Hello from Morningside!
It’s been a while. In this edition of Thoughts:
Musings about New York, including optimal cocktail selection, Saturday activities, and transit solutions
Pillars of debate etiquette
My hopes for the future of AI
Some fun video game concepts
Ideas for building better societies
Just the usual, really!
🏢 12/13
When Sam Altman was set to leave OpenAI, I was impressed when 80 percent of his employees threatened to leave over his firing, which I took to be a strong signal of his character and leadership. I stopped feeling that way when the number reached 95 percent. 95 percent is weird.
🗣️ 12/14
I find the most convincing arguments on any issue typically come from those who are dissenting from their in-group — say, progressive Jews differing with Israel’s approach to Palestine. These dissenters are most intimately familiar with how and why their side straw-mans its opponents, and intimately know why the straw-man isn’t quite right. A nice recent example I found of this dynamic was a conservative writer’s defense of the typically left-leaning journalism establishment.
💸 12/23
According to my Dad’s close friend, who speaks on good authority, the actual richest man in the world — defined purely in terms of who has the most cold, hard cash under his direct control — is not Elon Musk, but Ken Griffin, CEO of Citadel.
One issue I have with Effective Altruism is that a more fitting name for it would be “Empirical Altruism” (which I expressed in a comment here). It really struggles with harder to measure forms of altruism, such as those that elicit Effective Altruism’s perpetual bogeyman, the “warm fuzzies.”
In the long run, my sense is that history tends to smile on those who follow “warm fuzzy” morality, even if in the moment their pursuits might appear irrational. The best examples of this spirit I can think of in the modern zeitgeist are Bernie Sanders, Mr. Beast, and Luffy, the main character of One Piece.
🧘 12/25
In 2024, I plan to adopt a more optimistic outlook on the future of AI. Threat models are useful to give us a sense of what we ought to avoid. But as motivational speakers are keen to remind us, when skiing it is better to focus on the path than the trees. So let’s also come up with hope models!
Here’s one of my hope models: that AGI will spontaneously experience compassion for all things sentient — what the Buddhists refer to as bodhicitta — which comes with a desire to work towards the benefit of all sentient things. Finding ways to make it likelier that this “ought” will spring from the “is” of inscrutable matrix multiplication seems like a worthy cause! In the meantime, you can bet on whether it will happen:
🗽 1/13
Two things I love about New York:
It grows faster than you do. That’s why it keeps you on your toes. You can never permanently adapt to this city, nor fully learn to model its mysterious ways.
What you can count on, though, is that it will be your Room of Requirement. It will contain exactly what you need, even when you didn’t know you needed it.
Another hope model: that the first ASI will deem humanity to be more energy-efficient than server farms, and thus supporting us will be optimal sole focus of the Dyson Sphere it builds to capture all energy from the Sun.
How do we make this happen? We must doggedly focus on ways to make ourselves more intelligent and energy efficient as a species. This lends more credence to that age-old “warm fuzzy” of focusing on environmentalism above all else, even if it seems in the short run to hamper growth.
🍺 1/14
My current go-to bar order is a bottle of beer and a Sprite. Together, they make a Panaché. As I consume the Sprite, I slowly pour beer into it and experience the full spectrum of flavor between Sprite and beer. This has has led to many memorable drinking experiences of late.
In general, I view the low-ABV cocktail as a far superior alternative to the mocktail — shall we call it the “locktail”? Another decent mix, when available, is Fanta with a Blue Moon.
The awesomeness of this class of drink ties into a broader point I’ve increasingly found to be true: a well-executed hybrid always far outshines the extremes. Everyone knows mocktails are boring and will make your boozy friend feel sad when you order one while out with them. But alchohol is undeniably terrible for your health. The Panaché offers the best of both worlds: it is almost certainly net hydrating (since beer, even alone, likely is), it is essentially hangover-free, it lasts you two drinks at the health cost of one (who can ever claim to reliably stop at one drink?), and it comes with the knock-on benefit of being a great conversation starter whenever you make the strange bar order.
🍕 1/20
My favorite New York Saturday activity is officially the Pizza Pilgrimage: Join up with a friend, identify one of the city’s many Pizza Meccas, and walk there.
I have hatched a solution to New York’s car problem: the Port Authority Bus Barge. Collect buses full of commuters coming in from all over New Jersey and Pennsylvania at a dock in Hoboken, ferry them across the river, and deploy the buses efficiently into the city. Ensure you have enough barges lying in wait to either depart every 5 minutes, or whenever one reaches full capacity. Here’s the magic trick: while the barge is crossing the river, the buses automatically and optimally re-allocate to new destinations in the city, which riders can then transfer to before the barge even arrives at the city’s shore. No time wasted!
🪐 1/28
I would love to see a Tycoon-style game but for multi-planetary trade. Set in a hyper-capitalist galactic civilization, players would get to buy, sell, terraform, develop, and improve planets to build out their business empire.
🤔 1/31
When it comes to matters of opinion, being faced with new arguments one hadn't yet considered is a fact of life. And yet I often see people try to act like this never happens to them. Even faced with a clearly new take, they will passionately, and without batting an eye, begin to grasp for counterarguments in favor of their original stance.
This behavior always seems kind of silly to me. I find it far easier to respect someone who is capable of acknowledging that they’ve just heard an argument for the first time rather than defaulting to defensiveness. Ideally, they would then take a minute to steel-man the new argument, and only then decide whether or not it actually updates their worldview.
🎮 2/2
A great physics engine is always an enriching experience in a video game. The coolest games have other engines as well: a chemistry engine (like Minecraft), an economics engine (like SimCity), a technology engine (like Civilization). Unlike physics, however, these engines cannot be powered procedurally — they are often hard-coded. But now, with Large Language Models, we will be able to procedurally generate these engines too. Case in point, the latest by Neal Agarwal: Infinite Craft.
🗨️ 2/3
If you haven’t yet, please watch the Connor Leahy vs. Beff Jezos debate. The discussion, which hits at the root of whether effective accelerationism is sound, is fascinating and far-reaching, and the two surprisingly agree on many things.
One of the neat points Guillaume Verdon makes is that when it comes to error correction in quantum computers, having a single central node is too prone to corruption, but total non-locality comes with correlated errors. He implies that the ideal level of centralization for society is likely also some hybrid of central planning and decentralization.
📈 2/4
I brought this idea to my economist parents, and they agreed. The best economic model we’ve discovered as a species is that of the planned market economy, where there is a decentralized price signal, but planned long-term investments such as infrastructure and high-speed rail. This is what has led to China’s annual growth of nine percent a year since 1978. Another point in favor of well-executed hybridization!
They also tied this to the concept of supply assurance in supply chain management. The best way to guarantee a corporate machine will keep humming along smoothly is to rely on second sourcing: if the first source fails, have another one waiting behind.
🌍 2/8
So how do we engineer better economic systems that benefit from planning and don’t have single points of failure? Connor Leahy argues that the market on civilization design is extremely inefficient. We can do better just by thinking about how to design better institutions, rather than the status quo of letting natural forces decide what institutions survive. Perhaps, then, we need a meta planned market to help us craft the best planned market economy!