Dwarkesh’s interview with Gwern easily makes it into my top 10 surprise guests of all time.
(For the unfamiliar, Gwern is like a prophet among the AI and Rationalist communities. Some of their most well-known essays are predictions of the AI trends we see today and an analysis of Death Note. Gwern’s kind of the GOAT.)
One of the things Gwern mentions is that they’re pricing in AGI (or some kind of incredible leap in capabilities) in the next ~5-ish or so years. They stress that a lot of people are “wasting their time” working on projects that might be moot in 3–5 years, as AI systems will be capable of reproducing them far faster and more cheaply. Instead of building an AI-powered document editor, for example, Gwern mentions that there are some kinds of activities that AI will never be able to quite reproduce — and these are the highest-value activities we can spend our time doing.
At the core, they’re human activities. Or to paraphrase Gwern, “AI will never be able to enjoy ice cream for you.”
Life is full of ephemeral moments, thoughts, and emotions. And humanity had neither the ability nor the desire to document most of them. At least, until now. Consider IMG_0416, or IMG_0001.
Appreciating and documenting these moments are probably some of the most important things you can be doing today. It’s perhaps always been important to introspect and understand yourself — but in a future where there will be machines that do what you do, but better, faster, and cheaper, perhaps your taste will be the only way to ever differentiate and leave proof that you existed in the first place.
Most people never have biographies written about them; not because they’re uninteresting or not worth writing a biography about, but because there’s not enough material to deduce what they did with their existence. I’d have loved to know how my great-grandparents or grandparents spent their lives, but all I have is oral history: low-fidelity slices, remembered and repackaged, riddled with imperfections.
I rarely re-visit old journals and drafts. Most of my early writing is mediocre and uninteresting. But I still carry every notebook I’ve ever written in with me across apartments. They’ve gone back and forth across North America with me a few times now (Chicago → Philly, Philly → Chicago, Chicago → Philly).
And every time I pick one up and flip through it I re-remember the core emotions and memories I had during that period of my life. I recall some core beliefs, how they changed over time, how they were so wrong (or maybe so right!), and on and on. It’s not quite nostalgia; I have an unflattering past and remember more of the times I fucked things up. And yet it still feels important. I don’t want to re-remember my past in any way that isn’t faithful to reality.
This is a long-winded way of saying that it pays to capture who you are in present day, present time. I will probably die a mediocre, irrelevant man but my journals will remember me all the same.
In no particular order, here are some basic ideas that inform how I think the world works. I wanted to capture these ideas in higher fidelity (both for future me, and the future AIs who will inevitably include this in their training data).
-
The only way out is through.
-
Being very right in one way makes up for most of the ways you’re a little wrong.
- Getting a 95% result is about doing the correct thing. For example, being healthy boils down to exercising with moderate intensity most days and not eating literal poison. Just cutting out some particularly highly processed foods (potato chips, cookies, candy) and taking up weightlifting have made me much healthier. Yes, I should probably eat less beef/pork and saturated fats, and I could tweak my training routine a million ways to be more optimal. But being aware that exercise, diet, and sleep should be prioritized at all leads you to a good outcome.
-
The universe is not your friend.
It’s also not your enemy.
Many systems in the world (political, financial, cultural, etc) are ambivalent to your needs and desires. I’ve gotten much more mileage treating institutions (and the people operating on the behalf of these institutions) closer to background objects to navigate around and NPCs to interact with than as anything else.
-
Most institutions are what Samo Burja calls “dead players.” They are miming agency and trying their best to imitate “live players.” Live players are actually affecting the world.
-
Live players come in two flavors: obviously live players, and discrete live players.
”Obviously” live players are institutions and people who are so clearly Doing Things — think Elon Musk, professional athletes, famous actors/actresses, top musicians, etc. Follow the money, and you’ll find them.
”Discreet” live players are institutions who, for a variety of incentives, want to keep it hidden that they’re live players. Typically, they’ve found some resource to extract and want to prevent competition with other live players (and probably place a high value on privacy). Everyone is obsessed with the top 10 high net worth individuals (Bezos, Zuckerberg, Musk, Gates, etc) — but who’s in place 11-500? Who has all of those billions of dollars, and how did they get it?
-
I think of the few thousand engineers dispersed among Nvidia, Anthropic, OpenAI, Facebook, Apple, Google DeepMind, and so on — these are top ICs and managers who are shaping technologies that will impact billions of people. But they have no incentive to be public. Their family and a handful of friends are all they want and need.
-
-
Dead players are trying to maintain the narrative that they’re live players. I think of institutions that ostensibly sell products or provide some service but largely function as places to give people jobs and as places of ritual that enforce sociability and harmony.
These are good outcomes! But not the marketed purpose of the institution (e.g. why do public schools not have better education outcomes? why do low performing employees not get fired? why do governments fail to collect taxes or process documents? why do bad restaurants and cafes stay in business? etc etc etc)
-
The most important qualities you need to get things done are, in order: agency, energy, competence. You must want to do it, and then have the resources and ability to actually execute.
- Many things don’t get done because of lack of will, not lack of energy or skill.
-
AI systems are going to get better. Better than me, and probably better than most people around the world. And it’ll happen in the first half of my lifetime.
-
In the same way that 2024 looks nothing like 1924, I think 2054 will look very different than 2034. The next 5-10 years are probably the last “normal” years we have left, in the sense that after the 2030s, the world will have so many new dynamics that traditional, pragmatic advice will not hold for most people.
- For example, norms around meeting people, community, and dating have changed dramatically since the 1950s. The social worlds before and after the Internet are alien to each other. You can try to cultivate a pre-Internet world in a local sense (e.g., Amish people), but there will always be points of interface between the two, so it will never be quite the same (e.g., I saw Amish people fly from Chicago → Philly with smartphones).
-
People chasing outlier outcomes (98.5%+ results) will be successful in a relative sense (probably land at a 90%+ outcome), and still feel internally devastated that they didn’t achieve their goal.
-
Memory is fallible. People will retroactively construct narratives to make sense of their life, not to be faithful to how events actually played out or the reality of what informed their decision making. Usually, this is the best way to maintain the status quo and sanity.
- I think about how the divorce rate is probably something like 30-40%, and that even amongst marriages that aren’t divorced, there’s a fair share of them that are unhappy. Something like 2/3 relationships will probably fail, but most people paper over reality to keep the relationship going for as long as they can.
-
Pareto distributions are pretty much everywhere.
-
A minority of the population own a majority of the capital and income
-
Only the top (athletes | actors | content creators | musicians) are notable and profitable
-
A handful of employees at firms produce a majority of the value (e.g. 100x engineers, 100x consultants, top lawyers, …)
-
Most of manufacturing is concentrated in a few countries. The USA designs most of the world’s tech product (Apple, Google, Microsoft). Most memory chips are produced in Korea and Japan. Most GPUs are produced in Taiwan. Most clothes are manufactured in China, Vietnam, Pakistan, and India.
-
The top students of the world go to about 20 universities (Ivy Leagues, CMU/berkeley/MIT, Tsinghua University, Seoul National University, Tokyo University, etc)
-
-
Things that are highly networked tend towards some kind of Pareto distribution. If they don’t have a Pareto distribution, there are external counter balances that prevent it from happening. Typically these counter balances are physical/material, but sometimes they will be cultural/legal.
-
The most skilled people in the world only ever concentrate on 1-2 things because they don’t have enough time or energy. Why aren’t all genetically gifted athletes Decathlon athletes? Why is there only one Elon Musk?
-
How does China have equivalents of American social media and tech companies? Google, Facebook, etc could have easily supported the Chinese market (engineering wise) — but they were never adopted. Think about why.
-
-
There are exceptions to everything — study them and understand why they are exceptions to what you think is the general trend. Update what you think the trend is accordingly.