People watching, according to me, refers to the art/science/act of observing people, judging them, classifying them, analyzing them, finding them, etc. etc.
TODO: say something about Dunbar's number. there's a limit to how many people you can track closely, but most people are so bad that they're not worth tracking closely, so it sort of works out in the end.. you still want to get to the point where you're at least somewhat familiar with a ton of people though. a big problem that i think other people have (i.e. people who are not gung ho about people watching) is that they don't store enough people's names and what they're like, so they don't even catch connections that come up. like let's say person X mentions person Y as a cool person to check out. and then like five months later, person Z also says person Y is cool. but by that point you've forgotten that person X said this, so your "internal karma" for person Y has reset, instead of accumulating. you want to avoid this sort of thing happening. human memory sucks, etc.
Identifying good people
generic, high-level strategy:
- learning everything / having the right opinions about everything by just sticking to the object level is impossible -- there's just not enough time. At some point, you will need to rely on other people to identify interesting topics, or promote to your attention the correct views, etc. In a post-singularity utopia you will have enough time to derive everything from first principles, but we are not there yet.
- as with ideas, it's sometimes useful to become super excited about someone, and obsessively read everything they've written. It's ok to let yourself get super excited about someone to the point of obsession. Just realize that you will probably change your opinions later.
- when you see that an interesting person says something you disagree with, this is an opportunity for you to decide if you know better than they do, or if they know better (you can turn the latter situation into the former by doing additional research on the topic). see also Eliezer's rhythm of disagreement.
- randomly fact-check things using google/wikipedia/generic sources to score someone.
- as you become more knowledgeable about the world, things that used to seem to you like novel ideas/facts will eventually become "things everyone knows"/cliches/uninsightful. This is an opportunity for you to go back to people whom you've admired, to see if you need to tone down your admiration for them. if you find that they actually just repeat cliches that to your past self seemed insightful, you can downgrade them (but remember that insights become distilled and enter the "water supply" over time, so if they were one of the first people to start saying the now-cliche, you might actually want to upgrade your opinion of them. Someone had to invent an idea for the first time, and that person is probably quite a bit smarter than the people who are parroting that idea).
- in general, you won't be able to rank people who are significantly above you -- they will all just seem super smart. when i was a clueless teenager who stumbled onto LessWrong, everyone seemed super smart! i had to do a lot of learning about the world to get to the point where i could pick out people like eliezer, wei dai, carl shulman, and say "these people are the smartest on the site". see hiring and judging character for more on this. but as you get smarter yourself, you can come back to people who seemed super smart and see if now you can tell which ones are really smart vs merely seemed-smart-to-my-younger-self. spaced repetition is a thing, you don't need to judge someone all at once.
- read about gell-mann amnesia
- you can use recommendations to generate ideas for smart people, but you will generally need to put in the work to verify that the recommended people are good. e.g. if smart person A says that B and C are smart, you usually can't trust A to be correct about this. (it's really weird! i know a bunch of smart people who give lots of false positives about who else is smart. smartness is sadly not transitive.) you have to go through B and C's stuff yourself to judge for yourself.
Big picture vision
what is the point of all of this?
- there aren't many good people
- big insight of people watching: there aren't _that_ many people. if you really wanted, you could go through every single user on LW and check if they are smart or not. that will let you find all the smart people on LW. PW looks harder than it is because people aren't systematic about it.
- scope insensitivity in being able to tell who's good: if someone has produced 100x more good stuff than another person who has already produced a ton of work, it's really hard for most people to tell the difference.
- there aren't many potential candidates of good people either -- so it's actually possible to map out the whole space given ~1,000 hours of work.
- you will sometimes find that the people you eventually discover to be good are people you already encountered once or twice before and skipped over. this has happened for me with eliezer/LW in general, wei (2015 vs 2017), carl, jonathan blow (2016 vs 2020). you could maybe even count vipul too (if you count me checking groupprops when i was young and before i knew him). what does this mean? one thought is that there might be people you're missing out on that you skipped over. another is that there's actually this small core of good people, and because they're good you keep running into them (and eventually realizing how good they are). another is that there's a lot of redundancy when it comes to people watching; you don't need to get everything 100% correct on the first try -- instead, it's a long, multi-year process and you're allowed to make mistakes, you're allowed to change your mind about someone, etc.
- you can't convince people of the value of your people watching project or skills; they either have to already agree with your or they will never agree. The sad truth is that most people are really dumb, and they are also too dumb to figure out who is smart.
- It seems like there are fewer Eliezer/Wei/Carl caliber people in their 20s, compared to 15 years ago. What's going on?
- It kinda feels like the young people now in EA aren't as good as the young people in EA 5 years ago. What's going on?
- Most people don't even conceptualize people watching as a skill/discipline. They don't think of it as something you can improve on, where there is even an absolute scale that measures how good someone is or how good you are at figuring out how good someone is. In a specific context like software interviews, they might agree that you can interview people for a job in better and worse ways. They might even think you can improve your skills as a interviewer, and that there is a discipline of interviewing, of figuring out what works, what is a good predictor of job performance, and so on. But I think people just stop there, and don't generalize to the broader context, and ask "how can I find the smartest people, in a holistic/overall worldviews/overall ability to figure things out sense?"
- this people watching stuff is probably impossible to explain, and it's not clear that it's even worth my time to attempt this. when i say someone is "good" or "bad", you sort of already need to have been in conversation with me for like one year to even understand what i mean by that. what does it mean to say that "grognor was right about everything"? etc. There is just too much to explain, and someone who reads this won't be willing to put in the effort to understand all that. i think i'm partly just writing this for myself, to organize my thoughts after all these years of PWing.
- Output curve for good people
- brand association -- something I do much more than others (e.g. other people visit a website or look at someone's comment on Facebook and probably forget about it and don't look around). This is basically one of my secrets of being good at people watching. You need to make each person into a brand and keep them in your cache basically, so that you can develop a model of them over time. If you don't remember them, then each time you're seeing a new person (and you think of them as "randomly sampled LW user" or "random EA on the internet" or whatever), and you never get deeper into what they are about.
- Parasitizing on popularity
- Using "site:*.com/about" with identity labels  -- i actually think i might have found anatoly karlin this way. i don't remember finding anyone else that's interesting this way.
- there's sometimes lists like  but idk I've never found these super good for finding people?
- use spaced repetition or just a plain reminder system to check back on promising people -- e.g. I have to-do items in orgmode that just say "check back on X" where X is the url of someone's site, with a short description of why i found that person promising. young people especially can't be expected to have much online presence, so waiting gives you a way to guarantee that you "discover" them later.
- use tools like urlwatch to track people's websites (especially for those that are just static web pages that are infrequently updated)
- find hubs like rationality community/EA/michael nielsen twitter and use it to find more people.
- pay bounties to get recommendations for people (I am currently trying this out; i'll report back on results)
- IRL networks -- i haven't found this as useful for finding smart people. I think it's fine for finding decent people and for finding friends/people you can trust.
- find recommendations that smart people make (sometimes works, but in general goodness is not transitive)
- start with smart people you already know, and then look for people who recommend/talk about this person. e.g. as of early 2020 i think it's still a reliable indicator of some quality if someone talks about wei dai unprompted. But obviously this doesn't work for more famous people like eliezer. another example is gary drescher -- very few people seem to know him/talk about him, so it is evidence of goodness if people do talk about him.