The Peek behind the Curtain interview series includes interviews with eleven people I thought were particularly successful, relatable, or productive. We cover topics ranging from productivity to career exploration to self-care.
This seventh post covers “What skills have you put deliberate practice into trying to develop?” plus thoughts on good judgment and public facing work.
You can view bios of my guests and other posts in the series here.
What skills have you put deliberate practice into trying to develop?
Really understanding what people mean
I feel like a lot of my work involves trying to understand people who have big ideas and are kind of confusing and have left a lot of holes in their argument. Trying to figure out or reconstruct a full argument and then critique it. There's a lot of extraction from conversations that I do and I think I've learned to be good at. I like digging into people and I think I'm good at trying to restate what people said and really doggedly trying to make sure that I get it and can pass their intellectual Turing test.
I think I'm naturally inclined to really make sure that I have understood something in an intellectual conversation. I think it's helped over time to become a little bit less self-conscious about not knowing things. A lot of the game for me is to stop people and be like, "It sounds like you're saying this, is that right?" and really paying attention and not letting things go until I'm satisfied that I've got it.
Ajeya Cotra
Forecasting
I think there’s a variety of skills involved in forecasting. I do a fair amount of it on the Good Judgment Open and things. That's one skill which I think one can drill by practicing, which is what I did.
But the metacognitive skills, like knowing to ask yourself to make a forecast, are not trained by having questions you can answer. There, I've begun to try and make some steps on this. I had a fairly large Google doc of notes I make to myself to try and remind myself. I sometimes have a question like that, or prompts I try and make and then try and learn so that, “Oh, if I'm trying to do X, maybe I should think about Y and Z,” and then Y and Z might well become like fairly explicit forecast which I may not write down but have in my head and then use to inform action.
Like, “The timeline to completion here is more than six months, so you have to start now.” Or “My best guess to time of completion is six months, but the tail risk of it taking more than N months is X% and that seems like a bit too high for a specific good idea to start right now,” and various things of that nature.
Greg Lewis
Writing clearly
Really clear and good writing skill. I naturally tend to write fairly concisely, which is good, but I can work on having that be really clear, having it be pleasant to read, all those kinds of things.
Some of it is just taking advantage of opportunities that come up. CSET runs some trainings and things like that. I try to go to those and take notes, and really pay attention. There are books on how to write well. Try to read these books and notice the things that they point out in my own writing.
Being in a master's program is great for this because I get feedback from professors. I think if you're in a job that involves writing, then probably someone is going to edit you at some point and you can relate to that editing process as like, "Okay. I have to get through this hurdle. Let's just do it as trying to just move things forward.”
Then there's a different version which is trying to look at the edits and really take in like, "Why did this person make this edit? Are there patterns in the edits that they're making? Are they noticing things that could overall be better?" Whether that's just as you review someone's edits or, if you have the opportunity, talking to them and being like, "Hey, did you have any general advice for me or suggestions, and things like that?"
This is all pretty opportunistic, but definitely trying to make use of those things as they arise.
Helen Toner
Deliberately develop skills to patch weaknesses
Things that I am naturally good at, I just try to lean into and make the most out of and I use systems more for things that I am naturally not going to be amazing at and try to get myself up to an adequate standard. I don't know if that's really true but that feels true-ish to me.
Helen Toner
Listening
I spent a bunch of time and I'm still spending a bunch of time working on listening as a skill.
Generally being able to listen to a room or a person's perspective across time or understanding an organization's perspective across things, I think is in some ways about listening closely and actually just trying to be able to view and objectively parse things in a way that is coming from a place of wanting to understand. I've really enjoyed developing this as a skill, so I think that's helped me invest a bunch of time in it.
I think it's paid off in a bunch of different ways. I think the most obvious ways I can think of is that it's really helped me with this cluster around people and understanding organizations. I think it's also helped me a bunch in terms of leadership and management too. I think also it's helped me just become more attuned to communication and sentiment, which I think is also in turn helped me build up public speaking skills and communication skills more generally.
Definitely, I think it's harder than people tend to think it is. I think I'm still trying to make progress on it. Having spent a couple of years on it, I think there's a fair amount that I can still get better at. It's been fun realizing how much deliberate effort I can put into it and how much difference that makes in just being able to be more attuned to how to get better at it.
Jade Leung
Writing clearly and pleasantly
Being able to write well. That's a skill that I think is under-emphasized in research and ends up being just super, super important.
Feedback was super helpful. One of the best writers I know is Allan Dafoe, totally privileged to be able to work with him at GovAI. I think both being able to work with him on pieces of research, and also seeing how he would write and how he would revise, were super helpful. Literally to the level of: he'll do suggested edits and I would read them over and try to understand what kind of thing he was going for, instead of what was there. That, I think, did a lot of the heavy lifting in terms of making me better at it.
Then I think also just paying it more attention/spending more time on it sounds like a simple/obvious thing, but one thing that I learned was if I just read my work closely and actually looked at the sentence level edits that I could make to make it better, I just spotted a bunch more things that I usually wouldn't even have invested the time trying to look for, if I wasn't trying to focus on writing.
I think it's just wearing this extra lens of trying to improve one's writing. Then spending more time than I expected I would need to trying to make it better was actually just really worth it.
Jade Leung
Management
I've spent a lot of time reading different books to do with leadership or coaching or feedback and thinking through how to apply them in different situations. Talking through with people who have more management experience than me, how to navigate some particularly difficult situations. I think that has paid off pretty well.
Michelle Hutchinson
How important is public facing work, like EAG talks or publishing writing?
Quite valuable for research
In a research context, you need to see someone's written work to tell how worth listening to they are, or how much they have to say. I think outside the EA community it's really important, typically. If you want to be invited to different events, or invited to speak to people, that sort of thing.
If you have no public persona, whatsoever, I think this is going to be really hard for you. If there's not any public evidence that you are someone who works in this area who's a legit person. Yes, basically, I'd say overall quite important, with some minor caveats: if it's EA stuff, I think it is sometimes possible to get people's attention without public stuff, but almost always it's important.
Ben Garfinkel
Probably depends on the field
Policy stuff just tends to be very reputation-based, very prestige-based, very relationship-based. For me, in my space, it's been very helpful to have things like that. Not so much EAG talks, but more things like being a co-author on this paper, The Malicious Use of AI that Miles Brundage and Shahar Avin did. At the time, I felt it wouldn't really matter that much if I was a co-author on it or not, but just having published something in this space is good.
So I do think it's pretty important to me, but I think it's quite field-specific.
For me, having written things that are solid and that I feel pretty good about even if they're not groundbreaking. It's just really helpful as a way of showing what kinds of things I have thought about, what kinds of things people can expect me to know things about, versus just doing some ethereal tech and security something something, which is super broad in terms of the space.
Also it’s sort of a portfolio. Like, “Here's what it makes sense to talk to me about. Here is something that I definitely have thoughts on.”
Helen Toner
References can work instead
I haven't really optimized for this that much in my life so far and I don't really regret it. At least my experience so far has been, a lot of what I ended up doing is working on hard projects with people, and either getting things done via those projects and/or having it such that people that I work with are able to see what I'm capable of. Then references have done basically a bunch of the work for me in terms of replacing what public-facing work would have done.
Jade Leung
It’s good for your resume
It's good for your CV. I think one example is Matthew Rahtz, who was a CHAI intern. I think when I was evaluating CHAI intern applicants, I saw his name and was like, "Ah, I remember reading a replication of deep RL from human preferences that I think was by this guy." I checked and it was, or possibly he mentioned this on his application, and like, "Man, I feel like I'm willing to admit him just on that basis." This was not just the fact that he had replicated it, the blog post was also quite good. We still did do the rest of the application process as a sanity check, making sure that nothing else came up, but I felt more confident about Matthew than about basically any other candidate during that application process.
Rohin Shah
I hear the term “good judgment” or even just “thinking well” thrown around a lot to explain what allows people to be successful, prioritize useful things, that sort of thing. What skills do you think contribute to good judgment?
Calibration and logical precision
It still feels a little bit underdefined to me.
Part of it's just logical precision and reliably being able to tell when an argument is logically sound. It feels like it's useful to be able to write mathematical proofs. At least if you can write rigorous proofs that actually hold up or have a sense for when you've proved something, then that's a good sign.
Part of it's being pretty well calibrated. I think the mental motions that you go through to evaluate ideas or plans in a non-biased way feel somewhat similar to trying to be calibrated about a prediction.
Daniel Ziegler
Epistemic rationality, like learning what it felt like to change my mind
I think the version of it that I've thought most about is what the rationality community would call epistemic rationality or truth-seeking or holding true beliefs, probably, which I think is related but not exactly the same thing.
I think a really important phase in my intellectual development was basically reading the Sequences on LessWrong. For me, especially, one that felt especially important that changed how I think about things was the one called How to Actually Change Your Mind which was about first establishing the importance of being able to change your mind which is basically “If you think that you're already right about everything, then you're set, you don't need to change your mind.” Presumably, if you take a second to think about it, you don't think you're right about everything, and, therefore, you're going to need to change your mind about something.
Then I think really importantly, something I got from that series of blog post was just the way it feels from the inside, basically, when you're using confirmation bias or when you're wanting to believe a certain thing or stuff like that, and trying to notice the ways that that was going on in my own head. I think even just getting to the point of being able to notice them was a really big shift.
Helen Toner
Being grounded in how the world actually works
I think one bit of good judgment is about something like knowing how the world works and having a groundedness in reality which enables one to exercise good judgment.
I think the kinds of things that I mean here is just having an understanding of how stuff in the world actually works, how institutions are built and how they function, how people work, how they think, how things will be perceived by folks outside of your main group of people that you spend a bunch of time with and understand well. Having some way to assess how realistic it is that X things happen or Y person can do this thing or that Y person will do this thing.
Having more of this I think is one pretty core pillar of being able to have some starting point to exercise good judgment. I think having a very theoretical, non-grounded way of viewing the world ends up leading to poor judgment or at least not calibrated judgment because it's just not particularly realistic in terms of what assumptions are being made.
Actually trying to make stuff happen in the world tells you quite a lot. Trying to build stuff and trying to sell things and trying to pitch things to different people and trying to actually raise money and trying to do all those things. I think that tells you quite a bit quite quickly about a bunch of that stuff.
Jade Leung
Balancing inside and outside views
One thing would be balancing well between the inside view and the outside view, or gearsy models versus established practice.
One example that will be familiar to a bunch of people is the unilateralist's curse. It's usually phrased as an outside view thing but I think it's really a combination of the two. The unilateralist’s curse is just, to briefly explain it in case one of the listeners doesn't know what it is: if there are n people, all of whom can take some irreversible action. Let's say, publishing information on how to make a nuke, which is irreversible in the sense that once the information is out there you can't take it back in and hide it away again. You assume that all n of these people are perfectly altruistic, or they are perfectly valued-aligned. They're only going to take the action if they think it's net positive for the world.
You could imagine a naive algorithm where each person thinks about their inside view and says if I think this is net positive then I will do it. The problem is even if people are making mistakes at random, even if n-1 of the people think it's net negative and just 1 thinks it's net positive, the action still happens and then can't be reversed by the other n-1 people. Which seems bad if n-1 people think it's bad and 1 think it's good, then probably it's bad and shouldn't be done. That's the unilateralist’s curse.
The way I explained it, it sounds like an outside view argument but in practice when you're actually applying this to make decisions, you also have to decide who the set of n people are. That is a very inside view style judgment. Should I make my research public, my AI research public? If you include everyone in the AI research community and the AI safety community, it's just guarantee going to be: Yes, you should make-- Well, maybe not anymore but two or three years ago that's basically guarantee, you should just make your research public. Maybe we should be excluding the AI research community because they're not that value-aligned, they're not thinking about extinction risks very much. Maybe they're value-aligned but they're not thinking about extinction risk very much, so they're not well placed to evaluate this. We should defer to the people who have actually thought about this a bunch. Maybe that argues more for deferring to MIRI than maybe ML researchers who haven't thought about this as much. Then maybe you, as I do, that MIRI puts too much weight on some considerations and not on others. Then maybe that argues for down weighting MIRI and including other people.
In practice, there's quite a lot of inside view that goes into this and you need to decide at what point you stop applying your inside view and start applying the outside view. I don't know how to do this or better, I don't know principled algorithms of how you can do this and so I can't explain how to do this to anyone but it's something I often find myself doing. I don't know if I'm doing it well, but it's a thing I am doing and I think that it is important.
Rohin Shah