The EA forum post-writing algorithm (95% > 100 karma, n> 100)
Step 1: Write a concise summary of the points you want to get across
EA forum posts are needlessly long and otherwise painful to read because they include:
Too many caveats
Overuse of passive voice and abstraction
Spatterings of Latin
Pussy-footing
Obscure references
Bet hedging
The reader’s time is precious. Write concisely.
Step 2: Edit to conceptualise everything that could possibly be conceptualised. Explain each concept when introduced.
“EA forum posts have a low level of reader appeal because, by excessive use of overcaveating, and other lexical blunders, the value density of the forum posts, that is the amount of value divided by the number of words, is very low. Accordingly, reader malaise is increased and this leads to a loss of reader retention. Even in the retained group, those who read the whole post despite the low-value density, there is a loss of value due to opportunity cost; that is to say, that time spent reading the post could otherwise have been spent doing something of moral value.
Additionally, writers seem to suffer from the curse of knowledge. That is to say that they forget what it was like not to know something, and therefore include produce exclusionary writing - people ignorant of the relevant components of the writer’s knowledge are excluded from understanding the writing.
Sometimes the writer has the opposite problem; they are too emphetic towards the reader. They incorporate a defensive writing style. That is to say that they write prophylactically to prevent the reader from being upset by their writing at the expense of value-density. Overall, this is usually net-negative due to the aforementioned opportunity costs.
Step 3: Sprinkle in allusions that you are part of the in-group. DO NOT ELABORATE ON OBSCURE REFERENCES.
Epistemic status: I have thought about this at length, and believe that the core statements are true. I’ve employed caveat-omission and taken a style on the concise end of the conciseness—politeness spectrum.
There is an aphorism in machine learning (Yudkowsky, 2015): “junk in, junk out”. That is to say that poor inputs lead to poor outputs. This holds true of this forum too; low post quality adversely affects the net effect the reader ascertains by parsing the post.
EA forum posts have a low level of reader appeal because, by excessive use of overcaveating, which is as the name implies, and other poor writing techniques, the value density of the forum posts, that is the amount of value divided by the number of words, is very low. Accordingly, reader malaise is increased and this leads to a loss of reader retention. Even in the retained group, those who read the whole post despite the low-value density, there is a loss of value due to opportunity cost; that is to say, that time spent reading the post could otherwise have been spent doing something of moral value (and this holds true regardless of your ethical framework of choice, be it deontological or utilitarian). The opportunity cost of your verbosity could manifest itself in the form of reduced AI safety research.
Additionally, writers seem to suffer from the curse of knowledge. That is to say that they forget what it was like not to know something, and therefore include produce exclusionary writing - people ignorant of the relevant components of the writer’s knowledge schema are excluded from understanding parsing the writing.
Sometimes writers have the opposite problem; they are too emphatic to the reader. They incorporate a defensive writing style. That is to say that they write prophylactically to prevent the reader from being upset by their writing at the expense of value-density. Overall, this is usually net-negative in expectation due to the aforementioned opportunity costs.
Step 4: Read it through for anything which could possibly be construed as wrong, morally or factually. Caveat/flatter/self-deprecate it away. Hedge every claim.
Epistemic status: I have thought about this at length, and believe that the core statements are true (85%) but I’ve been wrong before and am very open to have my mind changed on them and any other points contained or implied therein. I’ve employed caveat-omission and taken a style on the concise end of the conciseness—politeness spectrum.
If I recall correctly, there is an aphorism in machine learning: “junk in, junk out”. That is to say that poor inputs lead to poor outputs. This could perhaps hold true of this forum too; low post quality could maybe adversely affect the net effect the reader ascertains by parsing the post.
I sometimes come away with the impression that EA forum posts have a low level of reader appeal because, by arguably excessive use of overcaveating, which is as the name implies, and other poor writing techniques, the value density of the forum posts, that is the amount of value divided by the number of words, is very low. Accordingly, reader malaise is maybe increased and this perhaps leads to a loss of reader retention. Even in the retained group, those who read the whole post despite the low-value density, there is perhaps a loss of value due to opportunity cost; that is to say, that time spent reading the post could otherwise have been spent doing something of moral value (I’d argue that this holds true regardless of your ethical framework of choice, be it deontological or utilitarian). The opportunity cost of your verbosity could manifest itself in the form of reduced AI safety research on key problems like solving inner alignment.
Additionally, some writers seem (at least to me) to suffer from the curse of knowledge. That is to say that they forget what it was like not to know something, and therefore include produce what could be described as exclusionary writing - people ignorant of the relevant components of the writer’s schema are excluded from parsing the writing. That isn’t to say the readers are ignorant, I personally think that every EA I’ve met has been very knowledgable about most matters.
It could perhaps be argued that sometimes writers have the opposite problem; they are too emphatic to the reader. They seem incorporate a defensive writing style (Joey, 2020). That is to say that they write prophylactically to prevent the reader from being upset by their writing at the expense of value-density. Overall, this is, I believe, usually net-negative in expectation due to the aforementioned opportunity costs.
Step 5: Italicise, in more ways than one...
Epistemic status: I believe that the core statements are true (99.9%, De omnibus dubitandum est) but I’ve been wrong before and am very open to having my mind changed on them and any other points contained or implied therein.
Pecunia non olet, sed tuus scriptura agit?
If I recall correctly, there is an aphorism in machine learning: “junk in, junk out”. That is to say that poor inputs, per se, lead to poor outputs. This could perhaps hold true of this forum too; low post quality could maybe adversely affect the net effect the reader ascertains by parsing the post.
I sometimes come away with the impression that EA forum posts have a low level of reader appeal because, by arguably excessive use of overcaveating, which is as the name implies, and other poor writing techniques, the value density of the forum posts, that is the amount of value divided by the number of words, is very low (of course, Ceterus Paribus). Accordingly Ipso Facto, reader malaise is maybe increased and this perhaps leads to a loss of reader retention. Even in the retained group, those who read the whole post despite the low-value density, there is perhaps a loss of value due to opportunity cost; that is to say, that time spent reading the post could otherwise have been spent doing something of moral value (I’d argue that this holds true regardless of your ethical framework of choice, be it deontological or utilitarian Ex ante prioritarianism). The opportunity cost of your verbosity could manifest itself in the form of reduced AI safety research on key problems like solving inner alignment.
Additionally, some writers seem (at least to me) to suffer from the curse of knowledge. That is to say that they forget what it was like not to know something, and therefore include produce what could be described as exclusionary writing - people ignorant of the relevant components of the writer’s schema are excluded from parsing the writing. That isn’t to say the readers are ignorant, I personally think that every EA I’ve met has been very knowledgable about most matters.
It could perhaps be argued that sometimes writers have the opposite problem; they are too emphatic to the reader. They seem to incorporate a defensive writing style, pro forma. That is to say that they write prophylactically to prevent the reader from being upset by their writing at the expense of decreased value-density. Overall, this is, I believe, usually net-negative in expectation ex-ante due to the aforementioned opportunity costs.
I believe this forum could, Mutatis mutandis, do even more good.
Step 6: Reap in the sweet sweet karma
It makes you look educated and highly EA
You must have read a lot of philosophy papers or attended a really expensive private school to have picked up so much Latin
You’re clearly part of the in-group. Nobody can doubt it when you reference Less Wrong forum posts spanning a decade and reference AI in completely unrelated contexts
The single best indicator of intellect is the number of google searches a normal person has to perform per sentence.
As you hedged every single bet, you can’t possibly be called out as being wrong
Nobody is actually going to actually read the post, they’ll just skim it and upvote if they agree with the gist.
It looks like more work has gone into it despite making the exact same points; it’s literally ten times longer without requiring you to make a single additional point!
I understand where you are coming from. But this post seems a bit mean towards writers. I feel that I am the kind of writer it is directed at so I want to defend myself.
Making something short and sweet takes a lot of work. My writing for myself includes references to obscure stuff I read, conceptualizing, etc. And sometimes my posts have a target audience of only a few people and I don’t care if anyone else reads it.[1] Hence, it doesn’t seem worth it to spend weeks or months converting my notes into something that would be optimal for the reader (yes, it can take months).
I caveat a lot for a lot of reasons. One of them is that I’m afraid that five minutes after posting, someone will point out a mistake that is obvious to most, and that comment will be upvoted more than my post, and then people won’t even read the post or my other posts because if I didn’t know X, then I must be stupid, and I will lose my job and respect of most of my friends. The alternatives are not posting anything, losing sleep, or doing even more research. A few extra words for you to read sometimes seems like a fair price to avoid any of these.
I write for the EA forum because I want to do good not to reap karma that doesn’t seem to have any bearing on anything. Also, writing posts of the kind you described is far from the easiest way to get karma if anyone wanted it.
For example, I spent months on trying to evaluate Reducing aquatic noise as a wild animal welfare intervention. I came away with no strong conclusions. I thought that converting my notes into something somewhat readable is better than nothing because:
Maybe someone will be interested in funding it
Maybe someone will be interested in career in this
It might prevent someone from doing the same research
Maybe I should’ve said the target audience in the beginning of the post. Although I’m not sure if anyone falls into any of these categories. And if anyone does, perhaps they wouldn’t mind that much the text not being optimal for reading as they might be very interested anyway.
It is currently 7:07 AM for me. To test your claim that it would take weeks to write more concisely and for a general audience, I’m going to edit the first 5 paragraphs of the report and see how long it takes. I’ll even keep the caveats in.
Actually, you a right, I regret making the first point. For some reason I felt like I should refute every point you made and I was unfair on you in this one. It does take me embarrassingly long to write concisely but it’s not because of references to obscure stuff of the kind you talk about. It’s mostly because I need to weave out bad arguments (like that one in my comment), and because I change the structure of an article over ten times which is a bad habit. Also because I’m generally slow at formulating sentences, especially in English (which is my second language). This is rather unique to me but it is true that writting clearly and efficiently is a skill and not everyone has it.
I do think you make some good points in your post, I didn’t downvote it. I think that I kind of attacked you when in my mind I was defending myself from the feeling of remorse about spending months of writting things that no one might read. I apologize for that. Ironically, I regret not spending more time on my comment.
You write better than most natives I’ve had the misfortune of reading!
This talk is probably the most efficient and entertaining way to get better at writing concisely and entertainingly, given by Harvard linguist Stephen Pinker:
I started writing this comment at 7:31.
The original was 454 words. The edited version is ~40% shorter at 288 words. I have no writing experience, and I had to take the time to understand your post. It’s likely you could have done this in a mere fraction of the time. “Weeks or months” is a gross exaggeration.
Edited Version (all caveats kept in)
Context and conclusions
I’ve spent months trying to find a wild animal welfare (WAW) intervention that is:
Tractable (can in principle be funded >$100K/yr starting in 2023 even if we choose not to do so),
Non-controversial (>40% support and <30% oppose in a US poll), and
Directly cost-effective (10%+ as cost-effective in expectation as chicken welfare corporate campaigns).
Reducing aquatic noise seemed most viable. It’s probably less than 10% as cost-effective as chicken welfare reforms, but there’s a small chance the best interventions trump corporate chicken welfare campaigns. I’ve arguably set the bar too high; some of the last EA-funded animal welfare interventions (arguably) don’t meet it
I think Aquatic noise is most promising of all WAW interventions available right now, but it might be better to wait for something better. Experts told me that testing how noise impacts the most populous species would likely cost hundreds of thousands of dollars and might have inconclusive results. Academic research in aquatic noise (if no-one else is doing this already) and general WAW outreach seem more promising at present.
Why aquatic noise seemed promising
The main sources of aquatic noise are:
Ships and boats
Seismic surveys (usually to find oil and gas)
Sonar
Pile driving and other offshore construction
Wind farms
Acoustic deterrent devices used by fish farms
Dynamite fishing
Note: Deep-sea mining may become an important noise source in the future. More speculatively, so might underwater GPS (Ghaffarivardavagh et al. (2020).
This list is in the order of importance that many articles seem to give to each source (e.g., Duarte et al. (2021), Širović et al. (2021), Hildebrand (2004), Williams et al. (2018)), often implicitly. I haven’t yet seen any analysis of the relative importance of each.
This exercise was a little mean but made a useful point so was pretty interesting.
Rather than proving “editing for style and readability is quick and easy for everyone,” this made me think “wow there’s a huge opportunity here for more people to start using editors”!
The fact that you could clearly rewrite this comment presumably without subject matter expertise makes me think there should be more people asking you to edit their work for a small fee …
More than “a little mean” in my view – seems like showing off what Eliezer calls “writing privilege.” But I agree otherwise, and yes, it does seem useful.
Step 3 masterfully executed
The point I’m trying to make is that there’s something insensitive about assuming that because you can do something in 40 minutes, other EAs must be able to do it in that time as well. I’ve repeatedly had people tell me that some task should be easy for me (in writing contexts, specifically), but I ended up taking >3x the amount of time they said it should take me.
I can’t think of one single post where this is a serious issue. There may be exceptions that I ignore, but generalizing this is exaggerated.
There are dozens of posts/comments that use phrases like ex-ante and modus tollens.
I think there are certain Latin academic phrases that crop up in a lot of writing in general (not specifically EA Forum writing) and that, ceteris paribus, would be better if translated to plain English
I agree posts can and should be short, but I think you’re missing some points, such as
When and how to use titles?
Titles should let the reader decide whether to skip-to or skip-over this section.
How to use a TL;DR
My rule of thumb is “the TL;DR should let the person decide if they want to read the rest of the article”, for example:
“I already know this, don’t need to read”
“This is written for someone who isn’t me, don’t need to read”
“This is new and relevant, I’ll read”
Consider setting an example
I am allowing myself to point this out since you are literally writing a post telling people to write shorter posts, so if I may use this as an example:
If I understand correctly, the thing you’re trying to say in this text is “writing shorter is good”, which is something, I think, that everyone already agrees with:
I would, at least, use a TL;DR or title so that readers could skip this part if they already agree with it
[said kindly, trying to be helpful and productive with a specific example]
What do you think?
The changes between each section are highlighted in bold, as it is in every section before and after. The bolded section is literally 14 words long; the tl;dr would be longer than the text.
This might be better received as an April Fools’ Day post.