Writing Style and the Typical Mind Fallacy
For a long time, Eliezer has been telling me I should write more like he does. I’ve mostly resisted, preferring instead to write like this:
Explain the lesson of the post immediately, and outline the ideas clearly with lots of headings, subheadings, lists, etc.
State the abstract points first, then give concrete examples.
Provide lots of links and references to related work so that readers have the opportunity to read more detail about what I’m trying to say (in case it wasn’t clear in a single sentence or paragraph), or read the same thing from a different angle (in case the metaphors and language I used weren’t clear to that reader).
At the recent Effective Altruism Summit I tried to figure out which personal features predicted writing style preference.
One hypothesis was that people who read lots of fiction (like Eliezer) will tend to prefer Eliezer’s story-like style, while those who read almost exclusively non-fiction (like me) will tend to prefer my “just gimme the facts” style. This hypothesis didn’t hold up well on my non-scientific survey of ~10 LW-reading effective altruists.
Another hypothesis was that most people would prefer Eliezer’s more exciting posts, while people trained in the sciences or analytic philosophy (which insist on clear organization, definitions, references to related work, etc.) would prefer my posts. This hypothesis fared a bit better, but not by much.
Another hypothesis was that people who had acquired an epiphany addiction would prefer Eliezer’s style, whereas those who just want to learn everything efficiently would prefer my style. But I didn’t test this.
Another hypothesis that occurs to me is that people with short attention spans could prefer my more skimmable style. But I haven’t tested this.
Perhaps the community would like to propose some hypotheses, and test them with LW polling?
- Engaging First Introductions to AI Risk by (19 Aug 2013 6:26 UTC; 31 points)
- Outline Summary: The Procrastination Equation by (26 Mar 2014 23:55 UTC; 11 points)
- 's comment on [Discussion] Are academic papers a terrible discussion forum for effective altruists? by (EA Forum; 11 Jul 2015 18:48 UTC; 0 points)
There is one person whose writings seem of better quality to me than either yours or Eliezer’s, and that’s Yvain. What do you think his writing style is?
(To be clear, I enjoy what you and EY write, despite the style differences, especially when you are at your best.)
Not sure what I’d call it, but I agree with Michael Vassar that the day Yvain began “Real Work” was “a tragic day for literary history.”
Time will tell—he’s still writing, and I find his dispatches from the front lines of patient care more interesting than some of his posts prior to that.
And it may yet be a good thing. Two quotes come to mind, one by Popper and one by Von Neumann:
I don’t think I agree with the Von Neumann quote (vide “pure mathematics”). One thing that does seem to guard against some of the problems discussed in the first quote is the rigor of proof (i.e. either be empirically-driven or formal (or both)).
Has there been anything since I Aten’t Dead?
Not that I know of, alas.
The writer who I like more than Eliezer, Luke or Yvain is Robin Hanson.
His posts seem supremely succinct and clear. There is no obfuscation of language and there is little attempt to add dressing or flavoring. Examples are extremely relevant. Typically, many possible hypotheses are considered and reasons provided for choosing a subset of these. He is also the best summarizer in the business.
I prefer Yvain’s writing style to Eliezer’s, but Yvain also has the problem of being too verbose for my taste. In terms of information-to-words ratio, my opinion is that lukeprog’s writings outperform Eliezer’s and Yvain’s.
That’s not a good metric to maximize… There is probably some broad optimum preferred by maybe 80% of readers, and too dense or too watery writing is definitely inferior. There ought to be some relevant research out there.
I agree, I think a good writer has a sense when a particular part of his argument is tricky or more difficult to grasp so he may add additional explanations or examples even though he has already made the point.
Speaking as someone with fairly serious attention deficit disorder, I find Eliezer’s style (and Yvain’s as well) is particularly good at holding my attention for long periods of time. Where a more bare-bones article would require me to use willpower to finish reading, I will tend to become hyperfocused (a less commonly recognized aspect of ADD) on one written in a style I find entertaining.
I have ADHD, and also find EY’s writing to be more engaging.
EDIT: Part of the problem with lukeprogs “more skimmable style” is that it is actually possible to skim and sortof get away with it, whereas in EY’s “more narrative style”, you lose the narrative flow if you start skipping, so there is a higher apparent cost to not paying attention. Hence I remain engaged with EY’s writing, and start skipping things in luke’s.
Thirded.
hm, that sounds bad.
hm, that sounds good.
hm, that sounds bad.
To me, Paul Graham is approximately the apex of good writing for the internet. I’d always assumed the sequences had succeeded despite, not because of, EY’s writing style. (I’ve wondered how many people were dissuaded from LW because they pick up things quickly, value their time, and were getting a frustratingly low insight/minute ratio from the Sequences.) It’s interesting to hear that others disagree.
I’ve occasionally had the experience of wanting to convey a concept from the Sequences to somebody who hasn’t read them, but when I try to find a good post to link them to, I realize that the description of the concept is spread out over three or more posts that each individually have a frustratingly low content-to-words ratio. (The LW wiki helps a bit, but there the descriptions are often too concise to be useful.)
I suspect that the popularity of the Sequences is both despite and due to the writing style. This problem with the style didn’t matter so much when the posts were being written and they showed up once a day in my RSS feed—in order to properly learn a concept, you need to encounter it several times with slight variations, and the actual message being spread out over many posts was originally helpful in this respect. It spread out the message over several days of reading and thus helped learn it better than if there had been just one clear, to-the-point post—that you read once and then forgot.
However, now that nobody is reading the posts at a one-per-day rate anymore, the style and format seems harmful. When you’re reading through a (huge) archived sequence of posts, unnecessary fluff will just create a feeling of things having being written in a needlessly wordy way. And it makes it very hard to usefully link to posts about specific concepts.
Are there some specific concepts that come readily to mind? Maybe we could experiment with making much more substantial, useful wiki pages, riddled with quotations / excerpts from LW and elsewhere. The current standard size and tone of the wiki pages could be preserved—as the lead sections for much longer articles.
(Broadly) wikipedian style is also useful for keeping articles organized and concise because of the section structure. Sections make topic divisions obvious, and ‘See more’ links can be included at the beginning of each section for more details.
We needn’t be bashful about copy-pasting large chunks of the Sequences where it’s useful to do so, and at the same time the more transparent structure of the wiki, and our ability to leave anything out that isn’t absolutely essential, would let us piece together more to-the-point presentations. The wiki would then become our go-to resource for quick but thorough introductions to concepts, and would include links to the relevant Sequence posts (and primary literature, etc.) for people who want more examples and poetry. (Both of which can be immensely important.)
Here’s one example.
I notice that comment was written two months before Eliezer published Skill: The Map is Not the Territory. Does this article satisfy you, or does it still deviate from what you’d ideally link to? If so, in what ways does it (or it + the sister articles) deviate?
That article has the problem that I was asking for a good page that would explain and summarize the meaning of the phrase. Skill never actually explains the phrase—rather it starts from the assumption that the reader already knows what that means. You may be able to figure out the intended meaning from the examples, but it would take some effort, particularly if you’re not already familiar with these kinds of ideas.
seconded.
For some people (myself included if I’m honest), a low insight/minute ratio might be a reason for liking the sequences. They enjoy seeing smart folks agree with and reiterate their beliefs. It makes the reader think “ah yes, this person is right, he is smart, just like me”.
I have a friend for whom this is the case. I’ve tried linking her to LW and she finds the sequences (a) move slowly and (b) mostly appear to reiterate stuff that she already knows about biases etc, from reading pop-psych/philosophy/science, books/blogs. I feel like she’d be a valuable member in most intellectual communities, but so far she’s been disinterested in this one.
Granted, another part of this is she appears to be disinclined towards “optimization” as a drive. Or something. I’ve still yet to fully understand this.
No idea whether it relates to your friend or not, but… the fact that the big names on Lesswrong seem to be really happy people appears to me to be anomalous. I seem to recall seeing a study somewhere that maximizing makes people less happy than satisficing. But optimization is just maximizing utility. It’s a puzzle.
Yeah. I think she doesn’t like the ideally of continually trying to improve stuff… so it does appear to be contrary to satisficing, though perhaps on a meta-level. Like, if something’s not working in her life, she’ll go and fix it, but she doesn’t operate from a perspective of continual growth. Ongoing growth, absolutely: Dweck-wise, she definitely has a growth mindset… but there’s no sense of “how can I make today marginally better than yesterday” etc...
Could you elaborate on the difference between continual and ongoing growth? Dweck-style growth mindset seems similar to LW-style life optimization on a practical level to me.
Uhhh, continual is a subset of ongoing. Essentially, my friend fixes things when they’re obviously problematic, and actually does a fair bit of self-modification in the process, in addition to modifying her environment. I think perhaps it’s somewhat like she just tries to find a local optimum, and then goes back to doing stuff with her life.
The mindset test basically asks two questions phrased a bunch of different ways, and those questions amount to “to what extent are your abilities (talent/intelligence) fixed?” My friend certainly believes (and demonstrates) that if necessary she can level up at any given thing, but for the most part her focus is on actually doing stuff, rather than growth.
I’ve tried a few times to sit down and read the sequences, but I’ve never been able to, for the reason I think lukeprog’s style is trying to address. There’s only so much background information I can stand, especially if the writer isn’t clear on where he’s going. And this is coming from someone who sat down and read Mencius Moldbug. And some of Nietzsche. And read HPMOR in three days.
I’m still interested in LW, obviously, but for… other reasons.
(As for my own style: I hate writing anything longer than a comment, so when I do, I have to keep myself amused, which usually means Carlylean bombast.)
I’m somewhat worried that this problem might be even more widespread. For example, “show, don’t tell” is the standard advice in writing guides, so I was intrigued to see my friend write:
I read somewhere that the taste of an experienced film critic is typically very different from the taste of somebody who only watches movies every now and then, and it seems like there should be a similar effect with writing—the kind of people who end up becoming authors or creating writing guides are likely to be atypical, and to have a atypical tastes. Since most writing guides seem to be based on the author’s personal taste as opposed to anything resembling an objective survey, I’m uncertain of how well their advice actually generalizes. (This being Less Wrong, I cannot avoid the obvious hypothesis that the “show, don’t tell” rule is partially an attempt to signal sophistication.)
I just had a conversation about this last night. I was saying that I don’t pay much attention to what movie critics say, and use the ratings of non-professionals because it’s just better correlated with what I like.
So, I’ll say this. Eliezer’s posts put the nail in the coffin of my Theism in the matter of a couple months. My mind had to wait to be told the story of how I could sit, ponder, and wonder for night after night and allow my entire mind to get on board with a major worldview shift. Each individual post assailed the narratives I was telling myself; without the conclusions given to me up front, I was able to travel the journey to the conclusion on my own, which ended up far more compelling than previously read arguments with a more explicit agenda.
That said, now that I find myself much closer to both of your worldviews, I find your posts much more efficient, and I more often recommend yours to others over Eliezer’s. I do this because I expect more impatience on the part of others than I do in myself.
This is a problem I feel with Less Wrong in general, although I know little of what could be done. I often recommend posts to outsiders, only to receive little else than skepticism and comments at the styles or apparent agendas of the authors. “But the logic and evidence awareness make them unassailable!” does little to established narratives. Narratives which suggest that coffee-table books by authors on the New York Times bestsellers list who give TED talks are the quickest route to insight. This was never a narrative I had, and was already looking for insight in new places, and so the initial thrust of taking the time to read the Sequences was all I really needed to be carried along by them more or less to their finish line.
Your point about different audiences is at point here. For most of the world, the Sequences and the scattershot back-referenced posts here are completely irrelevant to established patterns of trawling trusted mediums for dinner-party political wisdom. But giving in to that attitude would heavily decrease the value of this site to most of its current participants. Especially me, so don’t do it, yo.
This is a problem of writing in general: different audiences require different speed of explanation and level of detail. Write for beginners, and the experts will complain that your texts advance very slowly and repeat endlessly. Write for experts, and the beginners will complain they are not able to understand.
(Even worse, write a longer text for experts because the inferential distance is high; the experts will underestimate the inferential distance, pattern-match your text with writing for beginners, complain that there were too many details they had to skip, and then write a comment showing some basic misunderstanding they wouldn’t make if they wouldn’t skip those parts. -- “You wrote a lot of stuff about AI that everyone who studied AI knows, so I skipped it. Why don’t you simply make an AI that does not have goals? Also, the superhuman AI would obviously be smart enough to invent the correct morality. What? No, I am not going to read the Sequences!”)
So when you want to recommend something to other people, you shouldn’t recommend what is best for you now, because now you are a kind of expert (at least in the LW lingo) and they are beginners.
This is analogical to when someone asks me to recommend them some good texts online to learn programming. What are good sources for me is not a good source for the beginner. I prefer encyclopedia to find the missing details; the beginner needs a textbook to explain the simple concepts one at a time.
Fortunately, you don’t need a detailed model of which people prefer which writing style to encourage people to write posts on LW in a variety of styles!
Rather than “abstract stuff first, examples later” or “examples first, abstract stuff later” I prefer a hybrid approach: give as much of the abstract stuff as is necessary to motivate why you’re looking at the examples (which may be no abstract stuff at all), then give the examples, then give the rest of the abstract stuff. The main application I have in mind is mathematical writing, where sometimes a definition is very hard to motivate without examples, but sometimes the examples are very hard to motivate without looking at a previously well-understood abstraction first. The problem I have with always giving examples first is that I often don’t know what to do with the examples if they haven’t been properly introduced: where, in the filesystem of my brain, should I be filing these things?
Paul Halmos!
The difference between the two styles has to do with whether you are trying to get someone to change their beliefs and/or behavior. I suggest that if you run an experiment, you’ll find that even people who prefer to read in a factual style will be more likely to actually change their beliefs or behaviors when information is presented in story format.
(Heck, you might not have to run that experiment; ISTR that there’s already a study showing people shift attitudes as a result of identifying with story characters.)
Eliezer is right: stories are better writing… IF your intention is to influence your reader.
So, if you don’t care what your reader does with the information you give them, and you don’t care if people who aren’t already motivated to obtain that information get bored and tune you out, then feel free to simply provide facts. If you want people to care about the thing you’re writing about, stories are a requirement.
tl;dr: facts are far, stories are near.
(ETA: the book Wired For Story has some background information on the science of brains and stories.)
The stuff written by Eliezer that really stuck with me is the Sequences. The stuff written by you that really stuck with me are some of your instrumental rationality posts, like “How to Be Happy”.
It’s pretty clear to me that Eliezer’s dramatic, storytelling style is better suited for a series of essays designed to radically alter the way the reader thinks about truth, beauty, right and wrong, etc. And it’s pretty clear to me that your straightforward, just-the-facts style is better suited for posts along the lines of “here is what science suggests you can do to become happier/more productive”.
Consider maybe the idea that different styles are good for different things, and perhaps you and Eliezer should choose things to write about based on your respective comparative advantages.
Agreed, and I’d been meaning to ask Eliezer how he thinks about ways of gradually easing people into ideas which are new and possibly shocking for them.
Personally, I prefer Eliezer’s posts because I actually read Eliezer’s posts. (Also: Yvain’s posts.)
Of course the reverse is true as well (read because prefer). But the point is: if I don’t read something (because I don’t enjoy reading it), what good is it to me?
Hmm, I prefer to read Luke’s writing, but I find it best as a way of grasping information, not necessarily as a way of motivating me to action. Case in point: I read all of Luke’s posts on science of winning at life, and thought it interesting. However it wasn’t until reading the motivation hacker that I was motivated to move from knowing it to using it.
Do you prefer...
[pollid:538]
Do you read a lot of fiction?
[pollid:539]
Your background?
[pollid:540]
Do you consider yourself to have an epiphany addiction?
[pollid:541]
Bonus question
[pollid:542]
[X] Yvain’s style
I struggle to get through Yvain’s posts; it’s like trying to sit through a social psychology lecture delivered by Fluttershy. All ability to focus is blotted out by eeee I want to hug the author
I was concerned about this comment right up until the end, at which point I discovered that I had really enjoyed it intensely the whole time.
I am still interested in some good examples of Yvain’s posts which invoked this reaction from you, I have been reading somewhat more of his writing recently and though I often agree with his points, I don’t recall feeling such explicit urges.
PMed you because of reasons
Could you please provide some good examples which trigger this effect? I haven’t read more than ten or so of Yvain’s blog posts so my sample size might explain why I haven’t observed it yet.
I didn’t answer the poll because I prefer Luke’s style for instrumental rationality topics and Eliezer’s for epistemic rationality topics. It was unclear to me how I should have answered the questions.
I propose that another bonus question be added, one about preferred method of eating corn.
This poll does not show how people from one category voted in another, which is precisely the relationship that Lukeprog was looking for.
Unless you can do that with the raw poll data, but that just confused me.
Thankfully, the data is not quite that crippled! The data is reported in a… ‘long’ format, I think the term is, where each row is a single poll item response with a unique ID for the respondent. If you want to look at that sort of question, it’s up to you to aggregate the data correctly (eg with
grep). You can see this by looking at particular unique IDs, say that ofLeonhartand anonymous respondent11:There’s 5 entries for each, since there were 5 poll items, and and each poll item has its own unique ID as well. So if you wanted to know the relationship of an answer on poll item #538 and #541 based on how subjects answered #538, you’d get a list of everyone answered “0” in #538, and pull out their answer for #541 as well. That sort of thing.
(And now that I’m the topic, I wonder where my own writings fall, and how I would even know if I were insufficiently writing like Eliezer/Luke/Yvain.)
I like the your non-fiction style a lot (don’t know your fictional stuff). I often get the impression you’re in total control of the material. Very thorough yet original, witty and humble. The exemplary research paper. Definitely more Luke than Yvain/Eliezer.
I thought somebody else had previously done such analysis with raw poll data? But maybe I’m mistaken.
I voted “both equally”, but really I want Eliezer to continue writing in Eliezer’s style and Luke to keep writing in Luke’s style! Mostly I prefer reading factual things written in Luke’s style, but Eliezer’s style really seems to work for the sequences.
I put myself down for not reading a lot of fiction, but I was torn on that question, because while my book list is almost entirely nonfiction now, a few years ago it would have been the opposite.
Maybe say a little more about the evidence for this? Have you had conversations with people that confirmed this, or is it just coming from a strong prior?
I’ve now spoken to dozens of people who definitely prefer my style, dozens who definitely prefer Eliezer’s style, and dozens who are somewhere in-between.
And the universe of styles isn’t even defined by these two poles.
It occurs to me that any poll we run might be biased due to hypotheses being told to people before they answer the poll. Like, if I’m not sure if I think I have an epiphany addiction, and then I read that such are hypothetically associated with author X, then… well, I’m not sure how it changes my answer, but I know it makes it hard for me to feel like I’m answering fairly.
I prefer your style (rather, I really dislike Eliezer’s style). Possible data points: I read a lot of math: math blogs, math texts, math papers, and I have poor reading comprehension and reading speed. I don’t have a particularly short or long attention span, and I don’t really read much science or philosophy. I didn’t get a whole lot of epiphanies from the sequences, though it did have a strong influence on how I think (ie. my updates weren’t felt as epiphanies).
I like the structure of your writing. I like to build my mental categories from the top down, and structured writing helps me put things in mental buckets. For quite a while after reading the sequences, the whole idea of rationality was a big muddle of concepts and I had a hard time thinking about it as a whole. I had to think it over and do all the categorization by myself, which was a lot of work, and I don’t think I benefited enough from having to do that to justify the exercise.
I haven’t been conscious of your writing style, which probably means that it’s agreeable to me. Unless I were actively interested in it, the only time anyone’s writing style would come to my attention would be when it gets in the way or is somehow out of place.
If anything, your style of writing conforms closely to the SQ3R method of reading comprehension, which is obviously a point in your favor.
I generally prefer the more direct {lesson, evidence}. I have on several occasions thought that Luke has implemented this well.
But—I think we have evidence that EY is a particularly good writer of narrative. While also getting the content across. The epiphany hit is pretty sweet too.
Embedding lessons in stories (like the Blue and Green) makes the mind labile to their content and makes it easier to hang on to the memory and to retell to others. I imagine it comes at the cost to extra thinking and writing time to package lessons so.
Is that cost worth the marginal effort? I’m pretty sure the answer is ‘sometimes’.
And if you have a Luke and a Eliezer both on board, surely not everyone needs to their own lesson building, literature sweeps and narrative weaving (in the situations where those might be particularly useful).
Use comparative advantage?
Unstated in this post or the ‘typical mind fallacy’ link is the hypothesis that objectively not-better writing exists. To test this hypothesis, here is a sample of not-better writing that I have a high confidence will not be persuasive or interesting:
s lfas l sdkf jlds dj;lkg j;lasl dkjg lsakdjg ;lsasldkgj l;askd
If I have established that objectively not-better writing exists, that is a foundation to move away from. It is possible that in moving away from objectively not-better writing, we might move toward better writing. Further, we might move toward better writing that meets more than one preferred style of reading.
Writing is both art and craft. Unless you are a natural at writing, there are standard technical writing techniques you can follow (and must follow if you want to get published). I mentioned some of them a couple of years back. Keeping in mind a simple “state-explain-summarize” approach helps to make technical posts reach a certain minimum readability level.
And the same person might enjoy one style more in some contexts, and the other style more in other contexts.
I prefer an organized, expository style in an article. Right up front, where are we going, and what will be the payoff when we get there. Don’t leave me wondering what the point is, and give me some structure to hang the ideas onto.
Well, you’re no mysterious ancient wizard. ;-)
Applies to me at least. The “Does it bring out the snacks”-test encapsulates some of that elusive difference: the snacks come out when EY posts a longer post, they don’t in your case.
When Eliezer has edited my writing, he’s had a pretty good (gimlet!) eye for knocking out unnecessary adjectives, clauses, pushing for more specific examples, etc. So I find those sessions pretty helpful. Luke, have you had Eliezer edit your stuff to pin down what changes (and vice versa)?
We ran an experiment once. People can still participate.
If your last name begins with A-K, read this post and vote on it.
If your last name begins with L-Z, read this post and vote on it.
Data point: I read them both (in the order you asked, and voted how you asked), then I even compared them paragraph by paragraph, and I can’t really tell which is which. Or, at least, I can’t detect an obvious “Eliezerness” in either.
Details in rot13 to avoid bias if anyone else wants to compare them: Gur frpbaq irefvba frrzf “zber rqvgrq”, ohg gung’f zbfgyl orpnhfr ba n fvqr-ol fvqr pbzcnevfba gurer ner n ovg srjre cnentencuf. Nyfb, gur frpbaq vf n ovg zber fgbel-yvxr va gung gur hfr bs cnfg grafr vf zber pbafvfgrag. (Gur svefg frrzf gb fxvcf n ovg orgjrra cnfg naq cerfrag.) Tvira gur pbagrkg bs guvf cbfg, vs V unq gb pubbfr V’q thrff vg’f gur frpbaq bar Ryvrmre rqvgrq, ohg vs V pnzr sebz n qvssrerag pbagrkg V qba’g guvax V’q unir abgvprq nalguvat.
Personally I usually prefer your style, mostly, I think, because I am impatient. I want to know what the writer’s point is right away and then I like to get right into his supporting arguments so I can determine whether he’s made the case well enough to convince me. There are other times, when I’m more more relaxed and have more time when I may enjoy his style but they are the exception.
I think it is largely a question of preference. For LW’s purposes, wouldn’t the sole purpose of any communication be to provide an accurate map of the territory, regardless of style?
And I think the answer to which authors’ writing best accomplishes this would depend on the reader.
Personally, I have enjoyed EY’s writing as way to view a concept in different way—it’s a springboard; providing the energy and motivation I need to research the concept, and related considerations, further.
What are his reasons for suggesting this?
I agree. Personally, I prefer your writing style. Eliezer’s writings strike me as being excessively long-winded and imprecise. As other commentators have pointed out, his Sequences generally have a very low information-to-words ratio—most people, I suspect, would be significantly better off reading the primary sources instead of spending time reading the Sequences.
The thing is, for a lot of people Eliezer is just much more interesting to read, in spite of the long-windedness.
Do you prefer...
[pollid:537]