Hello from Morningside Heights! We’ve had a patch of beautiful weather here lately — the sky outside my window is currently a nice shade of dreamy blue.
This week’s post will be a little shorter to accommodate a new section: a summary of thoughts and ideas from last week. I’m hoping some of these can make it into fully-fledged blogposts someday!
Last Sunday was Juneteenth, marking an event tragically delayed by information asymmetry: the true end of slavery in the United States. It took over two years for the Emancipation Proclamation to finally take effect in Texas — compare that to the abortion bans that took place overnight on Saturday.
Today we live in an era where news and its effects travel faster, for better or for worse. And when it comes to information exchange, as with all manifestations of Capitalism, productive increases in efficiency have been accompanied with destructive increases in inequality.
One such increase has been the formation of problematic discrepancies in news access across localities in the U.S., as many small news outlets went out of business in the past decade. That collapse in accountability has led to an apparent rise in elections of scandal-ridden or corrupt politicians.
Information is power — it’s key to calibrating our actions towards building the society we want to live in. During his days as a political prisoner, Nelson Mandela would collect sandwich wrappers from trash cans, because they were made of newspapers. When news is that valuable to your ideals, no obstacle will prevent you from seeking it out.
But here is where another critical inequality has formed in news. While we face far fewer obstacles than ever before to getting the information we need, the gaps in the distribution of these obstacles have widened. Sure, it’s become easier to hear about troubling wars abroad. But it’s become even easier to watch a viral dance video on TikTok. The relative friction between information that distracts and placates us, and information that gives us agency, has grown.
My advice for alleviating this? Donate to Wikipedia. Or your favorite non-profit news source. I will match whatever donations you all make, please DM me if you do give something.
And now for a mixed bag of information from this past week.
💭 Weekly Thoughts
🤖 Sunday
I read another concerning M.L. researcher quote, that I forgot to mention in last week’s post, from MIT Technology Review:
When asked whether Gato was heading toward AGI, they wouldn’t be drawn. “I don’t actually think it’s really feasible to make predictions with these kinds of things. I try to avoid that. It’s like predicting the stock market,” said Kay.
Reed said the question was a difficult one: “I think most machine-learning people will studiously avoid answering. Very hard to predict, but, you know, hopefully we get there someday.”Hopefully? Why are we rushing into developing something so obviously dangerous? Unless he means hopefully we get better at predicting when it may come about, in which case I would ask for more urgency than simply sitting around and “hoping.”
I’m thinking of writing a guide to living life in anticipation of AGI. Should we go gently into that good night? Should we rage against the dying of the light?
Even when I accept my own powerlessness, I struggle to decide whether it’s best to simply take it easy and enjoy what time I have left, or strive to make everyone else’s remaining time easier to enjoy. I feel like writing a well-reasoned guide could help me (and hopefully others) find clarity in these matters.One way we could get around the “getting it right on the first try” problem with aligned AGI might be to launch our subsequent attempts at it into space. Picture solar-powered supercomputers sending intermittent progress reports back to Earth. Then we’d at least have a speed-of-light buffer against any rogue AGI out there, and if we stage our hypothesis-testing right, new iterations closer to Earth could protect us from runaway AI coming back to kill us. So there you go, policy proposal: restrict AI research to a one light-year radius from Earth!
A respite from all these scary robot thoughts came when my devices died and I had to simply let my scheduled Sunday post auto-publish (something I did as a safeguard against missing the deadline). I went to a Russian banya with a close friend whose own phone had recently passed. We enjoyed temperature extremes distraction-free, then went home and watched JCVD — very entertaining.
🧗♂️Monday
I went for a “swimb,” my favorite workout, with a buddy. A “swimb” is my name for an intense bouldering session followed by a few swimming laps. For this round, the gym had opened the pool’s glass ceiling, given how perfect the weather was. I sat in the hot tub for a while with my friend, gazing at the tall skyscrapers revealed by the opening — a moment of bliss.
🧑🚀 Tuesday
Went climbing with another friend. We chatted about Magnus Midtbø’s absolutely bonkers climb with Alex Honnold. And we talked about the legendary Tour de France stage that crowned the current cycling king Tadej Pogačar in 2020: La Planche des Belles Filles. At the crucial juncture, he turned off his heart-rate monitor and rode without a computer, much like, my friend remarked, Luke Skywalker in a New Hope: “Use the Force, Tadej!”
We also discussed this “swimb” notion I’m increasingly fond of, and which I’m hoping to someday develop into a new gym concept: “Swimb Spot.” We had fun brainstorming ideas for what would essentially be a fitness-oriented waterpark, featuring traverses over icy water, epic climbs overhanging deep pools, and plenty of opportunities for acrobatic dives. Coming soon to a warehouse near you.
☕️ Wednesday
The scaffolding in my part of town is so ubiquitous that I went for a walk in the rain without an umbrella and came home dry. Apparently landlords are just putting them up everywhere, by city ordinance, to protect sidewalks from the debris peeling off deteriorating buildings, with no actual intent to repair the aging façades.
I’m thinking of taking a video course. I want to make videos but I lack a solid accountability structure for it. The first I’m keen to produce: Bayesian-Optimized Mocha — using a reinforcement learning algorithm to explore the infinite combinations of ways to prepare mocha (think all the different coffee-making methods, multiplied by all the different ways of consuming chocolate). Then at the conclusion, I’ll share the best approach I found.
🍤 Thursday
I had Bánh Mis for dinner with a coworker. It was interesting to note the convergent use of baguette as a vehicle for fried shrimp in two antipodal parts of the French Colonial world: New Orleans and Vietnam.
My coworker and I agreed that buying Google stock was a good hedge against the AGI future, given DeepMind’s clear advantage in the race.
One thing language models like LaMDa seem to lack is an internal logical inference engine. Approximating human behavior might be better approached by slapping natural language interfaces onto a set of domain-specific internal models. Then maybe we can talk about sentience. After all some neuroscientists believe that consciousness stems from our ability to model our environment.
I’m thinking of writing a blog post about my college reunions last month. I’m particularly intrigued by the tactics Princeton uses: footing the bill for virtually unrestrained carousing — all the buildings are openly accessible and all the booze is free — in exchange for an atmosphere of freedom and nostalgia that is likely optimally conducive to alumni donations.
❓Friday
Here’s a question: throughout history, have the most cognitively powerful humans generally been more destructive, or more productive to humanity?
💀 More Destructive
🌻 More Productive
I went for a twilight bike ride in Central Park with Sasha, as the fireflies were coming out. Something about summer nights in the city feels like a perfect setting for practicing witchcraft. Book Club idea? Read magical fantasy and convene on summer nights in witch / druid costumes. Engage in arcane activities like collecting, feeding, and breeding fireflies.
🕰 Saturday
Networked note-taking tools like Roam and Obsidian are great, but I don’t always find myself automatically taking advantage of their back-linked structure. Lately I’ve been practicing what I call “situation exercises”: writing out all the places, topics, and people in my present, and exploring what they’re linked to in my graph. It was surprisingly gratifying on Saturday: I was able to start reading a book that a friend, who I was about to meet with for dinner, recommended long ago.
Clearly, I’m not someone who likes to leave things out or cut things short, which is one of the reasons I’m fond of Lex Fridman’s indulgently long podcast conversations. I’m curious what the longest one is — Charles Hoskinson went on for over 5 hours. Does anyone know of any longer?
See you all next Sunday!