Summaries are underrated

I will not eat the bugs, I will not read the The Precipice

[Epistemic confidence: 80%, some bets at the bottom]

Summary:

  • 1 min video summary of this article

  • There are not enough easy summaries of EA works

  • This makes it harder for people to learn their content, which costs time, energy and potential EAs

  • I estimate the time cost of this problem to be from $160k to $140mn (90% CI) with a median of $6.5mn

  • I reckon we could see huge improvement with about $20k of summary prizes

  • We should pay prizes for summaries, put our works in audio form, and have a community wiki somewhere

  • Some possible counterarguments are: (1) summaries are hard to judge, (2) summaries provide low resolution versions of ideas with no positive tradeoffs and (3) I am uniquely poor at consuming long works. I’m not convinced by any of these counterarguments.

The Problem: Not Enough Summaries

You hear about this cool thing called EA. You decide to learn more about it. If you are lucky, someone points you towards the EA Handbook. This is a focused summary of effective altruism. Here’s the opening page of the Handbook[1]:

It is very long. 8 chapters, with 92 posts.

And after that, if you want to understand EA in more depth, you are expected to read Doing Good Better. And the Precipice. And listen to 100 episodes of the 80k podcast. Or do an 8 week fellowship.

In short, good luck[2].

For lots of top-tier EA content, either there aren’t any summaries, or the summaries are of medium quality. Summaries and snippets make it easier for people to figure out if content is something they want to engage with. By having too few, we make it harder to enter EA and harder to upskill.

Let’s look at some of the foundational works.

Doing Good Better

The best summary of this that I found was by Lara M. As far as I’m aware, it’s not linked anywhere on the Forum, or included in the Introductory Fellowships. The EA Forum wiki page looks like this. That’s the whole article. Since sharing this I’ve been sent a number of summaries written by university groups. Someone looking for infromation about EA would not have access to these.

The Precipice

The best summary I found was on Wikipedia. It’s only three paragraphs long, so it isn’t detailed. Again, there’s no summary on the Forum

80,000 Hours podcasts

This is pretty good—the podcast has an introductory series and all podcast episodes have a written summary with some highlights.80k have a list of questions that each episode answers, allowing me to scan through and decide whether I want to listen.

https://​​80000hours.org/​​podcast/​​episodes/​​holden-karnofsky-open-philanthropy/​​

However, I still have to do a lot of clicking then scrolling to get here. There wasn’t a clear summary page I could find.There are no summaries of the podcasts on the Forum.

The Sequences

Lightcone gave out a short version of the Sequences n book form at EAG SF. This included selected blogs from the original Sequences, but they were still full-length.LessWrong has a set of summaries of blogposts from 2007. This is very good, but it only exists for 2007.

Forum articles

Many Forum articles begin with a summary. This isn’t good enough. I want to know if I should click on the article. You can hover for a preview, but I find this works only about 50% of the time.

Recently, Zoe Williams announced that she plans to publish weekly summaries of popular (40+ karma) EA Forum and LessWrong posts. This is great, I’ve subscribed.[3]

What about the EA Wiki?

The EA forum has a wiki, but it is used to give a brief description of a topic before showing a list of articles in that topic. It does not seek to summarise them. I have written before that I’d like to see a new type of post using the same technology but with a different framing.

As a final example, when the Economist published ‘What to read to understand “effective altruism”’ the top recommendation was.. The introductory series of 80k podcasts! This is about 20 hours of content. If a top flight news organisation can’t find an accessible introduction to point people then random members of the public are probably trying to find out about us and missing out.

In a better world, I suggest that for each piece of content we consider important, there would be:

  • A Forum linkpost[4]

  • An audio version of the text (this one is largely covered)

  • A Forum summary, perhaps a topic article which links to many posts

  • A Twitter thread

  • A Tiktok

  • Diagrams illustrating key points in the content

Currently, for many important pieces of EA content, these things are not available, which makes the content harder to access.

I think we have a great opportunity here: if we produce more high-quality (but short) summaries of EA material, we can lower the barriers to entry for new EAs, save existing EAs time, and help spread our top content more. Many people don’t have enough bandwidth for a 40 minute podcast:

  • Random twitter users

  • Busy startup founders and billionaires

  • Students who don’t know an EA

  • Politicians

I estimate that this costs the EA community $160,000 to $140 million (90% CI)

Inspired by Nuno Sempere and Sam Nolan, I used the estimation language squiggle to estimate the value lost to the EA community because of the lack of accessible summaries. I estimate the number of EAs, how much content they consume, and how much value summaries deliver despite being shorter than original works. Then I estimate how much of one’s reading might go to summaries instead (10% to 50%) and how valuable an EA’s time is. The estimate spits out the following.

current_EAs = 1500 to 12000
new_EAs_this_year = 500 to 5000
podcasts_listened_this_year = 0 to 250
podcast_time= .5 to 3
top_blogs_read_this_year = 5 to 500
blog_time = .1 to .5
top_books_read_this_year = 0 to 3
book_time = 5 to 20
how_much_shorter_are_summaries = .01 to .5
what_percentage_are_summaries_used = .1 to .5
how_much_of_the_full_value_is_in_the_summary = .1 to .95
value_of_EA_time = 10 to 1000
dollar_value_of_savings = (current_EAs+new_EAs_this_year)*(podcasts_listened_this_year*podcast_time+top_blogs_read_this_year*blog_time+top_books_read_this_year*book_time)*what_percentage_are_summaries_used*how_much_of_the_full_value_is_in_the_summary*(1-truncate(how_much_shorter_are_summaries,0,1))*value_of_EA_time

This is the money saved alone. Feel free to critique my numbers or do your own calculation. But to me, this looks like a solid case that summaries have a lot to offer us. Even spending $150k on summarisation would go a long way.

What’s more, people would probably still read the content, but the summaries would help them familiarise themselves faster and get a sense of what they are interested in.

Concrete solutions

  • You could write summaries

  • Pay for summarisation

  • Make audio versions of things (this is already happening!)

  • Write syntheses

  • Summarise works from other communities

You could write summaries[5]

I have mixed feelings about this. I would really recommend that if summarisation brings you joy, you should do it. But also, I think there is a supply and demand problem here. Summaries aren’t created because they aren’t rewarded with resources or prestige. That’s the problem we need to fix.

Pay for summarisation

We can offer bounties for summaries of content that are “good enough”. I created a bounty, funded by Superlinear, for summarizing books in Twitter threads.

Here, for example, is the prize for Doing Good Better. We have had 4 entries in the two weeks it’s been up and they have been great.

All of these cost $183. Arjun told me his Epistemic Learned Helplessness tweet had been viewed ~3000 times. Ines’ EA graphs was even shared more. At this price you could have good to great summaries of all the key works in EA for about $20k. That’s 500 twitter threads. I’m confident that I or others could judge that in 10-20 hours.

Making audio versions of things

Nonlinear have worked hard to turn many top works into audio, opening them up to a much wider audience. If you like podcasts, I’d recommend it.

It would be even better if you could filter their playlist by karma or cause area, or listen to summaries of posts before you jump into an entire post.

Still, this is really good!

More graphs

Someone who commented on this just really wanted more graphs.

Syntheses

I haveve written a few syntheses on the LessWrong wiki (the EA wiki doesn’t like syntheses; they see things more as topic headers than full wiki pages). Here are my attempts, which are early works in progress, if you don’t like ‘em, make ’em better.

Other summaries

I am surprised by a lack of summaries of the best ideas of other spaces on the forum. Some high priority topics for this might include:

  • Econ 101

  • Sociology

  • Psychology

  • Feminism

  • Individuals talking about their learnings from business

It seems to me that many of these communities will have something to teach us that we might as well have one summary of.

Possible counterarguments

It’s hard to judge what a good summary is

I strongly disagree. You could judge the quality of a summary by, for example:

  • Putting the summary up as a Forum post and seeing if it gets a certain number of karma. Note that I don’t think summaries get enough karma, but this can still be used to judge between bad (0 karma) and good (20 karma)

  • Getting a set of judges to assess it. Summaries are very short; you can read many of them in an hour. I’d happily do this.

Summaries share a lower-resolution version of the ideas

Sometimes this is true. However, sometimes low-resolution summaries can pique my interest in a topic, leading me to read about it more deeply later. Sometimes the idea wasn’t that well explained in the original work and the summary gets to the core of it. Generally, I would suggest that summaries often allow more people to access the core ideas of much longer works

I am unrepresentative because I have a very short attention span

I have worried about this idea. All I can say is that politicians, members of our outgroup, and billionaires all have short attention spans for new ideas as well. Maybe their attention span is more like mine than the median EA’s.

Conclusion

Summarisation provides a lot of value and we aren’t doing much of it. As a result, our ideas are harder to understand.

It’s easier to spread ideas in their clearest, most distilled form. We should not ask people to read whole books. We should certainly not ask people to read series of long technical blogs.

The ‘median voter hypothesis’ is the idea that you get roughly what the median person in your country wants. This idea looks roughly true to me. Likewise, I think that our ability to function flexibly as a community will be correlated to how well the median EA understands the most important ideas. If we make it easier for people to understand the core ideas of EA, the larger EA coordinate better and check the work of key decisionmakers. That’s what this sequence of posts is about.

I spoke this as a voice note to https://​​otter.ai which transcribed it, then I wrote it up and Amber Ace copy edited then I added a load more spelling errors in. I would recommend Amber for editing, you can message her here. Thanks to Sudhanshu Kasewa, who talked about this idea with me. Thanks to Arjun Panickssery, Emma Richter and Nuño Sempere and others who looked over early drafts of this. (If you helped, I’d love to thank, you but I need your permission)

This post is part of a test I’m doing on works-in-progress.

Some bets I’ll offer:

  • 80% of up to $500 that in 5 years time if we ask the forum question “was it good ex-ante to spend $20k on summaries” [or equivalent] the top postive answer will get more upvotes than the top negative one, with it being made clear that it resolves this bet

  1. ^

    Initially I googled the handbook and got directed here which sent me here which sent me to the actual handbook. beware of the leopard If you know who to tell to correct the fact that googling “EA Handbook” sends you to not one, but two incorrect locations, please tell them.

  2. ^

    In long: I have a 2 hour podcast you can listen to.

  3. ^

    I don’t know how to say this without sounding like an ass, but on hearing that Zoe wrote these summaries but didn’t publish them, I told Peter Wildeford to suggest that she did. Perhaps she intended to publish them anyway but it’s very plausible to me that we’d never have got a really high quality resource had noone.. asked for it.

  4. ^

    I think it’s a good idea to produce linkposts even for books because then they can be upvoted and downvoted and we can see their karma ranking, but I won’t discuss that here.)

  5. ^

    Rohin mentioned this and it was an oversight not to have included this discussion originally