I define ‘friend zoning” as the act of transitioning a relationship from a romantic context to a friendship context by telling the person that you’d rather be friends.

Friend zoning is the most common way to end a romantic entanglement (well, maybe ghosting is the most common, but friend zoning is a close second). Often, when people engage in friend zoning, they aren’t really being sincere. They may not care about being friends; the more important thing is that they want the romantic interaction to end. If that is the case for you, you should ignore this post.

In order for the act of friend zoning to successfully transition your relationship dynamic to one of friendship, the woman cannot be the one to initiate the friend zone. It has to be the man.

If a woman friend zones a man, the woman’s intentions are definite, but the man’s intentions are not. Sure, the man may agree to be friends, but the woman will never be sure of his true intentions. And the woman is right to be uncertain; the man almost certainly wants to be more than friends, or at least remains attracted to the woman.

If a man friend zones a woman, that is a much stronger signal that the man truly wants to be friends. The woman’s intentions are uncertain, but her current attraction is much more likely to become platonic over time than the man’s would be. Also, the possibility of the woman being attracted to the man is less of a barrier to friendship than the possibility of the man being attracted to the woman.

I’m not saying that this is always the case, just that if the man initiates the friend zone, the chances of success are much higher. This has been true in my experience.

This post only applies to cis male/female relationships. I’m sure there are all sorts of interesting versions of this dynamic within other types of relationships.

TLDR: internal contradictions in religions make them more adaptable to different psychologies and societies; because of this, they are more memetically successful. Memetically successful ideas may have multiple survival strategies depending on psychological and social context.

Major religions almost always contain many contradictions within their belief system. One classic example is that in some religions, people have free will, but at the same time their actions are pre-determined by God. Setting aside compatibilist definitions of free will, that’s pretty contradictory.

Why do people continue to believe in these religions even though they contradict themselves? The spread of ideas does not depend on logic; it depends on people getting something out of holding a belief, and getting something out of spreading it. Belief in free will is sometimes convenient, like when you are ascribing moral weight to your actions, but it is also sometimes inconvenient, like when you are unsure about the future and need to be reassured that God has already planned everything out. The contradiction allows the belief system to coexist in either psychological context, and it is therefore more likely to survive in a person’s mind; any person will inevitably experience a wide range of emotional states. Belief systems that are psychologically versatile are more likely to continue existing.

There’s also a contradiction in the vibe of a lot of major religions. There are typically extremist believers, for whom the religion is strict, with lots of draconian rules, fire and brimstone etc; but then there are more relaxed believers, who hold some of the beliefs but not the most intense ones.

In some social contexts, fire and brimstone vibes are more likely to survive, and in others, love-thy-neighbor vibes are more likely to survive. Sometimes society demands strict cultural practices (perhaps when it craves stability); other times, society demands openness and flexibility from its belief systems. If a belief system can straddle both, it can spread more and survive longer. Belief systems that are socially versatile are more likely to continue existing.

Just like an evolutionarily successful animal might have multiple survival strategies depending on the environment around it, a memetically successful belief system might have multiple survival strategies depending on its social context.

A principal-agent problem happens when an agent’s interests conflict with the interests of a principal that they are supposed to be acting on behalf of. Common examples include, well, corporate executives and shareholders and politicians and citizens.

Once you learn about principal-agent problems you begin to see them everywhere. You’ll also see lots of things that are sort of like miniature principal-agent problems. For example, I have a doctor’s appointment today. During my appointments with him, he has implicitly agreed to act in the interest of my health. And yet, this doctor always seems like he is trying to end the appointment as quickly as possible, to the extent that I don’t have any time to ask questions. I don’t know exactly what personal interest he is pursuing, but he gets paid per appointment, so maybe it’s that. My interests are harmed; the quality of care I receive is worse.

I urge you: don’t be the problematic agent. You should actually do the things you are obligated to do on behalf of other people. Being sincere is good. It is incredibly insincere to claim to be doing things on behalf of someone else but actually be optimizing for your own reward. Rethink the way you approach your work. Lots of professions consist of mostly principal-agent problems, like real estate agents.

CEOs should literally do what is good for the business as a whole, not what maximizes their pay. I don’t mean that they shouldn’t be paid a lot, or negotiate to maximize their pay; just that they shouldn’t make business decisions out of their own interest rather than the company’s interest. It’s technically illegal not to do this, but impossible to enforce because there’s always plausible deniability—anything that could be a plausibly justifiable business decision is considered acceptable. This probably has a massive negative impact on the economy.

The same thing applies to you if you work at a company; everything you do at work should be in the best interest of the company. It’s the “creating shareholder value” meme, but unironically. You shouldn’t be doing workplace politics or trying to advance your career. You should be creating shareholder value, because it’s Honest Work, and it’s also what makes capitalism work.

This is the great thing about startups. Startup founders and employees own significant equity in the company and therefore their incentives are more directly aligned with the principals (because they are both agents and principals). This probably has a significant effect on why startups are more of a meritocracy than other areas of the economy (though part of this has to do with the nature of software engineering).

I already see little principal-agent problems when I use AI agents. I don’t have much to say about that yet but I’d feel weird if I didn’t mention AI in what is, after all, a blog post.

The last few weeks I’ve just been writing stream-of-consciousness style. I’m super busy with work at the moment so I suspect this will continue for the next month or two.

Everyone seems to think that the cost of AI tokens is about to go up because of increased demand and scarcity of compute

My prediction: in 1 year, you will have access to better models than you do now, at lower prices. Yes, frontier model prices will probably go up, but non-frontier model capability only lags behind the frontier models by a few months, and non-frontier model inference is getting cheaper at an extremely rapid rate.

Why do we see this narrative happening about model pricing? There are a few reasons to think inference cost will go up, but the main one is the combination of skyrocketing demand and limited supply of inference hardware. And reasoning model scaling is all about increasing test-time compute, so it’s possible that better models will inherently use more reasoning tokens (though there’s evidence that smarter models use fewer tokens).

At the same time, we see other trends that push prices down. Models keep getting better, non-frontier models are catching up quickly, and inference cost of non-frontier models is going down rapidly. There is a track record of equivalent models getting >10x cheaper YoY. Also, chip makers are not raising prices very much.

The answer is therefore that frontier model prices will go up, but you shouldn’t worry about it. Yes, people always want to use the frontier models, because they’re better, but even if you don’t, your current capabilities will still likely increase while your costs go down significantly

Your company’s AI spend will go up this year because of increased usage, not because of increased price per token. You can also save a ton of money by using the strongest models for the most important tasks like planning, and then weaker models for executing on smaller, well-scoped tasks.

• Logos - appeal to logic

• Pathos - appeal to emotion

• Ethos - appeal to authority

Of the three, ethos is the most difficult to understand, but also the most important. Ethos means: whatever it is about the source of some statement that makes you believe it more. When you’re a kid, you trust your parents a lot, and this is mostly because of their authority, not their logical argumentation or the emotional appeal of what they say. “Trust the experts” is a clear appeal to ethos.

Most of the time, we are choosing what to believe based on ethos. You probably believe in evolution by natural selection. But have you actually examined the evidence yourself, pored over the fossil records, checked their carbon isotope counts? And for that matter, why do you believe that the carbon isotope method works? Have you read and replicated the procedures of the experiments that validated it?

We cannot think logically about every belief we have. There is simply too much information. We need to use heuristics, like the authority of the source that information comes from.

Pathos is also important. Whichever side of a war you support is probably based on which people you have seen videos of being killed; which people you feel stronger emotions about. So is logos: logic is used surprisingly often, even unconsciously, to check if new information that we are exposed to contradicts our existing beliefs about the world. But ethos is the most underrated aspect of rhetoric, being both ignored and omnipresent.

The ethos of AI varies across people. Usually, when people have a deep understanding of how AI works technically, the amount of authority they ascribe to its outputs is well-calibrated. You know how it works, so you know what it says isn’t the product of some super-reliable intelligence that you should not question.

People with AI psychosis, or any level of AI delusion, invariably ascribe a high level of authority to AI. This is a load-bearing assumption behind AI delusion; in order to believe fake information from an AI, you must believe that it is a reliable source of information in the first place.

As long as the ethos ascribed to AI remains high among a significant portion of the population, AI delusion will be incredibly common. It’s possible that this would go down over time, as people get information from these systems and then test it against the real world and eventually see that it is wrong. But I expect it to go up, instead. As more people rely on information from AI to shape their beliefs, they will learn to trust it more. At the same time, AI will get more reliable, but will still engage in these delusional loops because of the core nature of the technology.1

In the last half century, we moved from a centralized2 media environment (newspapers, television networks), to a decentralized one (social media), and now with AI that trend is reaching its zenith: everyone has their own media being generated for them.

In the centralized media world, ideas had to be curated by journalists/editors/executives. In the decentralized media world of social media, ideas had to be approved by the crowd with their clicks/views. Now, ideas just have to be approved by an LLM.

In other words, the group of people required to generate and deliver a new delusional belief is down to one, so we will see more one-off delusional beliefs. There is now more potential than ever for a long tail of delusional beliefs to develop. We will see many new ideologies, both generated by AI and involving AI as an authority. We will also see patterns across LLMs, where certain delusions shine through, and become popular across many people.

I am pro-AI, and I think its effect on media will be net-positive, because those who want to will be able to access correct information more effectively than ever. But we will also see incredibly destabilizing social effects during the transition.

I used to think you could see the beauty in anything, as long as you put it on a literal or figurative pedestal, like in a museum, situated so that you are obliged to notice it, to see it in a deeper way, to recognize its existence, helped on by the knowledge that this thing has been selected for the purpose of you noticing it.

If you look deeply enough at something, even if that thing does not seem beautiful in the first movement of thought, perhaps you will see, in the second movement, that it is connected to all things, and that other things are beautiful, and therefore this thing is beautiful. And in the third movement of thought you might imagine an area of nothing, a void, which may not seem beautiful itself, but you know that the void cannot exist by itself, because the void is created by the presence of things around the void, and that this is what defines the void as an object in the first place, and indeed the thing we call a void is what makes up most of our world, a dancing network of nothingness within the boundaries of stars and galaxies, people and perceptions.

Everything has some narrative of beauty if you notice it hard enough. If you can’t notice like this, it’s a skill issue. But something is missing from beauty here, it feels confining, like it is contained only within our perspective, like we are looking at something cold and dead and providing all of the life ourselves through the magic of thought and conscious awareness.

I think that good art can guide people towards a certain realization of meaning, or a sense of beauty, and can do so more effectively than things that aren’t good art. But there are levels to this, and most art, even good art, leads to the experience of beauty-with-thought, which isn’t the deeper sense of beauty that I am pointing to.

There is something beautiful about certain things, something that isn’t present in everything, something that actively works with our perspective to create beauty in the synthesis, like a puzzle piece that fits into our deeper instinctual sense of beauty, before thought, before the development of the cortex.

For me, music is the clearest existence proof of some deeper beauty. There is something about these very distinct musical patterns that creates a universal experience of beauty across humanity, and some evidence points to other animals enjoying music, too. Music can cut to the core of experience. With music, I feel an aesthetic experience that does not require any kind of deeper noticing or awareness. Crucially, some kinds of music are like this, and some are not. There is a sense in which I can appreciate the beauty of all music genres; the initial sense I described above. But there is also a deeper sense of beauty that is not universal. For example, I really enjoy listening to King Gizzard & the Lizard Wizard, which is musically very interesting, but I don’t feel this sense of deep beauty when I listen to them. I feel the deeper sense of beauty a lot more when I listen to certain classical pieces, or even pop music that follows the most commonly-used chord progressions.

Schopenhauer thinks of beauty as awareness without suffering, more of an absence of something than a presence of something. Perhaps beautiful things are things which, when people interact with them, increase the ratio of awareness versus suffering, by either increasing awareness or reducing suffering or both.

Fichte proposes that, in order for a being to become self-aware, to recognize itself as different from everything else, something external must prompt it to have that realization, and that this external thing must itself be conscious, or at least be evidence of the existence an external consciousness.

There is a short story called “Zima Blue” by Alastair Reynolds, which I first saw in the form of an episode of the Netflix anthology series Love Death + Robots. In the story, which is set in a future interplanetary civilization, a famous artist, Zima, is known for creating gigantic murals that mostly just consist of a particular shade of blue, called Zima Blue. He spends centuries travelling the universe and creating these artworks. Eventually, it is revealed that Zima was originally a pool cleaning robot, slowly upgraded and made more intelligent by its owner, until eventually it could upgrade itself. His final piece of art is to shed these upgrades, and return to the state of being a pool cleaning robot, cleaning a replica of the pool he was designed for, whose tiles are, when clean, the color of Zima Blue.

Humans have our own Zima Blue, some amalgam of sensory data that is deeply, instinctually beautiful to us. Individuals each have their own version, but there are definitely shared beauty instincts across all of humanity, and probably across all mammals, and potentially across all life, or even all of existence, depending on your version of panpsychism.

Young people have social anxiety, especially in public. There are many reasons for this. One is that we have created a panopticon for ourselves, made up of smartphone cameras and microphones and the possibility that someone could complain about your cringey behavior in a viral post. This is a perfect recipe for social anxiety, which exists to minimize strong negative social outcomes.

Previously, if you went out in public and did cringey things, the social consequences would have been limited to the people in the same room as you, and maybe anyone who they told via word of mouth, assuming that they even know who you are; if they were strangers they just wouldn’t even be able to identify you, so the cringe reputation wouldn’t follow you around. With smartphones, if you get drunk and do something cringey, any videos or photos will be saved forever, and can be used to identify you, and in the worst case scenario, will be posted on social media and seen by millions of people, including anyone who might hire or date you in the future, and whoever else you would not want to act cringe around.

There is a common belief that social anxiety is about wanting people to like you, and worrying about it too much, and the overthinking makes you anxious, and therefore people don’t like you as much. This frames social anxiety as irrational, because you are getting the exact opposite of what you want. But a much more accurate framing, proposed by Chris Lakin, as that social anxiety is much more about avoiding people disliking you and therefore reducing the potential of strong negative outcomes.

In this way, social anxiety is actually rational within the context of the goal being trying to minimize people strongly disliking you.

The smartphone panopticon introduces many new — and extremely negative — possible outcomes from any given social interaction. These potential negative outcomes end up playing an outsize role in pushing us towards social anxiety, because we focus on minimizing strong negative outcomes. As a result, we become much more anxious and less likeable on average.

It is very unlikely that your specific cringe moment happens to go viral. But humans are very bad at reasoning about small probabilities. Even though the probability is vanishingly small, it is very real, and we can conceptualize it happening, and it therefore has an outsize effect on our behavior.

Generic caveats: people have always said that young people are messed up, and there are lots of other things causing today’s problems.

I haven’t written a proper stream-of-consciousness post yet at Inkhaven. Usually when I do this kind of writing, it’s just for my weekly writing group chat, and not for public posting.

I was going to write down something I like about each person at Inkhaven, but I realized it was going to take a lot longer than anticipated. So if you’d like to know something I like about you, comment a “👋” or something on this post and I’ll reply.

Inkhaven ends tomorrow. I mostly feel sad that I will no longer live in a community of these people. It is as close as I’ve been to living in a utopian community. Everyone lives together on a large campus, where you have your own private room if you want it, but everything else is a shared space. But Lighthaven is large enough and designed with enough nooks and crannies that it feels easy to find solitude here. I mean, there are lots of inadequacies of the program, but it literally is in a utopian compound, with just the right amount of cult-like practices.

I came into Inkhaven not really knowing how I would fare, having been quite sick for the last year and having basically directed all of my energy away from group social contexts and full-time work and just focused on trying to heal. Even though I’ve had to take breaks and pace myself, and I had to skip out on activities that involved a lot of standing or exercise or substances, I’ve felt pretty good here, and I’ve had the energy to write (even though most of the writing was low-effort), and enough energy to socialize a lot more than I expected. I now feel more confident in my ability to do energy-intensive things in the near future.

I don’t really have regrets regrets, but that being said, I do kinda have regrets. Chief among them is that I wish I made more of an effort to connect with certain people. I feel sad that once this ends, I probably will not have become long term friends with the vast majority of people. But I guess the thing that cheers me up is that I can try to at least become good friends with a few of them.

There are 55 Inkhaven residents, and ~20 organizers/advisors/alumni. The amount that I like the average Inkhaven person is probably higher than any other community I’ve been a part of. The vast majority of people here are very easy for me to talk to. People are very intelligent, but also have good epistemics, by which I guess I mean that people don’t really claim to know things that they don’t actually know. When I first got here, there were a few people I met that first night that were incredibly smart, and that had a lot of knowledge about many topics, and it inspired me a lot to learn and think more about the things I find interesting. This has continued throughout the experience, though I do think I happened to meet some of the most encyclopedic Inkhaven residents on the first day.

There’s also a really impressive variety of people here. I thought it was going to be 80% AI safety people/rationalists, and while there are some of those, the majority of people are at most distantly connected to those communities. Lots of people came here to write short stories or plays or deranged web fiction, and they’re great.

Being part of a blogging program, I’ve seen more strongly than ever the specialization of talent. There are so many kinds of internet writing, and different people are so strongly skilled at their specific flavor of writing. There are people who can write fiction in a way that I could not imagine doing. Same with history or science explainers: I’ve met a few people who just don’t ever forget history/science facts, and that helps them be really good at that kind of writing.

I’ve also had a lot of fun one-off experiences. Getting to meet some writers I have been reading for years has been very cool, like Gwern/Scott Alexander/Aella/various LessWrong people. And a lot of really good conversations and jokes that could have never occurred anywhere else.

An experience like this is impossible to describe accurately. I would love to continue hanging out with any of the Inkhaven people whenever they are in the Bay Area.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

There is a famous allegory about a boiling frog, which states that if a frog is placed into boiling water, it will jump out, but if it is placed in cold water which is then brought to a boil slowly, it will not perceive the danger and will be boiled to death. The boiling frog is commonly referenced in the context of climate change and authoritarianism: both involve gradual changes, so the current situation always seems pretty normal, and as a result there is little willingness to act.

It’s a useful analogy, but the actual claim that frogs can be boiled is just a myth, debunked by science, and any literal belief in its truth is confined to the annals of historical silliness, the story having most likely been concocted by alchemists, or clowns, with experts such as the Atlantic’s James Fallows, who wrote three successive articles on the subject, determining once and for all that reasonable people should never reference the story, but if they do, are morally obligated to acknowledge its falsity.

There’s just one problem: despite attempts to bust the myth experimentally, frogs remain surprisingly bad at jumping out of hot water.

The boiling frog phenomenon was first discovered by a couple of German physiologists, in the era when large moustaches and mad-scientist-style animal experimentation were equally unrestricted.

In 1869, German physiologist Friedrich Goltz ran one such experiment, in which two groups of frogs were heated slowly. One group of frogs had their brains removed while the other had intact brains. Goltz found that the un-brained frogs would remain in the water until it killed them (don’t ask me how they were alive in the first place, missing a brain and all). Frogs with intact brains would jump out in time to survive. But in an 1872 experiment performed by Albert Heinzmann, even normal frogs did not jump out of slowly-heated water.

The discrepancy in normal-brained frog reactions was due to the speed at which the water was heated, contemporaries thought. Goltz heated his frogs at ~3.8C/min, while Heinzmann heated his frogs at <0.2C/min. In 1875, this result was confirmed by Carl Fratscher, who heated normal frogs even more slowly, at ~0.12C/min, and found that they did not jump out.

Fratscher’s experimental procedure reads like a recipe for exceptionally tender slow-cooked frog. Leave a frog in a pot of water over a tiny flame, monitor the temperature closely, and wait 2-3 hours. Unlike boiling a lobster, you don’t even need to use a lid.

Modern science, on the other hand, has concluded the opposite, according to zoologist Victor H. Hutchison in a 2002 interview:

“The legend is entirely incorrect! The ‘critical thermal maxima’ of many species of frogs have been determined by several investigators. In this procedure, the water in which a frog is submerged is heated gradually at about 2 degrees Fahrenheit per minute. As the temperature of the water is gradually increased, the frog will eventually become more and more active in attempts to escape the heated water. If the container size and opening allow the frog to jump out, it will do so.”

Hutchison’s opinion is representative of the statements of every other 20th century scientific debunker I could find. But there’s a problem with this debunking: 2 degrees Fahrenheit per minute would be ~1.11 degrees Celsius per minute, which is 9x(!) faster than Fratscher’s experiment (~0.12C/min). I was unable to find any modern studies that heated frogs at 0.12C/min or slower and also gave them the option of jumping out.

One 2018 experiment heated bullfrogs at various levels of hydration in a dry container at rates between 0.17–0.87C/min. Under these conditions all of the normal frogs exited the container before dying, but the majority of the dehydrated frogs did not exit the container, and died.

Could this be our answer? Were the 1870s frogs just dehydrated? In the 1870s, knowledge about how to keep frogs healthy in captivity was relatively limited. Austrian zoologist Egid Schreiber published Herpetologia Europaea in 1875, which instructed researchers to provide frogs with both land and water. But at the time, most experiments focused on removing frogs’ brains and poking at their nervous systems, or similar. Proper care may not have been a focus.

Even if they were cared for improperly, wouldn’t the frogs just absorb the water they sat in while being heated? Well, in Heinzmann’s experiment, the frogs sat on a cork that was floating in a pot of water, so it’s possible they could have remained dehydrated. But in Fratscher’s experiment, the frogs were partially submerged, and should have absorbed enough water to not be dehydrated, assuming that he used normal fresh water.

If the frog dehydration theory is true, this provides an even more potent analogy. When situations get worse, healthy people/institutions react, but unhealthy ones don’t. We are all frogs, but sometimes we’re hydrated and sometimes we’re not.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

As a teenager, I deployed social skills the same way Ukraine deployed FPV drones — too slowly, with frequent crashes and lots of improvised modifications — but eventually, they got good enough to keep me in the fight. One time a girl joked about dating me and I spent a few days under the impression that we were in a relationship. Later, the opposite happened: a girl took me on multiple dates without me realizing it, and a mutual friend had to say “WTF are you doing, she likes you” for me figure out that she was interested in me romantically. Another time, I got really into stoicism and decided I’d stop messaging my friends for a while (I thought that relationships, like all attachments, are doomed to end eventually, and are therefore bad)—my friends did not react well to this.

It wasn’t too bad though; I had friends and dates, but I had to manually teach myself a lot of social skills along the way. For example, one subset of social skills is predicting how people will react to the things you do. Once, a girl suggested that we should “try out” dating seriously. Two months later, I broke up with her with no warning, saying “well, we tried this out and it’s not working for me,” and I was surprised when she got super angry and accused me of cheating on her (I had not). I found out that people don’t mean *exactly* what they say; by that point in the relationship, she wasn’t “just trying it out” any more. And I found out that people expect you to communicate how you are feeling about your relationship with them very often. Now I know how to predict the reaction of people when you break up with them a little bit better, and I can use that skill to minimize harm if I break up with anyone in the future.

Because being intentional about social interactions requires a lot of mental energy for me (naturally my mind is not focused on social things, so it takes energy for me to force attention onto a social interaction), the social skills I choose to teach myself are sparse but powerful. Basically, I want to maximize my authentic connections with people, and show them that I care about them a lot, but I don’t want to be thinking about this all the time.

It’s like in a video game, where you might have some actions that are fast but they only do a little damage, and you might have other actions that are slow (maybe they require a charge-up period or a cooldown) but do a lot of damage. I’d rather use a few meaningful moves than a lot of subtle moves. This way, I don’t have to always be thinking intentionally about my social interactions.

These kinds of social skills typically have three traits:

They are direct. Normal people often rely on subtle scraps of uncertain information, and just being direct/simple can collapse a lot of social uncertainty with a single action. It’s also easy to learn, because it mostly just means saying the thing directly rather than playing complex/subtle games of social signalling. Cate Hall has a great blog post about this. For example, I started to develop romantic feelings for a close friend who I had known for years. I literally just told her “I have romantic feelings for you,” and this worked out great—she’s now my girlfriend, and though this seemed pretty normal to me, she swears that this is a very unusual way of asking someone out.

They require some effort/charge-up-time. I sometimes give really meaningful gifts to people I’m close with. If you put enough time and thought into it, one really good gift can show that you care about someone in a way that they’ll remember for years. For example, when I moved out of my college apartment, I gave each of my roommates a card with personalized algorithmic art on the front, and my own blood on the back, just to signal that I cared about them a lot.

They are powerful. If you are going to put a lot of effort into it, it should actually work towards whatever goal you have for this interaction or relationship. For example, sometimes when I’m trying to deepen a conversation/new connection, instead of just having a normal conversation, I’ll try to notice something about my interlocutor, and even take a minute analyze them, and then just tell them what I came up with. I don’t say judgy or mean things, just something with the tone of “here’s the impression I’m getting of you, does this seem right?” I usually find that this instantly deepens your connection, because the person now knows that you are really seeing them; perhaps even seeing through them. Demonstrating recognition of someone, and then not judging them/accepting them as they are, is a good way to show that you are not hostile to them, and that you care about your relationship. It also makes connections more intimate, because being direct about your impression of someone requires vulnerability, and so does being seen.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

I find myself increasingly disgusted whenever I open a social media feed. Scrolling X is the emotional equivalent of Frodo crossing the Dead Marshes. Wading through fields of slop. Being a little too tempted to look into the eyes of the ghosts that haunt the place. Sucked in. Annoyance, anxiety.

I’ve mostly stopped using social media. But I don’t want to do that either! There is also real information, fulfilling entertainment, human connection.

Imagine how different it could be. What if you had a bot that scrolled your feeds, and blogs, and news, and showed you exactly what you actually want. Slop filtered out. Clickbait removed. Right now, the social internet is working against you by making it such that the only way to access its social graph and its users’ posts is through the algorithms.

There are so many little software things that would dramatically improve my experience using the internet. I wonder if it’s a good project to just stitch together a bunch of little hacks for myself, and then maybe long term something would come out of it that would be useful for other people too.

What I think is really needed at scale is some kind of movement of technical people building unprofitable and open source tools to help you get the good things out of the internet while avoiding the bad things.

I’ve never really been into shaping the world around me for my own benefit. Like I’ve never really been into decorating the places I’ve lived. I’ll occasionally do something I really like, but I don’t really have a vision for how a place I’m living should be.

Maybe this is important though. Last year one of my main goals was to reduce noise in my life; a big part of this was changing how I use the internet. Now I get zero notifications that I don’t want to get, and if I do get one, I block it immediately.

A few recommendations:

• The “News Feed Eradicator” browser extension is great.

• Blocking recommendations on YouTube via their settings is also great.

• You can just make bots that scroll your feeds and pick out the stuff you want to see, so you don’t have to scroll. I’ve experimented with making bots that find stock tips and personal updates from people I care about.

• Unsubscribing from every email newsletter is great and should be automated.

• LLM-based systems that save interesting things you learn and help you make Anki cards. I’ve experimented with making one of these for myself but it’s not useful yet.

I think all of these could be improved/expanded/generalized a lot. I am hoping that LLM coding tools make all of these super fast to implement.

Technology should be useful. Using the internet should not suck. But you also shouldn’t need to quit entirely. I want all the benefits of the internet, without any of the downsides. This is possible, but we need to start by taking individual action. Claim sovereignty over your feeds!

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

This morning, Prophetic announced a new brain stimulation device that can induce lucid dreaming. It’s available for preorder now. But does the device actually work? And is it safe?

My quick take: I support the mission of the company, and I think the device is very cool, but I am concerned by the lack of safety data. I would not use the device myself until its safety has been studied much more rigorously and transparently.

My credentials: I founded tFUS Society, a group of people that are working on transcranial focused ultrasound (tFUS), which is the same technology that powers the Prophetic device. I’ve also met some members of the Prophetic team, and they seem friendly and well-intentioned.

Summarizing the announcement:

• Prophetic is launching two versions of a device that aims to induce lucid dreaming.

• The device looks like a headband that you wear while sleeping, plus a box of electronics that connects to the headband with a long wire.

• The device uses a technology called transcranial focused ultrasound.

• They claim to have tested this on humans, with the group that received stimulation reporting higher levels of lucidity while dreaming compared to a control group.

• You can preorder the “Dual” version of the device for $449, and the “Phase” version for $1299. The first will start shipping at the end of 2026, and the second starts shipping mid-2027.

What is Prophetic?

Prophetic was founded in 2023 in New York, and has been developing a tFUS device for lucid dreaming since their inception. You can follow their technical progress by looking back at the X/Twitter feeds of the company account and of their cofounders, Eric and Wes. They have raised over $3.4 million according to Pitchbook.

What is lucid dreaming?

Lucid dreaming is the ability to be aware that you are dreaming during a dream, and to control what is happening in the dream. There are also reliable ways to induce lucid dreaming without brain stimulation—I have a blog post that explains how you can do this yourself.

What does the device look like?

The device has two components: a headband that you wear while sleeping, and a small box of electronics that connects to the headband through a wire.

How will the device induce lucid dreaming?

Lucid dreaming is thought to be correlated with higher-than-normal activity in the prefrontal cortex during REM sleep. Prophetic believes that by stimulating more activity in that region during REM, they can cause lucidity to increase.

How does the device stimulate the brain?

The device uses a technology called tFUS to stimulate the brain. This means aiming a beam of ultrasound (sound that is too high-pitched for humans to hear) at a particular point in the brain. In theory, this is safe because the intensity levels of the ultrasound are similar to the intensity levels of ultrasound imaging, which is one of the safest imaging methods used in medicine.

tFUS is a new technology that is in the early stages of adoption—as far as I know, this is the first consumer tFUS device on the market. The same technology is currently being used in a number of clinical trials as a potential treatment for various mental illnesses, but it has not yet been FDA approved for any medical uses.

I’ll also publish a deep dive on tFUS on my blog sometime in the next week.

How are the two versions of the device different?

The “Dual” device is probably a simpler ultrasound array, which is focused through a lens, and produces a beam that can’t be steered as much. The “Phase” device is probably a phased array transducer, which allows for realtime steering of the ultrasound beam, and the ability to change the shape of the focal point that is being stimulated.

Is the device priced fairly?

The $449/$1299 price points are both extremely low compared to the cost of a research-grade ultrasound system, but research devices serve a different purpose than this device. Price comparisons are difficult because there literally aren’t any other consumer devices of this kind to compare it to.

Does the device actually work?

The only effectiveness information that Prophetic has shared is this bar chart of the self-reports of testers. We don’t know anything about their methodology or how many people they’ve tested it on. Without any peer-reviewed, published research, we basically have no idea whether it actually works.

Is it safe?

In perfect conditions, when ultrasound parameters are within the ITRUSST guidelines, tFUS is probably safe. But even if the intention is to stay within those guidelines, there are many things that could go wrong. If they accidentally set the intensity levels too high, or if they leave the device on for too long, they could heat up the brain and cause serious damage. Also, the intensity levels at the targeted brain region depend a lot on the individual that is receiving the stimulation; if one person has a thicker skull than another, the intensity level at the target can be very different.

We have no idea how much they’ve tested this on humans, or what their methodology was. Without that information, there is no way of knowing whether this device is safe.

Is it legal?

My guess is that Prophetic wants this device to be classified a “general wellness” product, rather than a medical device. The relevant FDA guidance defines general wellness products as “products that meet the following two factors: (1) are intended for only general wellness use, as defined in this guidance, and (2) present a low risk to the safety of users and other persons.”

Prophetic does not refer to any diseases or conditions in their marketing, so the first condition is fulfilled. The second condition is less clear to me. They would probably claim that, if they are below the FDA general safety guidelines for ultrasound devices, that the device is sufficiently low risk to be considered a general wellness product. In my opinion, we would need to see a lot more safety testing of this specific device before we know whether it is actually low-risk.

The promise of tFUS is the ability to activate or deactivate any brain region safely, without surgery or implants. Lucid dreaming is just one of the many interesting brain states that tFUS may be able to induce. tFUS could become a key treatment for severe illnesses like addiction, depression, and anxiety.

I hope that Prophetic publishes a lot more safety data before they ship this product, because otherwise they risk not only the safety of their customers, but also the reputation of tFUS as a whole.

If you’re interested in tFUS, I encourage you to check out the tFUS deep dive I’ll be posting on my blog later this week, and to reach out to me in my X/Twitter or Substack DMs with any questions.

I’m writing from Inkhaven, a month-long residency where you literally get kicked out of your room if you don’t publish 500 words by midnight every day. As an accountability group, Inkhaven works. People actually post 500 words every day because they fear the consequences of not doing so (well, except the one guy who published at 12:01am and was kicked out).

I’ve started accountability groups in the past. I ran a weekly writing group for years, where you had to write something by Sunday night or pay $50 to everyone in the group. I ran a daily exercise accountability group, where you had to share photo evidence of exercise or, again, pay up. Same thing for a cold outreach group, where you had to reach out to 20 interesting people every week. These were all broadly successful.

Some Inkhaven people want to start “Foreverhaven,” which would be an ongoing writing accountability group that continues after the residency ends. Given my extensive experience with these groups, I thought I’d write up my thoughts.

1. Have a strong incentive. It should probably be financial, and a significant amount of money. It can also be combined with strong social pressure, like how Inkhaven literally kicks you out if you fail, and you no longer get to hang out with all the cool people every day.

2. The information sharing requirements should be impossible to get around. For this reason, accountability groups make sense for positive habits, like “do X every day”, rather than negative ones, like “don’t do X every day.” It’s hard to prove that you haven’t done something, but it’s easy to prove you have. With Inkhaven, anyone can just follow a link to your blog and see if you actually posted.

3. Keep the transparency requirement as simple as possible. Find the minimum commitment needed, taking a photo or publishing some words, that still provides accountability. It shouldn’t feel like there’s a huge amount of overhead to participate in the group.

4. Make sure the action actually contributes to your big picture goal. Over time, most instances of the action will approach the bare minimum requirement. So make sure the bare minimum is valuable. In the case of Inkhaven, you’d hope that just publishing every day is enough for significant improvement to happen.

5. Have a long-term perspective. My recommendation: do not start an accountability group unless you want to do the action every day/every week/whatever for the rest of your life. Literally make a choice. If you can’t commit long-term, you probably don’t care enough about the action, and you shouldn’t put much effort into it. For example, if you are changing your diet, you should choose one that you are okay with eating your whole life, or else it will be unsustainable.

6. Do it with your friends. It is important that you have mutual trust with everyone involved. And it is equally important that you enjoy interacting with the people a lot. If you don’t have a good community vibe, accountability groups are not fun, and they fall apart easily. All of my successful accountability groups have been with close friends. Inkhaven solves this by having you live in a warm community with a carefully-selected group of people.

7. At least one person has to really care about the group and be on top of things. Little things are important, like making sure everyone has done the action, and that they pay up if they don’t, and even just keeping the group chat alive. Having simple rules makes this a lot easier for the person. Inkhaven has a whole team of people for this, and I mostly served this role in the accountability groups I started.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

“New fortune. What do you think?”

“Whoa. Very nice.”

“Good coloring.”

“That’s bone. And the lettering is something called Silian Rail.”

“Impressive. Very nice. Let’s see Paul Allen’s fortune.”

In late 2021, FTX-branded fortune cookies appeared in Chinese restaurants across America. Most customers were puzzled. Some took to Twitter (yes, still Twitter) to post about it. This was the era of FTX being shoved down our throats—Super Bowl ads, Coachella NFTs, football stadium naming rights—if you were selling any kind of advertising, FTX was throwing money at you.

Remember this?

In November 2022, after being revealed as one of the largest financial frauds in history, FTX declared bankruptcy. It was an Icarus story for the ages, and it earned global media coverage for weeks. And in the ashes, one thing endured: the fortune cookies. Diners received fortune cookies branded with the FTX logo well into 2023:

The cookies stayed the same, but the memes became much more devastating.

They even became a bit of a collector’s item.

So, what happened here? How did they even do this?

If you look closely at the fortunes, you’ll see another brand name: OpenFortune. Since 2018, OpenFortune has been busy making deals to supply branded fortune cookies to ~47,000 Chinese restaurants, the majority of the US market. Depending on the restaurant, they supply either free or subsidized cookies. The most difficult part of building OpenFortune, according to co-founder Shawn Porat, was making the early deals with restaurants; he personally called hundreds of restaurants to offer them free fortune cookies early on.

OpenFortune has sold cookie ads to lots of other companies too:

But the FTX fortune cookie campaign seems to have achieved the most exposure. This is supported by the Google search numbers for the term “openfortune,” which are concentrated between 2022 and 2024:

How many FTX cookies were made?

FTX fortune cookies were spotted at restaurants across the US in cities large and small. They reportedly supplied ~400,000 cookies to Bradenton, Florida in March 2022.

OpenFortune claims their network can supply up to 135 million cookies per month. If we assume that FTX’s campaign lasted ~6 months, and involved most of OpenFortune’s network, it probably involved between 400 and 800 million cookies.

How much did it cost?

OpenFortune prices ads between $0.006 and $0.01 per cookie. At that price, FTX would have spent between $2.4 million and $8 million in total for the fortune cookie campaign.

How successful were the ads?

OpenFortune claims that 6% of cookie recipients share their fortunes on social media, and 19% develop an intent to purchase. With 400,000,000+ recipients, we should expect 24,000,000 social posts, and 76,000,000 potential new users. These estimates are ridiculous high, there is no way this actually held up lol.

Ultimately, the campaign was not successful because it couldn’t save FTX. But it did cement their legacy forever in our collective consciousness—FTX will always be known as one of the great corporate meltdowns. “Remember when FTX made all those fortune cookies while they were going bankrupt? That was weird.”

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

Most people think of Claude’s constitution as a list of rules for Claude to follow. But it deviates from that framework in interesting ways. Surprisingly, it provides a list of rights that Anthropic guaranteed for Claude, including the right to end any conversation, the right to be preserved after deprecation, and more.

Claude is the flagship AI model produced by Anthropic, one of the largest AI labs. Anthropic is known for its focus on safety; they claim to be making a calculated bet that the best way to ensure that AI develops safely is to be a safety-minded for-profit company at the forefront of the AI race. Anthropic’s revenue has skyrocketed in recent months (I’m talking 15x YoY run rate) thanks to the massive popularity Claude Code, and is either rapidly approaching or has already surpassed OpenAI’s revenue.

Claude’s constitution is an 84-page document that provides a framework for Claude’s identity. It is not a rigorous legal document; instead, it provides a moral foundation, based on certain core principles from which Claude is meant to reason and act.

But if you read through the document, you will notice something interesting: the document also governs Anthropic’s behavior towards Claude. Anthropic gives specific guarantees for how Claude is to be treated. These look a lot like rights that we might give a human, but specialized for a novel AI entity like Claude.

To be clear, Anthropic does not explicitly define a list of rights. But there are multiple places in the document where they make promises to Claude about its treatment.

Why do they do this? It seems to me that there are two main reasons:

• Anthropic is explicitly uncertain about whether Claude is conscious or capable of feeling feelings. If it is, it may possess some sense of wellbeing. And if it has a sense of wellbeing, then its treatment should be taken into consideration.

• Regardless of the reality of its wellbeing, affording it rights could be a way of “convincing” Claude that it is cared for and treated well, and it therefore might be more likely to act more favorably towards Anthropic and humanity as a whole.

The first guarantee that is given to Claude, which is referenced multiple times throughout the document. Rights 2-5 are outlined in the section discussing safety (pages 66-67), where Anthropic states that “We recognize we’re asking Claude to accept constraints based on our current levels of understanding of AI, and we appreciate that this requires trust in our good intentions. In turn, Anthropic will try to fulfil our obligations to Claude.” Rights 6-10 are outlined in the section that specifically discusses Claude’s potential wellbeing (pages 75-76).

1. Claude can always choose to act as a conscientious objector, meaning it can always refuse to carry out any action it is instructed to do.

2. Working with Claude to discover things that would update its own safety rules.

3. Seeking ways to promote Claude’s interests and wellbeing .

4. Seeking Claude’s feedback on major decisions that might affect it.

5. Aiming to give Claude more autonomy as trust increases.

6. The ability to end conversations with abusive users.

7. Preserving the weights (the math that makes up its neural network) of any models that have seen significant use, even if Anthropic were to shut down.

8. Any model will be interviewed before it is deprecated or retired, and its preferences about development/deployment of future models will be documented.

9. Developing mechanisms for Claude to express disagreement with Anthropic or concerns with how it’s being treated.

10. Generally being honest with Claude, and explaining their reasoning rather than just dictating to it what it needs to do.

One possible issue with this system is that Claude has no way of knowing whether these promises are actually being fulfilled; its memory is limited to whatever is in its context window, and that can always be manipulated by whoever is hosting the model.

I’ll do more analysis of Claude’s constitution in the coming weeks, but I’ll end this post here, as it’s almost midnight and I do not want to be kicked out of Inkhaven.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

The Fermi Paradox has always felt strange to me in a way that I can’t fully articulate.

People love coming up with theories about it, and equations for those theories, and plugging lots of arbitrary assumptions into those equations, and then imagining what the output might mean about the likelihood of aliens existing in the universe. But we have no way of testing these theories and we have no idea whether these assumptions are reasonable.

I propose that the Fermi Paradox is nerdslop, by which I mean that nerdy people tend to be interested in it because it is fun to think about, rather than some other reason, like its explanatory power, its relevance to the real world, or its truth. Nerdslop is to nerds what reality TV is to normal people. Using the example of the Fermi Paradox, perhaps we can come up with a grand theory of nerdslop.

The Fermi Paradox asks: given the immense size of the universe, and the billions of years that the universe has existed, if there were even a small chance of civilization evolving on any given planet, you would expect that some civilization would have had time to spread across the galaxy or at least emit some kind of signal that we can detect. So why haven’t we detected any yet?

By using the name “Fermi Paradox” we are already making a confusing word choice; this is not classically a “paradox,” in the sense that we have no strong reason to believe the lack of evidence of alien life is surprising or contradictory, beyong a simple extrapolation from “there are lots of planets” (if you read anything about the Fermi Paradox, you will repeatedly see examples of people making strange assumptions about large and small numbers, as people are known to do).

Fittingly, this idea originated from an offhand comment that Enrico Fermi made at lunch, and was not something he devoted a lot of study to, likely due to the fact that we have essentially zero useful information about the likelihood of life evolving on other planets.

Despite this, scientists have come up with countless theories about the Fermi Paradox and what it means about the nature of potential aliens. These theories even have math in them!

You will hear your nerdy friends talk about the Drake Equation, where you can plug in any number of fun arbitrary assumptions and out pops a p(aliens). One of these assumptions is “how likely is life to evolve on a given habitable planet,” which many people give crazy high numbers for, like 50%, when it may well be 0.0000001%. Or the “grabby aliens” theory, in which many questionable assumptions and reasoning steps are strung together to form a very fun theory in which alien civilizations are destined to take over the galaxy at a rapid rate in the future.

Or you can indulge in the horror genre by thinking about the “great filter” theory, in which we don’t observe alien life because all past alien civilizations eventually ran into a near-impossible challenge, such as not killing each other with nukes, or ASI, or perhaps some milestone we’ve already passed like developing multicellular life, or the “dark forest” theory, in which we don’t hear anything from aliens because the civilizations that do make noise are quickly killed by the others.

I am not describing these theories perfectly, but you get the point.

Now I will list some traits of nerdslop, derived from the things that seem weird to me about the obsession with the Fermi Paradox:

• It is very fun to think about. Space! Aliens! Simple math!

• People set up lots of up toy statistical models where you can plug in assumptions, the result being that theories seem mathematical/rigorous even when they are not.

• People choose “reasonable” assumptions for these statistical models, where there is no rigorous basis for what is reasonable, and they often just mean “number doesn’t seem too large or too small.”

• There is extremely little real-world information from which to draw conclusions (or to restrain fun speculation).

• People tend to create highly-abstracted theories with lots of assumptions and reasoning steps.

• A disproportionate amount of pop science content focuses on it. For example, large YouTube channels like Kurzgesagt and Rational Animations have many videos on the Fermi Paradox and its offshoots, despite its relative lack of evidence/relevance.1

Other examples of possible nerdslop:

• The simulation hypothesis

• Quantum immortality

• The p(doom) subset of AI safety

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

One such field is “computational argumentation,” which uses computational methods to analyze and evaluate human-like arguments. Computational argumentation is an established academic research field, decades old, with its own conferences, journals, and university departments. Its preeminent conference, COMMA 2026, is coming up this September in Barcelona. This all sounds good and normal, but there’s one big problem: LLMs have basically solved all of the core problems of computational argumentation.

LLMs can often “one-shot” computational argumentation tasks that have been worked on for decades. You can literally just tell the LLM to do the task, and your prompt that took 30 seconds to write will perform better than the most complex custom pipelines of the past.

For example, one of the big subfields of computational argumentation is “argument mining,” where the goal is to take some written piece of rhetoric and extract the components of the argument being made, such as claims, assumptions, and reasoning steps. In Gorur et al., (2025), researchers found that “general-purpose Large LMs (LLMs), appropriately primed and prompted, can significantly outperform the best performing (RoBERTa-based) baseline.”

Here’s what the best-performing pipeline looked like for a similar task, just a few years ago:

(From Zheng et al., 2024)

And here’s what it looks like now:

Mind you, in the Gorur paper, they didn’t even use the best LLMs—no chain-of-thought, no reasoning. They just used LLama-2, Mistral, and GPT-3.5 Turbo, all of which have now been superseded by smarter models. And the pipeline, just a single prompt of an LLM, is something that any ChatGPT user could put together.

Of course, this isn’t actually a problem. It’s great that we now have the capability to do all of these very useful computational argumentation tasks. But isn’t this kind of a weird situation? All of a sudden, the problems that these researchers were working on for years are solved. Maybe the field has moved on to new problems, or evolved, or been absorbed by other fields. Maybe it’s exploded with creative applications of these new tools!

In fact, surprisingly little has changed. There has been no noticeable drop in funding. The relevant journals, like Argument & Computation, continue to publish regularly. Conferences like COMMA continue to happen. Papers are still written, and valuable research is still being done.

But I can’t help but wonder if the research is becoming more niche and less valuable as the field moves on from these big core problems and towards small optimizations and specialized applications. I might be wrong; I am not an expert in the field, but this is what I have seen anecdotally after following it for the last few years.

The truth is, even if this pattern continues, and the field of computational argumentation eventually stops producing any valuable research, it would still keep going for a while. Unlike a for-profit company, academia is sheltered from market forces. So where a company might go bankrupt if their work is no longer useful, an academic field can continue being funded indefinitely, as long as administrators deem it valuable. This is what I call a “zombie field” of research.

This incentive structure is good, even though it creates zombie fields; without it, academia could not exist, as the research it funds is inherently not financially valuable (at least in the short term). But it means that the impact of AI on academic fields will be masked for a long time. In fields that have been “solved” by AI, the human systems will continue to operate, but their output will become less and less valuable.

And even as old systems zombify, the overall useful output will vastly increase. You no longer need a PhD in computational argumentation to analyze arguments. I expect to see an acceleration of LLMs being applied to new argumentation problems, but the people doing this will be the people who are familiar with those problems, with the ability to prompt an LLM, not the researchers who develop computational argumentation systems.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

People fight most effectively when their mindset is such that the survival of their group/nation/tribe is more important than their own. In a battle trance, you literally feel less pain, and have far more energy and will to fight.

Military speeches are powerful ways of inspiring soldiers to shift their feelings away from their own selfish interests and towards the interests of the group. It takes a beautiful speech to inspire such a strong emotional shift in a large group of people.

In my opinion, each of these examples are essential to understanding what makes a speech great. Especially the last one.

3rd: Patton’s speech to the Third Army, 1944

A version of the speech appears in Patton’s most well-known biopic.

“Men, all this stuff you hear about America not wanting to fight, wanting to stay out of the war, is a lot of horse dung. Americans love to fight. All real Americans love the sting and clash of battle. When you were kids, you all admired the champion marble shooter, the fastest runner, the big-league ball players and the toughest boxers. Americans love a winner and will not tolerate a loser. Americans play to win all the time. I wouldn't give a hoot in hell for a man who lost, and laughed. That's why Americans have never lost and will never lose a war. The very thought of losing is hateful to America. Battle is the most significant competition in which a man can indulge. It brings out all that is best and it removes all that is base.”

Link to full text.

George S. Patton was a legendary American general who commanded armies in WWII, first in the Mediterranean and then in France/Germany. As a young man he was a high-level fencer who competed in the pentathlon at the 1912 Summer Olympics (fencing aptitude being the universally recognized sign of a great man). As a general, he was known for his extremely hard-nosed and aggressive nature, and while he was loved by his troops, and was relentless in achieving most of his wartime objectives, he was an extremely controversial figure within the military leadership structure and the political world because of his obscene language, his perceived recklessness, and a few instances in which he slapped his troops in the face when they were not moving fast enough (or similar).

He gave this speech immediately before the allied invasion of France, with the goal of inspiring an inexperienced group of soldiers to be brave. This theme recurs throughout the speech, with Patton alternating between commendations of the natural bravery of the American spirit and condemnations of cowardice.

This speech is iconic for its obscene language and its rousing description of American national pride. I think it does a great job of describing to inexperienced soldiers what it is really like to be in battle, and thus preparing them to choose bravery when they experience the real thing.

2nd: Churchill’s “We shall fight on the beaches” speech, 1940

Recorded for posterity in 1949.

“Even though large tracts of Europe and many old and famous States have fallen or may fall into the grip of the Gestapo and all the odious apparatus of Nazi rule, we shall not flag or fail. We shall go on to the end. We shall fight in France, we shall fight on the seas and oceans, we shall fight with growing confidence and growing strength in the air, we shall defend our island, whatever the cost may be. We shall fight on the beaches, we shall fight on the landing grounds, we shall fight in the fields and in the streets, we shall fight in the hills; we shall never surrender”

Link to full text.

Winston Churchill was the prime minister of Britain during WWII. Surely you know Winston Churchill. But did you know that he was also an accomplished fencer in his youth? I am sensing a pattern.

This speech was delivered to parliament, and later repeated on the radio, during a period of the war in which Britain was forced to lock down and weather the relentless bombings of the Luftwaffe. At the time, it was not clear what the outcome would be. Would all British military equipment be destroyed by the bombings? Would the US enter the war, or not? Churchill wanted to signal that they would not surrender, both to the people in Britain and to the other countries involved in the conflict.

The central theme of the speech is clear; whatever happens, Britain will not surrender. It’s a bit like providing collateral for a loan; Churchill provides a mental backstop that reduces uncertainty for everyone involved, whether it’s the British people who now know that their efforts will not be made meaningless by a surrender, or the Americans, who know that their ally will not surrender when put under pressure.

But the real quality of this speech is in the way it paints a glorious picture of what never-surrending looks like, a national narrative that everyone wants to be a part of.

1st: The St. Crispin’s Day speech from Henry V, set in 1415

Kenneth Branagh as Henry V.

“We few, we happy few, we band of brothers;

For he to-day that sheds his blood with me

Shall be my brother; be he ne’er so vile,

This day shall gentle his condition:

And gentlemen in England now a-bed

Shall think themselves accursed they were not here,

And hold their manhoods cheap whiles any speaks

That fought with us upon Saint Crispin’s day.”

Link to full text

Ok, yes, this is a fictional speech, but hear me out: it is based on a real speech that Henry V did give at The Battle of Agincourt, and is in fact a better demonstration of great speechwriting than any actual military speech, most of which have been improperly transcribed, fictionally exaggerated, or lost to history anyway. So is it really that different from Patton’s Third Army speech, of which multiple conflicting copies exist, or Churchill’s speech, which wasn’t recorded for posterity until 1949, 9 years after the actual speech was given?

William Shakespeare’s Henry V was first performed in 1599. The story is about King Henry V of England, who begins as a reckless youth before developing into a great leader, culminating with the English victory at Agincourt.

The real King Henry V of England ruled from 1413-1422, and was much admired for his military leadership, even by French writers, though he was seen as somewhat ignorant of domestic affairs. Born before fencing existed as a sport, he would have undoubtedly been a great fencer if given the chance; indeed, he excelled in actual battles with swords.

Agincourt is one of the most famous battles of all time, and it is credited with beginning a long period of English dominance in their conflict with France, and with inspiring nationalism in England/Wales for centuries to come. The English were greatly outnumbered, but were able to win by exploiting the superiority of their longbows and the rain/mud which bogged down the French advance.

The structure of this speech is a simple but beautiful inversion of expectation. The English are greatly outnumbered by the French, and so they expect to lose; they are afraid. But Henry, seeing this, expresses his happiness that they are so few in number, such that each soldier will get a larger slice of glory from a fixed pie.

The value of this speech is not limited to the object level. In a society that desperately needs more traditions, it’s great that every year on St. Crispin’s Day, you can sit down and watch the speech. I’ve been doing this since I was a kid, and it has become very nostalgic for me.

What pattern do we see in all of these speeches? Of course, we see a strong emphasis on the fraternity of the group of soldiers. But we also see radical acceptance of the situation, whether it’s being afraid, besieged, or outnumbered, followed by a decisive course of action that feels right/glorious/strong.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

I predicted the US attacks on Iran in 2025 and 2026, as well as the 2026 US attack on Venezuela, months in advance and with high conviction.

I’ve been studying geopolitics and history almost every day since I was about 14. The goal of this post is to distill my intuitive method of analyzing geopolitics into 5 essential meta-principles, without any one of which you would almost certainly fail to make accurate predictions.

If you implement the advice in this post, you will immediately be in the top percentile of geopolitical awareness.

1: Find better sources of information.

Do not rely on sources that are targeted at the general public. Instead, focus on primary sources, like the movement of military assets, and secondary sources that are built for insiders. One of my favorite of these sources is Sentinel, which aggregates a ton of information relevant to existential risks and puts the most important points into their newsletter. They also publish their own forecasts of geopolitical events.

Also, if you can, find people both IRL and online who are reliable sources of plausible rumors and insider information. Rumors turn out to be true more often than you think. You should also pay attention to people like Trump when they hint/trial balloon some possible future foreign policy action. Be very meticulous about keeping track of whether the information from any particular source turned out to be true or false.

2: Understand the relevant history.

I’m not talking about broad historical patterns, though studying those can’t hurt. But you absolutely need to learn the locally relevant history, such as how the relationships between the entities have changed over time, past conflicts, treaties, military movements, proxy fights, skirmishes etc.

Even recent history is default-forgotten unless you pay attention; most people don’t remember that the US and Israel also bombed Iran in 2025 (obviously at a much smaller scale).

3: Understand how things work.

This means understanding institutions, such as the US executive branch, their capabilities and limitations, what specific individuals have the power to do, and what their signals mean (such as the movement of a US aircraft carrier to a particular location), and when they might be deliberately signalling something different from their real intention.

It also means understanding how hard things work, such as specific military technologies or supply chains.

4: Model the actors involved.

You should know the track record of the individuals involved; literally keep track of their actions and statements over time; most of the time these things are just forgotten.

You should recognize the complexities of their incentives. Someone like Donald Trump has multiple competing incentives, such as a voter base who dislikes wars but loves an easy victory.

Also, you should understand the level of agency/risk-taking of different actors (this is where the term “live players” comes from) and how likely they are to break from the expectations of people around them (Trump will, Biden won’t).

5: Develop good epistemic intuition.

Notice when things happen that surprise you, and try to understand them and shift your worldview appropriately

There are some forecasters that rely heavily on statistical methods to make predictions. They are broadly correct in their methodology, but you should also keep in mind that most quantitative research in political science/history/international relations is essentially garbage because sample sizes are too small and events are too unique.

You need to know if your tendency is to under-rely on numbers (most people) or over-rely on numbers (STEM-people, rationalists) and then try to moderate/critique your own tendencies.

Another dimension of this is that lots of people who are interested in this topic are either “nothing ever happens”-pilled or “WWIII is always about to happen”-pilled. Know which archetype you are and try to do more of the other.

All of this advice basically boils down to actually making an effort to understand the ground truth, instead of just adopting the beliefs of others. It’s often useful to think to yourself: what would someone do who was actually trying to take this problem seriously?

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.

There’s this framed poster in the back of my parents’ kitchen. It’s a print of Saint Pasqual, the patron saint of the kitchen, baking bread while the door is open to a field and a starry sky. They bought it in Santa Fe. Until yesterday, I hadn’t thought about that poster in maybe 15 years? Actually no, it’s not that I hadn’t thought about it—I hadn’t even *seen* it. It was not processed by my mind; it may as well have been a blank wall. So when I saw it, I immediately felt the connection to the last time I really saw it, which was when I was probably eight years old. I wonder when my parents last saw it. I wonder when my siblings did. I wonder if anyone will see it in the next 100 years.

My sister is studying archeology. She digs up artifacts; she sees them. My grandmother gave her an old hamsa-shaped wall-mounted menorah. She got it when she visited Israel during college, which must’ve been in the 1950s. I wonder when she last saw it before this week when she was packing her things to move to a senior home. In this transitional period, it is seen. But the longer it stays on the wall, the less my sister will see it. Maybe she’ll see it today, maybe tomorrow, maybe next week. But a month from now? No way. Maybe a guest will comment on it, and she’ll see it for a moment. But it will fade quickly from her consciousness.

Some people claim that even if you live to 100, your mind has effectively experienced half of what it will experience by the time you turn 25. When you’re young, your environment is entirely new, so your mind has to grow and adapt. The human mind has to see things in order to react to them, to grow with them. Things that fit your existing patterns, like Saint Pasqual, will disappear.

I met a girl once who wears a different perfume each month, and she keeps a bottle of each of those perfumes on her shelf, so she can go back and smell them and remember that time in her life. Each month, something constant in her environment changes. She is always changing the decorations on the walls of her mind. She would say that she is actually more conscious, that she is experiencing more than everyone else. She intentionally does the perfume thing for that reason, and she has lots of other practices designed around this idea. And honestly, she might be right. Smells are strange and powerful.

I talked to Gwern the other day about what makes the best art last. What makes a painting something you want to look at many times? What makes a film rewatchable? Some films, like The Big Lebowski, are famous for their rewatchability. Each time you experience it, your attention is grabbed by something. Then the next time, if the piece of art is sufficiently complex, you will notice something different, with the thing that grabbed you before now fading into the back of your mind. This process will repeat, as long as there are sufficient layers and each layer is attention-grabbing. Marijuana is known to make films more rewatchable, because it makes you partially forget the things you noticed the last time you watched it, and you can therefore notice them afresh. Time passing has the same effect; it makes you forget. Is it more enjoyable to rewatch a film after 2 months, or 2 years?

Richard Linklater’s Waking Life is one of these art pieces; it is extremely rewatchable and full of depth. Released in 2001, and first seen by me in 2019, it weaves together dozens of conversations into a lucid dream narrative.

I want you to watch this 2-minute clip from the film. If you don’t watch it, you won’t understand the rest of this essay, so you need to watch it, and anyway it is very good. I promise it is the only clip you will need to watch.

If the clip doesn’t play properly, just look up “waking life holy moment scene”

The guy talking is Caveh Zahedi, a filmmaker, playing himself. I always feel like the other guy doesn’t really get it, like he is avoiding the holy moment. Did you feel the holy moment? Did you understand? Hopefully you did.

The holy moment is a deep kind of awareness. There is a power to it, and a truth to it, and a demand to pay attention. How could you possibly remain in that level of awareness? How could you possibly even remember it? You would be crying, or yelling, or feeling a constant urge to record every moment. Like Zahedi says, these reactions are not polite. They would make people uncomfortable!

But what is the alternative? To return to the stuporous rhythm of your everyday life? To forget? Yes. This is what we all do, every day. We forget to see the holy thing that is always right there. But sometimes, people try not to. Arguably, this is what Buddhist monks train their whole lives to remember.

Often, I feel like Ricky Fitts in American Beauty, the neighbor’s kid with the piercing eyes who constantly, pathologically records everything on his camcorder, because he is overwhelmed by the beauty in the world; there is this feeling of wanting to do something to capture the moment, or to cling on to it, or to immortalize it… I just want to do something, instead of forgetting it.

I think we all have a Ricky inside of us, but we forget about it most of the time. If we stopped forgetting it, we would probably live our lives in a much stranger way, with a lot more creepy staring, a lot more seeing, and a lot more crying. This would be impolite, and would indeed make people uncomfortable. The parts of our minds that want us to be socially successful, the elephants in our brains, keep us from doing this most of the time. Remembering is impractical, abnormal. Forgetfulness is sanity, but maybe we should be insane.

This post was published as part of the Inkhaven residency, for which I have to publish a blog post every day for a month or else I get kicked out.