Are you working on something important?
Why not?
[Hello everyone! Neil here. This email is the 7th in a feed where I update my close friends ∼ weekly. You may read it, skip it, delete it, or tape it to the wall and pray to it every night. — — If you want to support me: reply with a one-sentence email, even if you didn’t read the entire update. You can literally just answer with "whee” and I’ll be happy. React in any way shape or form. Thanks! To unsubscribe, either mark my emails as spam or defeat me at ping-pong-chess. Good luck.]
Si vous ne parlez que français, vous pouvez utiliser Google Translate si ça vous intéresse!
I.
Calm week.
(That’s not necessarily a good thing btw. “Calm” sounds nice, but isn’t. I should be doing more.)
No LessWrong.com posts written. Why?
Friction: it takes work to follow an idea until it becomes interesting enough to be worth publishing
And so I fiddle with the idea for a time and then get bored and get on to the next without publishing
And so my cadence is unbearably slow, and that makes me less likely to publish, which slows me down even more
Read this blog post if you encounter similar challenges
IEP test/concours/thing completed yesterday. A room, 6-hours worth of essays, and the only criterion for admitting you until any of their schools. There are worse methods of filtering students!
Blue Dot AI course is still good. If anyone wants to peak at any of their resources, their curriculum is public. All we do in our two-hour weekly zoom calls is “discuss”, so you’re not missing anything.
I got accepted into the Sciences Po / UCL dual degree, and that’s likely my best choice
Sciences Po is a good target for influence given it’s at the center of French governance
Don’t worry about UCL, that’s only relevant in 2 years (far future)
This isn’t my work at all, but I’ll sneak it in anyway. There’s been some big news in machine learning interpretability! That’s the discipline that tries figuring out what the heck is going on in the models we build.
This post is probably more accessible than the original post, but both are necessary to get the full picture
It’s technical-ish and took me several re-readings to understand. I don’t do much math.
The takeaway is: ChatGPT doesn’t just predict the next word in a sentence, it builds a model of the world and simulates itself running through that model. Only then does it predict the next word. It’s a lot smarter than you think.
Nick Bostrom still hasn’t put my translation of his paper up, but he has a good excuse: his institute just shut down!
II.
Do you actually show up?
And remember: To be a Plot Character, you’ve got to involve yourself in the Plot of the Story. Which from the standpoint of a hundred million years from now, is much more likely to involve the creation of Artificial Intelligence or the next great advance in human rationality (e.g. Science) than… than all that other stuff. Sometimes I don’t really understand why so few people try to get involved in the Plot. But if there’s one thing I’ve learned in life, it’s that the important things are accomplished not by those best suited to do them, or by those who ought to be responsible for doing them, but by whoever actually shows up.
- Eliezer Yudkowsky, 2014
You’re far from being the smartest or most conscientious person to work in whatever field you’re planning on integrating. Like, far far. In almost any category you compete in, you’ll be left in the dust. The others are just too good.
Yesterday at the IEP test, I was reminded that a lot of the kids in the room have been planning for this test for years. They paid hundreds of euros to get into those fancy online thingies that prepare you. They’ve been studying at least 30 cumulative hours for this. Meanwhile, my full plan amounted to the chunk of the history textbook I crammed at 11PM the night before.
Tests of this kind are linear. Parcoursup will tell you how you ranked. There’s only one available 1rst place; so too for 2nd place and 3rd place. If there’s anyone, out there, at all, that’s both smarter and more conscientious than you, you’re doomed. It’s zero sum.
This is why I assert that you’d be mad to compete in tests of this kind any longer. You can generalize from the IEP test: don’t compete when you’re sure to lose. Before entering a field: check whether it’s just an IEP test with extra steps under the hood.
(Why would you want to be the best, you ask? Because most fields work by cumulative effect, where the already-successful find it easier than others to become more successful. Success curves tend to be Pareto distributions. So whatever you care about, you get 10 times more of it if you’re the best. A worthy goal.)
III.
The cool thing with life is that there are two ways of cheating. The first is combining two different fields and being the best at both at once. There’s a term in mathematics/economics called the “Pareto frontier” which is fancy for “best possible combination of two metrics”. You might spend 75% of your skill on studying statistics and 25% of your skill on studying oviparous eggs; or 50/50 or 40/60. Whatever the arrangement, you’re on the Pareto frontier.
If you’re a little smart/conscientious, you can be the best in the world at statistics/oviparous eggs. Take a look at this graph:
Here, both Alice and Bob and Carol are at the Pareto frontier between the skills of gerontology (old people) and Bayesian statistics (subcategory of statistics).
So you can escape IEP-test-like competition by being both good at history essays and statistics, machine learning, psychology, evolution, etc. You might be the only expert on Earth in both history and evolution. If there’s any value in that intersection—that is, if you get paid for solving the problems in your Pareto box (see graph)—then you’ve found a good career.
Read this short post for a good explanation.
IV.
Richard Hamming’s favorite activity was sneaking into labs. Imagine you’re a biologist giving shrooms to rats, a mathematician solving Famous Impossible Problem, or an economist studying 14th century globalization. Suddenly, Richard Hamming bursts out from under your desk, grabs you, and shakes you violently by the shoulders screaming “ARE YOU WORKING ON THE MOST IMPORTANT PROBLEM IN YOUR FIELD?!?!?”
Richard Hamming was not particularly liked, because most scientists would answer “No I’m not working on the most important problem I know about—hey how’d you get that security badge?”.
In his words: “the average scientist spends almost all his time on problems which he believes will not be important and he also doesn’t believe that they will lead to important problems.” In other words, most smart, conscientious people have failed step one, which is to actually show up.
V.
You see this reflected in the real world. Most low-hanging fruit problems, which if solved would let you save thousands of lives, are so low-hanging that there’s a good chance you’ll be the most competent person working on it. All you have to do is seek a promising-looking avenue which lets you combine two or more fields—and then pivot that toward the most important problem you know about—and you’ll succeed more than you can imagine. Keep me updated.
Are there two fields you enjoy that might make a good combination?
What’s the most important problem you know about within those fields?
Do you actually show up?
[Email me!]
Appendix: here’s Richard Hamming’s speech:
