[Epistemic status: A proposed causal mechanism for something I’ve been using for a while which seems to work empirically (n=1!)]

So, the tails come apart. I really like the elegance of this theory and how it seems to touch every part of life. It seems highly applicable.

For example, say you wanted to become the world’s most successful author. There is no single talent for you to learn which makes you the most successful. You must be a good writer but also a good publicist and business manager. Your writing skill can be broken down into various skills such as humour, prose, research ability etc. You need to get good at hiring the right people.

You don’t necessarily need to be the best at any individual skill. The tails come apart so the person who writes the world’s most beautiful prose might not be able to find themselves an audience if they’re lacking in other skills.

This is very similar to Scott Adams’ concept of a talent stack. He claims that having many good but unspectacular skills, which support each other, is a more efficient strategy than working on a single amazing talent.

The tails coming apart is a good causal mechanism to explain how the talent stack might work and significantly increases my estimate of how useful the concept is. I’ve been thinking in talent stack terms for a while and it seems to work for me.


I work in mechanical engineering. The talent which sets me apart the most is maths (I studied maths and physics, not engineering). Relatively few people I work with can follow in detail the maths I do, and probably none could generate the work themselves. Within the less wrong community I’m sure I would stand out a lot less but, in my day-to-day work, I could improve my maths overnight +1σ and no-one would notice the difference.

So in order to improve my overall engineering value it is a better use of my time to focus on some adjacent skills which either:

1. Fill in gaps in my knowledge from not studying engineering (low time commitment for big relative gains)

2. Move me into the top 10% within the company in a skill which is useful alongside maths

Sometimes these are completely new skills, sometimes they may be things I’m already good at which can be modified to be more useful at work.


Part of me thinks this should be obvious. I suspect some people have a vague model for something similar in their head.

What surprises me is how few people act on this.

Most people seem to just improve the skills related to the tasks that they’re carrying out. It’s important to maintain skills in this area but the fact that you’re being paid to work on them suggests that you’re already towards the tail of that particular skill. Getting +1σ better in a skill which you’re already highly competent at is difficult and improves your relative skill over a smaller proportion of the population.

It is rare to see people work on skills that they’re not familiar with and that don’t come up in their normal work, in anticipation of it being useful later on. With proper planning it seems that this is a better use of time.


Toy model time.

Consider two uncorrelated skills which you might learn. You are at the mean skill level (µ) for skill 1 and at µ+2σ for skill 2. Assume a normal distribution for population skill level for both. You have the option of increasing either by 1σ. You only care about getting ahead of as many people as possible with as little effort as possible.

Going from µ to µ+σ increases your position by 34% of the population, going from µ+2σ to µ+3σ only overtakes 2.1% of the population. Even if the skill 2 is more important, it needs to be 16x more important to give the same impact.

Of all the people who are above µ, 32% are able to get above µ+σ. Of all the people above µ+2σ, only 8.7% are able to get above µ+3σ. Training skill 2 by σ is 3.7x harder (rarer) than training skill 1.

Taking into account both impact and difficulty, training skill 1 is 60x more efficient than training skill 2.

Let's move on before I get arrested for crimes against maths.


By way of example, here are some of the things I’ve deliberately learnt which aren’t directly related to my work but are adjacent enough to be helpful.

Electronics. Most mechanical engineers have little knowledge of this so getting into the top 10% is relatively easy. I bought a raspberry pi and did a few mini projects.

Programming. I was already a fair programmer (by mechanical engineer standards). The most important additional thing I did was teach myself visual basic so that I can create macros for MS office programs. I also did a bit of Python and SQL as they're used at my work, even though it isn't for something I work on.

Customer understanding. I noticed in my company that a lot of people didn’t really know what factors cause our potential customers to choose a supplier and how they weight each factor. Hanging around with sales-y people and asking questions gave huge relative gains very quickly

Data science/machine learning. This one was mostly for fun and my knowledge is very basic but, because no one else on site has much knowledge in the subjects, I can do things that no one else can.

Mental maths approximations. I was never amazing at mental arithmetic. Getting better at this sometimes helps with work but it is most useful for impressing to people. Other people can’t relate to “real” maths as they don’t really understand what’s going on but they can relate to mental arithmetic and anyone doing it fast looks amazing.

Psychology. My current topic, I’m hoping it will be useful for career building. First book is Influence by Robert Cialdini as everyone seems to recommend it!

I’d like to claim that I have a mathematical model to determine which skill to work on next but in truth the decision is probably made mostly based on which seems like most fun. This may actually be a valid method of choosing as the more fun something is, the closer it is to a neutral hour. In truth I didn’t pick the method for any such principled reason!


What adjacent skills could you learn to maximise your impact?

New Comment
7 comments, sorted by Click to highlight new comments since:

Joel Spolsky's post on barriers to entry makes a similar point and goes one step further: instead of starting with a list of skills, start with a list of reasons why people aren't buying your stuff.

That’s a great post,thanks for linking. I especially like the counterintuitive removing barriers to exit. I’m not sure that what I’m talking about works if applied to businesses - generally in business it pays to be tightly focussed. Being 10% better at what you do than your nearest competitor can mean that you get a 99% market share. The same may apply to employees in some very focussed professions but probably not many.

You might look into the Lotka curve, the 'o-ring model', or Shockley's analysis of productivity as a lognormal phenomenon: "On the Statistics of the Individual Variations of Productivity in a Research Laboratory". The 'adjacency' of skills clearly must break down at some point, otherwise specialization & division-of-labor wouldn't be so useful, and there are probably a lot of other things like hard thresholds for skills to be worthwhile, but probably many people aren't at the optimal balances of skills/traits - I think yak shaving may be a particularly strong symptom of when certain skills or tools have been underinvested in.

I've had similar success with this. Being the programmer with system admin and org dev skills has served me well! Similarly I think I do well in my research being the AI safety phenomenologist.

To say a bit more on this, I think there's also an adaptive behavior some people pick up in this direction that leads them to seek local uniqueness. Sometimes this backfires and the person becomes ostracized as a weirdo, but when it's successful they have something that makes them unique, special, and, importantly for their chances at success in survival and reproduction, indispensable.

Of course I don't think most people think of doing this deliberately in this way. I myself feel it as a motivation to be special, so sometimes pursuing a skill feels worth the effort because I expect it will make me special. But that's a difference of reasons why this behavior might be selected for and what it feels like to perform the behavior.


Over at MR Cowan was just talking about how the great Venture Capitalists were all generalist, not specialist. This post seems to be in the same vein. I tend to be very sympathetic to the idea that humans will do best if they are not overly specialized. Might be something from my dad -- he always told me I should get a trade and a profession and that way I will always be able for find a good paying job. That is a type of generalist.

I also agree that it's not merely being something of a jack of many trades but to have those skills that are complementary with the others, not substitutes ( the anti-correlated relationship romeostevensit mentions).

Clearly it's an example of the whole being greater than the sum of the part idea.

The most valuable stacks are going to be ones that are in demand while having anti-correlated component skills.