Hide table of contents

Let me try to summarise my reaction here, because there is no way this post can make sense without a preamble into methodology…

Just because you have a large overhang of purchasing power doesn’t mean you should explicitly search for ways to spend it. In fact, conjunctive search is just exponentially inefficient with the number of weakly correlated criteria you're compromising between (e.g. searching for plans that cost money that also seem like they are impactfwl enough to be worth the cost). Especially if you’re searching for peak impact where correlations break down and tails come apart. I think EA Global conferences still persist as the standard for international community building because it satisfices for impact while also looking like it at least does something legibly impactfwl with the funding overhang.

The general rule is that you should first do something like searching purely for high impact, and only then do accounting to check if you can afford it–or whether it costs anything at all. Spending should (almost) never be a proxy to optimise for.

But we needn’t keep it up just because we’ve grown accustomed to it. When you start worrying about having prematurely overoptimised yourself down a narrow part of the search tree, the sensible thing to do is to take a step back and check whether you’ve missed any fruitfwl branches at the base. It would for the same reason be a mistake to look just at current practices and try to nudge them in your preferred directions without even considering the zoomed-out strategic perspective. So here are some of my not-entirely-marginal suggestions.[1]

I think the EA community is better served by...

  1. Optimising for valuable long-lasting friendships
  2. Moving almost all real-life EA conferences online
  3. Facilitating better online communication channels and meeting places, and optimising them at least as creatively as real-life conferences have been optimised
  4. If you’re trying to facilitate long-term connections, optimise specifically for that
  5. Notice and take advantage of potential network effects to maximise happiness
  6. Concrete suggestions under the pony
Princess Celestia knew what she had to do. She had to satisfy their values through friendship and ponies.”


  • The metric EAG(x) teams often use to measure impact is number of connections, where a “connection” is defined as “someone you feel comfortable reaching out to to ask for a favor.” This doesn’t distinguish between purely career-related connections and wellbeing-related connections, yet I think their paths to impact are very different, and if you want to maximise either, you should optimise specifically for that. I’ll call the latter “friendships”, and say “connections” to refer to the all-inclusive category.

Possible misconceptions about EAG

Whatever you think about the value of real-life vs online friendships, keep in mind that the connections EAGs generate are almost exclusively online. People fly in from all over the world, make new friends over the weekend, and fly back separately. Whether they meet online or irl, that won’t make a difference to how they communicate post-event if they wish to keep in touch.

Now, this isn’t a complaint, since online connections can be really valuable. But in order for two people to keep in touch after having connected at an irl conference, they have to initiate it via a separate communication channel that they don't have the habit of using yet. This might not sound like a big deal, but the bonds are fresh and uncertain, which means we should be even more wary of underestimating trivial barriers. If you want to encourage connections that have a higher chance of actually lasting, it makes sense to try to facilitate their creation via the same communication channel that they would use to keep in touch later.

"I'm a bit skeptical of the value of "connections" at these types of events, and I tend to think that they are a bit fake. But you could test this, e.g., you could ask a random number of participants to actually follow up on their reported connections, and see whether they respond, or you could ask in a few months whether people actually followed up.” – NegativeNuno

It’d be interesting to measure how long those connections last, but also the extent of attendee overlap between EAGs (since the marginal benefit decreases), and the difference in important EA friendships before and after these events. If it’s the case that most of the connections that people make during the event are reported by the kinds of people who tend to make a lot of connections, then the marginal welfare-add of those connections isn’t going to matter as much. 

And to me, that seems kinda plausible. I get the heebie-jeebies when I think of attending crowded irl conferences like that. I imagine that the people who tend to attend these conferences are most likely to be proactive about finding connections, and more often friendship-satured. Which would mean that they’re the target population that is least likely to counterfactually benefit from iterating their “connection” count.

If I were trying to optimise for facilitating valuable EA friendships, I would focus on the people who aren’t already friend-saturated, and I would try to enable those friendships to form via a communication-channel (virtual or otherwise) that they can continue to use for the long-term.

I think it’s fairly common that the value of connections is pareto-distributed. My closest friend has been more than a thousand times as usefwl to my expected altruistic impact than my median “connection”. Hence why measuring the number of connections sounds severely misguided to me, especially when you have opportunities to optimise more closely for the correlates of peak friendships.

After we hosted the Unconference in the EA Gather Town for ~60 participants, we expected some of them to start coworking there later. But we had maybe one or two people who returned to check it out. The general lesson to me is that short-term events rarely inspire long-term engagement unless you optimise specifically for that.

Target populations

The EAGxNordics team from 2019 tried explicitly to target less-connected, less-activated EAs, and wrote an excellent postmortem on it:

“Our main target group was people with good EA understanding who aren’t taking action, and we therefore sent out invitations to these people before publishing the tickets to everyone. … In the end, we were not successful in reaching our target group. We now think an EAGx is not the best way to reach such people. … We also see from the results that most attendees (79%) already had a strong EA network and were taking significant action, making it hard to achieve our main goal of “activating” more EAs.”

Contrast this with the EA Pen Pals project beta run, which attracted approximately 33 people who were randomly matched for online 1on1s.

“We originally designed this project to help people with limited access to a local EA community, and it seems we attracted our target demographic - 53% of respondents had no local community available, and only about 50% attended any local meetups.”

17 of them responded to the survey, and 5 of them responded ‘Yes’ to “You made a friend!”. 5 made a friend! That’s huge compared to input. I think the people who are differentially less attracted to real-life EAGs and more attracted to connecting to people online (e.g. because they’re shy like me) will benefit more from organised efforts to facilitate this.

I’ve encountered the argument that relentlessly resourcefwl people–the people who are most likely to do more good in the world–will take care of themselves, so we needn't try to proactively help them along except for putting them in touch with the EA community. But that thesis has sufficiently important counterexamples that I don't put much weight on it.

“When I consider applying to EAG, I ask myself whether it's worth taking up a slot that would otherwise go to some bright-eyed college student who has been dreaming of going to EAG for years and is going to consider it the highlight of their life. … I think there could be hundreds of people in this category, including some of the people who would benefit most from attending.” – Scott Alexander

Alas, the Pen Pal project didn’t get funded, and the main bottleneck their users identified for the usefwlness of such a service was not enough users. When you have a project with potentially exponential network effects that is bottlenecked by a coordination problem, you can’t cut its funding before it reaches maturation and claim that that was a fair test.

While friendships aren’t the only thing these conferences are optimising for, it’s one of the things. And given that I don't think they serve that function very well, marginal (and purely-optimised) efforts to actually serve that function might be more promising.

I can’t remember where I heard it, but “leadership is the power to make compelling invitations.” I don’t have leadership, so I can’t convincingly invite a large enough group of people all at once in order to try to defeat a coordination problem that will benefit everypony.

Network effects

Moloch the vast stone of war! Moloch the stunned governments!

If you’re trying to find huge opportunities to create value, look for feedback loops. Effective altruism itself seems to have been experiencing some exponential growth in popularity lately. This is a general possibility for things when popularity depends on its popularity–aka has a network effect.

I think the value of a group of friends increases pretty sharply with increased interaction. Luise talks about the theory of change for EA hubs like the Bay Area–she calls it an “all-year EAG”.

“Interacting with passionate, value-aligned people on a daily basis feels very motivating and nourishing. Being able to talk about your work with others and get excited together is nice. I personally have never worked so much in my entire life and it’s by choice.”

University groups are one of the most cost-effective wellsprings of value in the world that I know of. But what makes them effective? Well, for one, they have the luxury of selecting from a very pre-selected group of talented people who haven’t yet settled into their life plans. But another reason is that they have the opportunity to spend a lot more time together because they’re on the same schedules and live close enough that they can form strong friendships. One of kuhanj’s main pieces of advice is to

“Focus on retention and deep engagement over shallow engagement. Highly engaged members planning on dedicating their career to doing the most good will likely be much more impactful than people who are sympathetic to the ideas, but not enough to change their career (or donation) plans. Additionally, highly engaged members will help you grow your group.”

And if you think these effects are exclusive to irl interactions, I’d like to know why, and whether you’ve spent more than 5 seconds thinking of creative solutions to that obstacle.

What sphinx of cement and aluminum bashed open their skulls and ate up their brains and imagination?

The reason large-scale online hubs haven’t spontaneously sprung up yet is not because the community wouldn't benefit greatly from it, it's rather because it's a coordination problem, and no one has tried investing enough resources into making it happen. There have been numerous local attempts at getting something like this off the ground, but it can't sustain itself unless it reaches critical mass, so just trying to do it locally was doomed to fail from the start.

Moving EA conferences online seems good, but I’m much more excited about the huge potential upside of long-lasting projects with compounding network effects

The EA Pen Pals project and the Virtual Coffee matching thing on EA Anywhere’s Slack are welcome experiments, but … I don’t think you can point to this and say that we as a community actually tried. Projects like this have had maybe 1% of the concerted optimisation power applied to them compared to the history of all EAG(x)s and other irl projects for facilitating connections. This branch has not been explored yet, and there could be fruits here.


So far, I’m only aware of six online community building projects that don't seem to have hard caps on their upward potential.

  • EA Anywhere is the largest and they’re doing weekly events, discussion groups, reading groups, hosting randomly-matched 1on1s and they have an active slack. They're also hosting the next EAGxVirtual in October.
  • EA Gather Town is a project I'm part of to build an always-on virtual meeting place for coworking and having conversations. We explicitly try to optimise it for being the online equivalent of an EA hub/office. It's been surprisingly successfwl, but further value is still bottlenecked by coordination. We have plans to do something about this, though!
  • EA Virtual Reality is an active community of EAs in VR. They have an ongoing weekly alignment course, frequent hangouts, talks, and they've introduced a fair amount of people in the neighbouring transhumanist community to EA.
  • EA Co-Working Discord is another coworking platform hosted on discord where people use the in-discord video feature to cowork.
  • EA Focusmate is a coworking service with an active community of EAs who schedule times to match up and cowork with each other.

If you would like to help with ambitious projects in this direction, or just offer support to help with coordination problem bottlenecks (e.g. assurance contracts), send me a message or book a call. : )

  1. ^

    To be clear, I’ve not spent much time taking an unbiased look at the base of the search tree here. After landing on my perspective, I’ve spent most of my search time looking for arguments that support it rather than the reverse. I've also tried to convey the supporting case without being as epistemically rigorous as I would have been if I were a decision-maker in this area. I just don’t have access to enough relevant information to make these decisions, so I think I serve my role here best as someone who highlights possible opportunities.





More posts like this

Sorted by Click to highlight new comments since:

Hi Emrik,

[I work on the CEA events team]

Thanks for writing this up :) I don’t totally follow your argument here, sorry. I’ll respond below to some specific points, but please let me know if I’m misunderstanding your core argument.

I agree online conferences are valuable for many of the reasons you outlined and that’s why I’m excited about EAGxVirtual (applications launching soon!). I also agree that deeper, long-lasting connections are important and that our events should try to create them (I think retreats are particularly good at this, for example).

However, I disagree with some of the takes here:

Moving almost all real-life EA conferences online

At a high-level, virtual events score lower on “likelihood to recommend” and “number of new connections made” than in-person events. I strongly think they’re still worth doing largely because it’s too costly for many people to attend in-person events.

Furthermore, I don’t quite see how favouring virtual conferences leads you to recommend cancelling in-person ones - you can have more of both!

“The connections EAGs generate are almost exclusively online”

What’s your evidence for this? My sense is that being physically co-located with other EAs is highly valuable, and that many people move to EA hubs for this reason

 Hence why measuring the number of connections sounds severely misguided to me, especially when you have opportunities to optimise more closely for the correlates of peak friendships.

I would guess that number of connections correlates quite closely with new peak friendships (the more people you meet, the more potential close friends you meet), so I don't think it's misguided. 

After talking a bit with Ruth Grace Wong about it, I admit to having framed this entire thingthink I should just reframe this entire thing. I’m not trying to reduce irl activity. I’m more concerned with the idea that irl EAG(x)s are short, and connections made are largely long-distance. People meet wonderfwl people they’d like to keep in touch with over the weekend, but because they live far apart, it’s hard to maintain the connection.

As you point out, moving to a hub makes it a lot easier to maintain connections because there’s less stress involved with seeing each other frequently. But realistically, most people aren’t going to move to stay in touch, either because they lack the opportunity or inclination. And even if most people did move, I think there’s still a lot of value to making it easier for people to keep in touch after they fly back to their respective homes from just having met at EAG. I think the virtual aspects of community building have been neglected, and marginal efforts to optimise this can pay manyfold greater dividends. : )

“I would guess that number of connections correlates quite closely with new peak friendships”

My best friends have all come about because we’ve shared spaces where we serendipitously see each other frequently (e.g. the Gather Town coworking space we use). Regular contact is really important. I’ve met several people whom I would consider very usefwl to interact frequently with, but because those interactions would have to be scheduled, they rarely reach their potential. I think people maybe underestimate how much of a hassle it can be for most people to keep scheduling things, even if the call itself is very positive. The scheduling aspect just makes it a lot costlier to maintain.

I’ve had conversation-length interactions with many more high-compatibility EAs than the number of EAs that I’ve shared office space with, yet it’s overwhelmingly the EAs I’ve shared daily interactions with who have ended up important to me (important in the sense that they keep me going and help & inspire me to do good). On Bayes then, this implies that conversation-length first-impressions have very inferior conversion rates to lasting friendships compared to online shared office spaces--even if the latter generates fewer first-pass "connections". Hence why the metric itself seems ill-advised when applied to different activities.

This is part of my reasoning for why I want EAG(x)s to be specifically optimised as on-ramps for regular meeting-places online (virtual hubs), like EA Anywhere, mastermind groups, EAVR, or especially coworking office spaces like EAGT or Complice.

The latter spaces also serve as places for the community to just hang out online, and attendance is less explicitly focused on having impact.

The sign of having a healthy community is that when people burn out and need to take a break from trying to have impact, they don't also feel the need to take a break from the community. The community is their refuge.

Anne Wissemann (from the EA Workspace project in 2015) also talks about how important it is to have spaces to casually bump into each other for the maintenance of friendship:

“Many of us have full schedules and Skype calls to uphold friendships pose a real cost on time and flexibility and can be a pain to arrange. Keeping in touch via the EAW is easy - you’re bound to bump into each every once in a while if you work at similar times. It doesn’t cost you any additional time since you were having a break anyways, and you can make progress on your projects in the meantime. Win-win!”

The effect both on happiness and akrasia on finding important value-aligned friends in a daily-interacting coworking environment is hard to understate, and the solutions are far from saturated with all the people who could benefit from them.

Btw, I'd be eager to talk about some plans we have to grow the online office/hub thing, and how CEA might be able to help, if you have time and inclination. Feel free to book a time with me if that sounds interesting to you! Otherwise, thanks for commenting on my post and have a nice day. : )

For the pony!

I think the virtual aspects of community building have been neglected, and marginal efforts to optimise this can pay manyfold greater dividends. :)

I agree with this! :) 

On Bayes then, this implies that conversation-length first-impressions have very inferior conversion rates to lasting friendships compared to online shared office spaces

I think you’re updating a bit too much on your own experiences here. This seems possible to me, but I also know that many people find virtual interactions more draining (myself included).

Very inferior also doesn’t sound right to me - I regularly talk to people who make close friends at conferences and retreats (again, retreats seem better for the thing you’re describing) and our data suggests this is a regular occurrence.

I don’t have data for the virtual spaces > close friends pipeline, but I would be surprised if it was markedly better. But let's do both!

This is part of my reasoning for why I want EAG(x)s to be specifically optimised as on-ramps for regular meeting-places online

Promoting these things alongside hubs and future events seems good :) Again, I don’t see why we should privilege online spaces to the extent that you’re pushing here, and I disagree that EAGx events should specifically optimised for anything.

The plans for a virtual EA office look really exciting :) I’ve requested comment access to that doc :) 

I think I have a tendency to explain/talk things with much simpler models than I actually operate with, and I should adjust. But to be clear, I think yeah, EAG(x) shouldn't be exclusively optimised as on-ramps. What I meant by "specifically optimised" wasn't "exclusively", but I realise I wasn't being very clear. And I agree that retreats sound more conducive to friendships than online interactions for most people.

What I hope for is just that the a virtual hub/coworking office that's always accessible (this could be EAGT, but anything that serves the purpose is fine) is advertised at EAG(x)s as a way to keep in touch with the people you meet. And perhaps that CEA helps advertise the coworking week starting at the 24th to give it some credibility. But that's only if you think it's a good idea, and I don't want to come across as too much of a salesperson. :p

I'm on a vacation 'till Monday, so I won't be optimising the document much until then, but thanks for noticing us. Have a happy weekend! 

More from Emrik
Curated and popular this week
Relevant opportunities