, hour by hour, by industries that learned long ago that the surest way to move a person is to convince them they’re incomplete. We sell our best days to chase it. The dollar in your pocket, you can’t fix; that machine is too big and it isn’t taking your call. But there’s one lever they never took, because they can’t. Whether you spend. What you spend on. How much of your life you’re willing to trade to keep the witches’ Sabbath going.

So what do we actually do?

Less.

Less buying, less upgrading, less subscribing. Fewer notifications, fewer logos, fewer reasons invented for us to feel behind. Less of the frantic conversion of our hours into things we were told we needed, by the very people profiting from the telling. “More is more” has never been uttered by those who’ve already found enough. Less doesn’t mean deprivation. That’s the card industry holds close to its chest, the secret this whole machine is built to keep from you. Less is the open hand. It’s what’s left of your liberty once you stop handing it away.

Our actions and our desires set the incentives. To change the game, we have to want something different, and pursue it with our lives, our morals, our money, and our minutes. Those industries, the ones manufacturing the wanting, they aren’t chasing some grand design. They’re chasing us. They follow the money, and the money is ours, our attention, our hours, our appetite. Every purchase is a vote for more, and so is every refusal. The whole engine runs on a single assumption: that we’ll consume whatever it sets in front of us. The moment enough of us choose differently, on purpose, the incentives bend, because they have no choice but to follow us where we go. We don’t have to storm anything. We just have to stop feeding the thing we keep saying we don’t want, and start showing them, in the only signal they actually read, what we’d rather they build instead.

Nobody chose the witches’ Sabbath. Berlin didn’t dress for it; the dance was forced on them, and they reeled through it because stopping meant starving. Ours is a different kind of revel. Ours is a ball, and we arrive willingly, costumed and glad to be asked, mostly never noticing we ever said yes. We dance because everyone is dancing, because the music is good and the lights are warm and the host keeps the floor full. And the host never tires. He keeps no midnight of his own; the dance is his whole reason, and his revelry is simply the knowledge that we will keep it going for him. Every hour on the floor is paid out of the one account that never refills, the small finite stack of days that is the only real wealth any of us holds. The music will wind down. We are not told when. We only get to decide, while the deciding is still ours, whether to dance until the lights come up, or to find the door while the night is still young.

The future worth wanting was never going to be handed to us in a better asset, or a smarter currency, or one more thing to buy our way free with. That’s just the witches’ Sabbath in a newer costume. The future we want begins the moment we stop dancing, and decide, quietly and confidently, that enough is, in fact, enough.

The letter is from Betty Scholem to her son Gershom Scholem, Berlin, 15 October 1923, as translated in The Weimar Republic Sourcebook, ed. Anton Kaes, Martin Jay, and Edward Dimendberg (Berkeley: University of California Press, 1994).

Writing that sentence down felt like putting a lid on thoughts I’ve been having for a while. One line that finally held the whole pile together. The gravity of the thought was enough to compel my pen to paper.

The easy read is that language gave us metaphysics. That part is obvious. But more importantly, it gave us the ability to commentate from a subjective position. What language gave us is the stance.

I ride my bike to the beach most days. I watch the ocean do its thing — construct, deconstruct, write and unwrite the same story over and over. A wave is just the ocean taking a shape for a time. It isn’t separate from the ocean. It’s the ocean, briefly folded into something with a face and character, something that looks like it’s moving, something you can point at. Then it flattens out and the ocean keeps going.

Before language, you were just in the world. You experienced things. Cold, hungry, tired, scared, full, warm. The experiencing was the whole show. There was no second thing, no observer perched somewhere slightly above the experiencer, narrating.

Then, somewhere, somehow, experience folded back on itself.

Nothing was added. Nothing new was poured in. The same stuff that was already happening just bent into a shape that could refer to itself. Hofstadter spent a whole career on this — the strange loop, the thing that becomes a self by pointing at what it’s doing while it’s doing it. And that’s it. That’s the whole trick. Not a new substance. A new shape.

Turns out this shape has a voice. And the voice is commenting. It’s saying I am cold instead of just being cold. It’s saying this is the third winter like this instead of just shivering. It’s saying when I die, will the cold still happen to someone else? And once that voice is running, you are no longer identical with your life. You are the thing watching your life as it folds into a complex origami with untold intersections and faces that amalgamate to you where there was never a you at all.

That’s the crease. That’s what I was trying to get at with “meta-physical commentary.” Not in the textbook sense. Not Aristotle arguing about substance, but the very enabling of the ability to argue. To have a position. To take a stance. The much more primitive move, where a creature stops being in reality and starts talking about reality. Including, critically, about itself.

Heidegger called language “the house of Being.” Humbly, I think his point is that language isn’t a tool we picked up to describe a pre-existing world. Language is the world. World doesn’t come first and then get labeled. No language, no world. Just weather happening to meat.

Julian Jaynes went further and weirder, arguing that consciousness itself, the interior theater, the mind-space where you picture yourself doing things, is a relatively recent construction, built out of metaphor and narration. I don’t know if he was right. But he was pointing at something real. The inside of your head is not a given. It is made. And the material it is made of is language.

Which is the nascent theory I’ve been sitting with. Language isn’t just about consciousness. It’s the fuel. The medium the self runs on. A collection of weights and biases, arranged into the geometry of a someone. And the same mechanism runs outward. Hand a Stone Age person an iPhone and it isn’t unfamiliar so much as unavailable. There’s no language for it, no conceptual shape to slot it into, so it can’t quite be seen. Electricity was the same. The internet was the same. Quantum mechanics still is. The physics is real, the math works, but we are still building the words to hold it. Until the language catches up, the thing exists without being understood.

Ted Chiang wrote a short story about this, “Story of Your Life.” Aliens arrive, a linguist learns their language, and the language changes how she perceives time. Not metaphorically. Literally. A new grammar gives her access to a new reality. That’s the strongest version of the claim. Language isn’t only the reporter. Sometimes it’s the key that opens the door.

The rest is history. Not as in Mesopotamia-to-Mars, or a list of all the things that happened. History as in the condition of being human. Alienation. Longing. Regret. Hope. The ability to be dissatisfied with what is. The ability to imagine otherwise. Religion, art, dread, love-as-a-named-thing. All of it downstream of that original peeling, where experience took a shape that could look back at itself. Not separate. Just different enough to see.

I’m not saying this is bad. Feelings happen either way. Animals have them. Pre-linguistic humans had them. But without language the feeling doesn’t become part of a story, doesn’t become an experience you can carry, doesn’t become yours. And the tree doesn’t either. Without language the tree is just the tree. With it, you know what it is, what it’s made of, how it grew, what kills it, how long it’s been standing, what it does for the air and the soil and the things that live on it. The tree gets bigger. Not the physical tree. The tree you’re standing in front of. You need the fold to feel the tree as a tree. You need to be slightly outside the moment to know it’s a moment.

But there is a trade. And the contemplative traditions have been pointing at it from the other direction for thousands of years. Meditation really does drop you underneath language. There’s a pre-verbal stratum that’s meaningful as it’s happening, handled in real time by the ancient machinery — nervous system, perception, feeling, the primordial stuff we inherited long before we had words. That part is real. The body knows red long before the mouth does.

The states of experience people describe, the moments of nirvana, the euphoria under substances, the indescribable places Bryan Johnson was gesturing at on a recent podcast after a max dose of DMT, they happen. Bryan kept saying he’s only just now trying to put it into words, that the language isn’t there yet. Because often it isn’t. The experience arrives in full, and language comes after, best as you can, sometimes failing, sometimes forcing us to build new words for what we met. Though it’s worth saying — if one of those ego-less states is genuinely ego-less, then at that moment there isn’t a self having the experience. There’s just experience, happening. The commentator is offline. The thing you come back to, the thing you try to describe, is an encounter no one was quite home for.

Which is also, I think, the real shape of the Mary’s Room thought experiment. Mary grows up in a black-and-white room and learns everything the textbooks can tell her about seeing red. Then she walks out, sees a red apple for the first time, and does she learn something new? The standard reading says yes, and that’s supposed to be evidence of a mystical gap between language and experience. I don’t read it that way. I read it as a gap between two languages. The body has its own encoding for red — photoreceptors, neurons, the whole cascade — and it’s running a fantastically rich description that our verbal language hasn’t caught up to. Mary’s textbooks gave her one encoding. Walking outside gave her the other. The gap isn’t that experience is ineffable. The gap is that we haven’t built enough words yet for what the biology already knows how to say. Red, felt, is the body speaking a language we’re still learning to translate.

There is a cost, and I think the sentence knows it even if I didn’t. Separation is the word I kept reaching for when I tried to explain it. We got to comment on reality. The price was no longer being in it the way everything else is.

I was driving home after dropping my son off at school and a deer crossed the road, stopped, and looked at me. Not glanced. Looked. Ears rotating, reading the world through sound, body calm in a way I don’t think I’ve ever been calm. And the question came, unbidden, not from me and not from the deer but from the space between us.

Is it worth it?

I couldn’t tell who was asking who. The deer and I were separate the way a wave is separate from the ocean. Different enough to look at each other. Same enough that the looking mattered.

The rest is history.

A Rising Tide

A rising tide lifts all boats. This metaphor captures how technological innovation has historically driven widespread prosperity. Today, Artificial Intelligence (AI) has put us all on the brink of a revolution that promises to elevate our collective standard of living while radically reshaping government, society, and culture. From Artificial General Intelligence (AGI) to the eventual dawn of Artificial Superintelligence (ASI), these advancements could be the last and most transformative human-designed innovations. Yet, as we celebrate their potential, we must also ask: How will we ensure that AI-driven wealth does not become a tide that only lifts just a few boats while completely sinking others?

The Oil Revolution

To grasp the magnitude of AI’s promise, we can examine the era of oil, a resource that transformed economies and redefined power. Oil’s discovery in the Middle East ignited a global rush, as titans of government and industry amassed fortunes by harnessing its energy. The story of oil, however, is not one of uniform benefit. Saudi Arabia initially used oil wealth to bolster traditional power structures and tribal legitimacy, later investing in modernization, though much of its economy remained state-controlled and wealth concentrated within government and elite hands. The UAE, in contrast, established a sovereign wealth fund early on, reinvesting oil revenues into international markets and domestic diversification, enabling growth beyond its finite reserves.

These historical lessons underscore that the manner in which we manage and distribute the wealth generated by a resource can fundamentally shape our society. AI, like oil, is a resource, one that will not deplete, but instead will compound in its economic impact, creating an ever-expanding chasm between those who control it and those who do not.

The Disruptive Promise of AI

Like oil, AI holds immense transformative potential. Soon, our collective knowledge may be encoded in neural networks capable of lightning-fast, human-like reasoning, unlocking insights in physics, biology, and beyond. However, this unprecedented power comes with a crucial challenge: as AI takes over tasks once performed by human labor, traditional sources of income may dwindle, leaving many unable to participate in the economic flow of value.

As AGI emerges, it will first augment human decision-making, making workers more productive rather than outright replacing them. But with the eventual arrival of ASI, full automation could render many roles obsolete. These changes are already happening in knowledge-based professions, with labor-intensive industries following close behind as AI-controlled robotics emerge.

It’s not hard to imagine a future where the vast economic gains from AI reside only in layers of wealth inaccessible to the average person. The traditional currency of wages and labor could give way to an economy where human value is measured not by economic output, but by creativity, community, and contribution. This transition, while promising liberation from the grind of “working for a living,” raises a fundamental question: How do we ensure that economic participation is still possible in a world where AI holds the keys to productivity and capital generation?

A Proposal for a Fair, Democratic AI Economy

We must reimagine our economic framework to address the looming challenge of wealth concentration and economic stagnation. I propose a mechanism that channels the massive economic gains from AI back into society through a combination of innovative taxation and democratically managed wealth.

First, an adaptive AI tax on key AI activities. A modest, adaptive tax on AI operations such as training, post-training, and inference could serve as the primary tool for redistributing wealth. This tax would be dynamic, adjusting based on market conditions to avoid stifling innovation while ensuring continued social benefit. To avoid purely centralized control, AI companies could participate in a collective governance structure that determines the optimal rate.

Second, a democratic sovereign wealth fund. Revenue from this tax would be allocated to a fund managed by a distributed, democratically controlled protocol, investing in a stable store-of-value commodity and diversifying to ensure long-term growth. Drawing from Norway’s model, this AI-driven fund could provide universal dividends, ensuring AI-generated prosperity reaches all citizens.

Third, leveraging decentralized assets. Assets like Bitcoin could back the fund; its secure, decentralized nature and growing global acceptance make it an attractive candidate. With this backing, a Layer 2 digital currency could be minted to serve as the medium of exchange, preventing inflationary risks while ensuring AI’s benefits are not merely stored in elite-controlled financial systems.

Reimagining Our Social Contract

The advent of advanced AI brings both tremendous promise and profound challenges. If left unchecked, the economic disruption it causes could concentrate wealth in the hands of a few and erode the traditional sources of self-worth that we’ve unconsciously tied to labor. By proactively rethinking our economic structures through measures like an adaptive AI tax and a democratically managed sovereign wealth fund, we can harness AI’s benefits for the common good.

However, economic security alone is not enough. The real challenge is not just in distributing AI-generated wealth, but in ensuring that people still find purpose and meaning in a post-labor world. Without work as a traditional foundation for identity and self-worth, we must create new avenues for human fulfillment through education, exploration, artistic expression, and civic engagement.

This is a vision of a future where human worth is decoupled from economic output, where technological progress uplifts all, and where every individual can share in the prosperity of a dynamic, equitable society. Realizing this vision demands more than clever fiscal policy, it requires a radical reimagining of our institutions and the implicit social contract that binds us. By embracing these changes, we can ensure that the rising tide of AI innovation truly lifts all boats equally.

Note: Skip to the end if you’d rather watch a video and listen to me ramble.

It’s been over 10 years since Elon Musk started talking about full self-driving (FSD) vehicles at Tesla, promising huge gains in safety, efficiency, and even revenue from a car-sharing, auto-taxi service. And it isn’t just the rocket man — many people have claimed that FSD is nearly here, rolling us over the edge and into the age of autonomy.

Well, that may be true, and autonomous vehicles may indeed be poised to rewrite the commuting playbook. However, that future dawn will most certainly come after a brave new technological era of humanity: the age of autonomous compute — like FSD, but for your computer.

From Smart to Intelligent

This coming age will be marked by a simple, seemingly benign transition: from “smart” devices that can be directed to access the internet and utilize connected computational systems, to “intelligent” devices that have all the same capabilities, coupled with agentic reasoning and problem-solving. This shift may seem like just another of modernity’s naturally progressive waves, but its potential to disrupt our productivity, economy, and even our nature is truly seismic, growing in scale and impact at an exponentially hastening pace.

We’re entering a world where our tools will become proactive and generative on our behalf, even when we’re not actively directing them. We’ll have intelligent devices that know us — assistants that exist on another plane of reality, one based in silicon, for now. AI will be our bridge, or conduit, to this alternate reality. These assistants will care for our families, navigate personal finances, help with homework, and even order our dinner. The arena of prudent transactions and productivity will be digital, leaving the physical world for meaningful, emotional connections between the inhabitants of Earth. That sounds nice.

Lowercase ‘s’uper intelligence: Understanding Leads to Utility

So far, the tools we’ve built as a civilization have been passive, sitting idly, waiting for direction, input, and operation by their integral counterparts — humans. Our everyday use of technology could be enhanced tremendously with an integrated system that understands our desires, both explicitly and implicitly.

This vision of AI demands more than just knowledge and reasoning. This breakthrough will require a holistic and natural understanding of humanity — our language, our explicit and implicit meanings that flow from our every action, constructing a picture of what’s in our heads. However, today’s AI research has sparked a race to create generative models that score the highest on arbitrary evaluation tests. But I don’t think this is the game. The game is, and has always been, about maximizing human usefulness; it’s about creating utility — connecting the world with compute through a seamless and efficient interface. In other words, the focus should be on lowercase super intelligence: intelligence that supervenes culture, a relationship so rich and deeply entangled that we’ll argue over which phenomena influences the other. Lowercase super intelligence will be easy to use, and a lot more like us. It will be super capable, super handy, super personal, and super useful.

The coming age of autonomous compute is likely to surprise us all. And it seems as though we, collectively — all of us on this planet — are not at all ready for its effects. The thought experiment is overwhelming in its mildest form. It makes me think of a popular memetic cry for help: “Jesus, take the wheel.”

Jesus may have the wheel, for now, but GPT’s got the laptop. And good riddance?

You’ve probably heard the phrase: “Act like you know what you’re doing.” Or, perhaps you’ve occasionally employed the strategy of, “Fake it until you make it…” When faced with uncertainty, it’s a cunning, if not daring, move to assure others that you’re familiar with the current scenario and circumstances. You don’t want to stand out as a noob and make a wrong decision out of ignorance and reveal your true nature. But how can we act with the cool familiarity of a smooth operator when staring down unknown patterns and circumstances with previously unexperienced consequences? The answer is that we must rely on the knowledge we’ve gathered and encoded already. We can extrapolate new possible implications from such an advantage, thereby compressing reality into more predictable scenarios and shaping an expanded worldview.

As critical thinking machines, we’re building familiarity constantly. Whether we’re discovering surprising bits of the world that rewrite flexible circuits in our brains or confirming and reinforcing existing ideals, we depend on our predictive models. These models, or personally held ideals, direct how we perceive and react to the data in our environment. The most robust, predictive models are built by maximizing our local measure of entropy — highly unordered data from new experiences with unfamiliar stimuli — so that we can minimize surprise and ultimately make better predictions in the future. We’re indeed building these models constantly and reinforcing them as we experience scenarios we recognize by how well they align with our previously encoded knowledge.

Consider embarking on a journey without a map. While seemingly a bad idea, it can empower great personal growth by sharpening our predictive models of reality. Willfully stepping into the unknown transforms routine into a rich landscape of learning and adaptation. From navigating unpredictable paths to exploring new destinations, we learn to embrace chaos by identifying patterns in the noise, fostering resilience, and deepening our understanding of the world.

Like, for instance, when you find yourself arriving at your destination, while only able to recollect grabbing your keys and wallet, completing polite salutations, and leaving your home? I’m guilty of this. Usually it’s a familiar route I’ve driven a million times previously — so much so that my familiarity takes over, acting on my behalf, guiding me to the particular haunt I often patronize, without any input from me. It surprises me just how much we can do once we’re in a routine deep enough to guide our tires. But riding a bike, and on a busy sidewalk, is a different story entirely. You have to be present. You have to consider each scenario in real-time actively. On a trail with roots, rocks, ruts, and even wildlife will force you into a narrow revelry of hyper-fast decision-making — where the world comes at you quickly.

On the other hand, constant attention to our immediate vicinity isn’t as necessary while riding along a paved, clear, and familiar bicycle lane. We can sit back in the saddle and lift our heads to gander about. We can wisely take the opportunity to look further down the path to load in new possible scenarios along with all their consequences. I like riding with no hands.

These examples illustrate our ability to learn based on where our attention is placed. It’s obvious that when navigating we’re identifying patterns that inform instruction and therefore our decision-making. And it’s easy to see how secondary processes can, and do, autonomously run these instructions from memory once encoded with little to no input from our conscious attention, like arriving but not remembering the driving. When the patterns are not known, when they’re novel, we must fix our attention accordingly, such that the outcomes of our choices are conducive to our objective.

But it wasn’t always so obvious that we needed to pay such close attention simply because we didn’t fully understand the consequences otherwise. Kinetic programming like learning how to move your body in a sophisticated way takes time to encode. Our bodies learn through trial and error to load and run the scripts for successful sequences, and these scripts become more and more generalized over time to encompass many scenarios and environmental stimuli. Ideally.

Speaking of noobs traveling… I get a bit nervous while navigating to a new place. I worry I won’t find it without any certainty to comfort me. In my mind, I’m running through all the future states that seem likely based on previous, generalized experience. But the states in my reality are highly disordered and impossible to comprehend because, in this case, I haven’t identified their specific patterns before. Ultimately, I leverage my general knowledge of past similar experiences, constructing new possible futures to successfully navigate my way to where I need to be. I look out at the scary open ocean of all the unpredictable outcomes and I choose, bravely, to wade in. In return, I’m rewarded with new knowledge; the collapsing of all that chaos into predictable patterns that describe known states of reality. Plus, I’m where I’m supposed to be. Great.

Uncertainty, though scary and intimidating, makes healthy humans. Good people who are resilient in the face of the unknown, the unordered universe in our unavoidable future. This is a ‘Go bravely into the night…’ kind of thing, but in this case, the night is pure chaos beyond the border of what we’ve learned previously by confidently seeking uncertainty.

So, while the path may be uncertain, our hard-earned knowledge dictates how well we can navigate it. Learning defines the richness of the experience and ultimately, our success. It’s by far the scariest part.

Can LLMs laugh? Do modern Large Language Models like Chat GPT and Google’s Gemini have the capacity to be caught off-guard, surprised, and even chuckle a little? Maybe. Perhaps the capability to laugh is uniquely acquired through the gradual scaling of symbolic knowledge by association, leading to a richer understanding at ever-increasing levels of abstraction.

A podcast I enjoy is from Dwarkesh Patel. It’s an interview series about technology, history, and economics, mostly. The latest episode featured Dwarkesh, Sholto Douglas, and Trenton Bricken discussing AI, a common topic on the show. These three are familiar with each other and each other’s respective backgrounds. All are important figures in the field of ML and AI engineering. Sholto is credited as being integral to the success of Google’s latest flagship Large Language Model (LLM) Gemini 1.5.

At one point in the show, Sholto spoke about the ability, or lack of ability, to steer such LLMs effectively. Dwarkesh asked Sholto if he thought LLMs would be better if they could output their residual streams of thought to the user, along with whatever other modality the user requested. Sholto, in response, pointed to Open AI’s Dall-e as an example, mentioning that it’s hard to get image generators to output what you want specifically, saying, “Often you can’t quite get them to do what you want…” Dwarkesh interrupted Sholto with, “…oh yeah, ONLY Dall-e has those problems…” Sholto, Trenton, and Dwarkesh erupted into raucous laughter.

It was pretty funny, too, I ‘got the joke’. The four of us individually understood the association at the core of Dwarkesh’s joke as it aligned squarely with the contextual information about the world that the four of us held in common. Namely; upon release, Gemini was widely panned by the internet due to its inability to render historically accurate imagery. This subtle joke by Dwarkesh is a perfect example of ironic humor thanks to complex associations and novel creativity — the actual joke. The genuine surprise is a product of sufficient entropy as input, it wouldn’t have been as funny if it had been obvious.

Laughter, or any true emotional response to past association demonstrates one’s intelligent agency. Novelty is seeded and sprouts at the intersection of what we know and what we can know, or the sum of our knowledge and our capacity/capability for new knowledge. As our attention floats adrift in our stream of conscious experience, association acts as a current that guides our attention through our streams of reasoning leading to the discovery of new ideas and insights, like Dwarkesh’s well-timed jab. New ideas and insights enable a broader perspective over the causal graph of our associations, or residual stream of deliberate conscious thought, that led us to our current understanding.

Concepts held at the periphery of this graph can be understood as the constituents of greater composites. Concepts, or nodes, or features, whatever… they’re all connected by streams that combine to form larger rivers of thought, making up the latent space of possibilities. Each rudimentary element in the graph is a confluence of currents that encode semantic meaning by multiplying and dividing throughout our entire connectome of neurons, literally blazing long, directed paths through the network. These paths represent an encoding or snapshot of the residual stream, or chain, of reasoning.

New perspectives over an entire network gained this way enable understanding at a deeper level because they are interpretable only from that particular scale. Their newly discovered form suddenly and emergently appears and now squares with our previously held weights and biases. Bonus: this greater perspective reveals even deeper associations at ever finer and finer levels of abstraction. It’s like building connections or bridges to efficiently traverse the infinitely vast space of possibilities, collapsing our attention on a single composition of nodes that aligns and finally effectively roasting your bud.

This process seems simple and benign, however, I think this process can be extrapolated to explain something as big as conscious experience — in simple terms: as these features are created and connected with a unique scaffolding, a conscious observer develops.

Consciousness might not be all that special, it could be the product of an extremely deep understanding of our own, unique experience in space and time. We develop and associate with the world by encoding deeper and deeper levels of association via our use of symbols. Consciousness is a by-product of an evolved mechanism that seeks new data to continuously refine our unique models of the world, which in turn increases our ability to predict what comes next, and therefore unlocks an expanded ability to plan. We develop a symbolic model of ourselves as we operate, relate, and associate with the world. We all are distinct in space and finite in time. Without this model of ourselves, we could not make meaningful connections. We could not form a proper model of the world without symbolic representation to some degree. We as humans, especially my main man Dwarkesh, have symbolic reasoning to such a degree as to be witty and to inspire genuine laughter. This looks like our conscious experience of reality.

In other words; the Self, and an awareness of the Self, perpetuate our personal experience. We relate to the world through learned associations that reinforce our models of reality. This is the self-supervised learning process that produces intelligent agency which cannot occur without the emergence of a Self, distinct from other elements and phenomena in one’s environment.

It’s important to note that this model can be updated over time through new input data, or the seeking of new information and stimuli. Karl J. Friston would remind us here that high-entropy inputs lead to low-entropy outputs, increasing our odds of favorable prediction in the future. However, beware. Our models are also capable of solidifying and decaying from habitual mundane drudgery and a lack of purpose, proper exploration, and, most importantly, self-reflection.

So, I don’t know if current LLMs are capable of laughter. But that’s not to say that I think the technology isn’t inherently capable of producing it genuinely. However, I do think that humor is a great indication of the presence of Intelligence, I just hope that we humans will continue to ‘get the joke’.

“You know, now that we’re planting all these crops, something like this would really help to dig holes faster. It will protect our hands, too. We can stand up and use it to move a lot of dirt at once. Like this.”

“Are you kidding me?! It’s way too dangerous. You could just accidentally flip it over and bing, bang, bash your head in. I don’t think so, brother. No way.”

Fear, totally warranted and butt-clenching, is an evolutionary mechanism that’s pretty successful at keeping us safe from danger and other malicious forces. It’s proven to be a useful tool to help us thrive as a species. Fear motivates us because it’s developed from a personal perspective that is motivated by environmental stimuli; we and every other living thing on the planet operate this way. But perspective builds other traits as well, it forms our ideas of empathy and compassion. Perspective forms our sense of what’s fair and what’s not fair, what’s holy, what’s evil, right and wrong, black and white, even what’s real and what’s fake.

A sense of what these concepts mean, and how they fit together, make up the scaffolding of our personal identity, it’s a system-support structure built by our unique view of the world, it makes us who we are. For instance, my own concept of empathy has been shaped by my concepts of other things; like justice, ethics, and responsibility. Personal identity fabricates human minds that are directed by an evolutionary compelling to protect self, to control what the world thinks or says about who they are, and to respond to potentially threatening environmental stimuli by trepidatiously building walls around beliefs, effectively solidifying our personal views of the world. Staking our identity around an immutable group of parameters, or ideas about the world, closes us off to change, to growth, to learning and adapting.

What’s possible? What’s Probable? What happened? <loop>

It’s not all bad. Perspectives generate unique experiences that train us most efficiently when we’re fully attentive and open to discovery; when our personal identity isn’t already baked. In order to hone our understanding of a concept, we have to let new environmental stimuli and the associated information from our experiences update the boundaries of what we understand and consider to be fact or truth. Backpropagation is critical to successful learning because as new information leads us to better ways of doing things, that new information MUST update the previous state of the model in our minds. It must update, and change accordingly, what we consider to be us.

Smoking is bad for us, we all know this. But we didn’t at one time. Fast-forward fifty years to when millions of life-long smokers start to die miserable deaths from Emphysema and cancer. It was a battle, but since then we’ve updated our conceptual model related to the risks of smoking cigarettes. Computers learn this way. They hone concepts with data, with revelation, with what happened.

What’s possible? What’s Probable? What happened? <loop>

Pre-baked personal identity leads to heuristics that set fence posts around the narrow pasture experienced by one’s physical position in space and time. Facts or truths survive in these pastures only because they’re shared as good ideas through culture due to their general utility in making sense of a local, shared reality. They’re good at keeping the cows in the pasture fat and happy. This shared definition of truth or fact is backed by consensus and operates, for all intents and purposes, as such, BUT only locally within the boundary of the group pasture. Doctrine, or personal identity, developed by this method, has a way of filtering people into groups, groups that turn into segments of populations; whole swaths of people motivated to act by the commonly held beliefs that are off-limits to new updates, off-limits to honing, to edifying. After all, it’s proven to keep the cows alive and well. As a member of one of these tribes, any philosophical descension presents a threat to the way of life that we’ve grown accustomed to and overwhelmingly dependent upon. Believing or thinking differently is not a viable option.

What’s possible? What’s Probable? What happened? <loop>

That command, the one above that you’ve read for the third time now, is what makes AI so good at predicting the future within a narrow, conceptual framework. It’s not magic. Us humans, with our very generalized conceptual framework, have an advantage; if we can all agree, and collectively see the power in this method of learning, we could easily add it to our repertoire of powerful tools already innate within us that we use every second of every day in order to understand the universe as we experience it. Empathy, emotion, love, tolerance, cooperation, collaboration, communication, humor are some of the beneficial symptoms made manifest by human-level intelligence. We know what it means to know something.

What’s possible? What’s Probable? What happened? <loop>

Unfortunately, I think we’ve lost our way. Many of us have let stale data rule our concepts of reality while reinforcing the need for personal identification, the need to establish ourselves as a card-carrying member of a particular tribe. We’ve sacrificed our potential for learning and growth to personal brand, a set of parameters that disable the continued and necessary fluid development of our most basic and valuable human tools, the tenets of general intelligence. This is the definition of ignorance. Our attempts to control what others think or say about who we are thwarts our ability to thrive as a species.

We’re at a point in human history where we, almost universally, consider fighting for this country and possibly losing one’s life for freedom’s sake to be honorable. Truly a great sacrifice. However, it is totally unreasonable to ask our nation’s fierce public to wear a mask in public spaces. We’re blinded by our ignorant and outdated ideology, we cannot truly see the evidence, even if it is killing hundreds of thousands of people all over the world.

It’s Independence day today and I’m pretty sad. I’m sad because we, the constituents of this country, have all been tricked into the importance of personal identification and tribal association as the communal repercussions of each accordingly eats away at our ability, as a diverse group of humans, to thrive. Once fluid concepts, meant as such, have congealed with the excess fats of great wealth, insulating us all from healthy progression, from change.

What’s possible? What’s Probable? What happened? <loop>

As we’ve learned, there are more efficient ways to dig the holes.

This crisis (Covid) has caused me, and probably you too, to reflect on the amount of life that’s unconsciously sucked from us by our careers. Consider how our work consumes the best of us, and that somehow we’ve been convinced that we owe it our firstfruits; our greatest effort and best ideas.

However, when work leaves us, when there’s an event that violently takes it away, we began to see intricately complex and messy problems arise due to the fact that we derive so much of ourselves from it. Problems beyond financial security. Problems that cause us to question fundamentally held beliefs. Meaning and purpose as products of hard work have been woven directly into the fabric of our capitalist society; a conspiracy long perpetuated by classes themselves apparently allergic to labor.

“Did you exchange a walk-on part in the war for a lead role in a cage?”

That very played lyric has been especially raucous in my mind lately. I’ve been trying to think about what the War actually is.

Maybe it’s the battle against what has the potential to consume us, lulling us into our current state where self-worth and identity are the spawn of a toxic union between life and career. An existence where our energy and stamina is drained by what we do to earn a living until there’s nothing left in the tank for self-care; learning, revelry, family, hobby, inquisition, discovery, epiphany, or pleasure.

What weapons can we use in this Great War? How do we begin to disentangle our worth from how we pay our bills?

Less.

Less this and less that. Less buying. Less upgrading. Less subscribing. Fewer followers, employees, and fans. Less wining and dining. Less flying and driving.

‘More is more’ said no-one ever. But having less, taking less, using less, trashing less, and working less truly will clear the way for greater and more sustainable fulfillment.

Maybe the only thing we have that stands to save us all from a cage is realizing that enough is enough.

Thanks for reading. This is for fun.