Head of Online (EA Forum, effectivealtruism.org, Virtual Programs) at CEA. Non-EA interests include chess and TikTok (@benthamite). We are probably hiring: https://www.centreforeffectivealtruism.org/careers
Ben_West
Karma: 4,599
That makes sense, thanks!
Thanks for sharing this! Do you have a sense for what the denominator is? I’ve previously tried to get some sense of this, and found it pretty challenging (mostly for obvious reasons like “people who have left EA are by definition harder for me to contact”).
I’m guessing 3-5 people is like 1 in 50 of the EA’s you know, over the course of a ~decade?
Thanks for considering this. It does seem possible to me that more people should change how they interact with EA.
I feel a little surprised by you describing this as “leaving” though – I think basically everyone I know in EA has “left”, by your definition. Do you have a different experience?
E.g. I personally meet all three of your criteria: my friends are mostly non-EA’s, even though I work at an EA organization I don’t rely on it for income, and my sense of self worth is not wholly tied to impact. But I would feel pretty surprised if someone described me as having left EA?
Hmm, if we are still talking about comparing CEA versus Ashby, I’m not sure this carves reality at the joints: it’s certainly true that people with zero experience have an uphill battle getting hired, but I don’t think CEA is unusual in this regard. (If anything, I would guess that we are more open to people with limited experience.)
the thing I find confusing is that you “didn’t have particularly strong opinions about whether EA jobs are still hard to get.”… So I don’t really understand why you present a lot of data that all points the same way, yet remain unconvinced by the conclusion they lead to.
I think I’m largely like “bruh, literally zero of our product manager finalist candidates had ever had the title “product manager” before, how could we possibly be more selective than Ashby?”[1]
Some other data points:
When I reach out to people who seem like good fits, they often decline to apply, meaning that they don’t even get into the data set evaluated here
When I asked some people who are well-connected to PMs to pass on the job to others they know, they declined to do so because they thought the PMs they knew would be so unlikely to want it it wasn’t worth even asking
I acknowledge that, if you rely 100% on the data set presented here, maybe you will come to a different conclusion, but I really just don’t think the data set presented here is that compelling.
- ^
As mentioned, our candidates are impressive in other ways, and maybe they are more impressive than the average Ashby candidate overall, but I just don’t think we have the evidence to confidently say that.
offer rate is more relevant to selectivity (if you disagree, could you explain why?)
I think it’s pretty uncontroversial that our applicants are more dedicated (i.e. more likely to accept an offer). My understanding of Ashby is that it’s used by a bunch of random tech recruiting agencies, and I would guess that their applicants have ~0 pre-existing excitement about the companies they get sent to.
I don’t see any metrics that suggest the opposite is true, or even that it’s a close call or ambiguous in any way.
The statement in the post is “CEA might be slightly more selective than Ashby’s customers, but it does not seem like the difference is large”. This seems consistent with the view that CEA is selective? (It also just implies that Ashby is selective, which is a reasonable thing to believe.[1])
--
As a meta point: I kind of get the sense that you feel that this post is intended to be polemical, like we are trying to convince people that CEA isn’t selective or something. But as you originally said: “the authors don’t seem to take an explicit stance on the issue” – we just wanted to share some statistics about our hiring and, at least as evidenced by that first comment of yours, we were somewhat successful in conveying that we didn’t have particularly strong opinions about whether EA jobs are still hard to get.
This post was intended to provide some statistics about our hiring, because we were collecting them for internal purposes anyway so I figured we might as well publish. We threw in the Ashby thing at the end because it was an easily accessible data point, but to be honest I kind of regret doing that – I’m not sure the comparison was useful for many people, and it caused confusion.
- ^
It sounds to me like you think Ashby is selective: “the Ashby benchmark (which itself likely captures selective jobs).”
- ^
Thanks yeah sorry, there is a greater change in the percentage of drop off for Ashby on-site → hired, but because we start with a smaller pool we are still more selective. 1 in 7 versus 1 in 5 is the correct comparison.
This data shows clear, consistent, and large differences all suggesting that CEA is much more selective than the industry benchmark
I guess I’m flattered that you trust the research we did here so much, but I think it’s very much not clear:
The number of applicants we get is very heavily influenced by how widely we promote the position, if the job happens to get posted to a job aggregator site, etc. To take a concrete example: six months ago we hired for a PM and got 52 applicants; last month we opened another PM position which got on to some non-EA job boards and got 113 applicants. If we hire one person from each round, I think you will say that we have gotten more than twice as selective, which is I guess kind of true, but our hiring bar hasn’t really changed (the person who we hired last time would be a top candidate this time).
I don’t really know what Ashby’s candidate pool is like, but I would guess their average applicant has more experience than ours – for example: none of our final candidates last round ever even had the job title “product manager” before, though they had had related roles, and in the current round neither of the two people at the furthest round in the process have ever had a PM role. I would be pretty surprised if Ashby’s final rounds were consistently made up of people who had never been PMs before.[1]
The conclusion of this post was “Overall, CEA might be slightly more selective than Ashby’s customers, but it does not seem like the difference is large” and that still seems basically right to me: 1⁄7 vs. 1⁄5 is more selective, but well within the margin of error given how much uncertainty I have.
I think the OP’s summary of the industry benchmark exercise is extremely misleading
Thanks – I just cut that sentence since my inability to communicate my view even with our substantial back-and-forth makes me pessimistic about making a summary.
- ^
In general, I would guess that CEA’s applicants have substantially less experience than their for-profit counterparts, as EA is quite young, but our applicants are more impressive given their age. E.g. we get a lot of college student applicants, but those students are going to prestigious universities.
Some thoughts:
Entrepreneurship and strategy:
I think lean startup seems better than everything there.
If you’re looking for something in particular drop, maybe good strategy, bad strategy?
Change of heart: I don’t think this has survived the replication crisis well, I would drop it.
Why We Love Dogs, Eat Pigs, and Wear Cows: I didn’t think this was a very epistemically rigorous book, I would drop it
Sorry for my slow response here, I missed the notification about your comment.
If EOIs are hard to get, that seems relevant to the question of whether EA jobs are hard to get since EOIs are quite sought after
I think maybe we just didn’t explain what EOIs are well. As an example: we had a product manager EOI; once we opened a full hiring round for PMs we contacted all the people who filled out the EOI and said “hey are you still looking for a PM position” and then moved the ones who said “yes” into the p.m. hiring round.[1]
I’m not sure why Ben thinks hires as a “percent of applicants who get to the people ops interview stage” (the only stage where CEA is more likely to hire, and not an apples-to-apples comparison since CEA has a work trial before it and Ashby doesn’t) is the right metric
My conclusion was: “in some ways CEA is more selective, and in other ways we are less; I think the methodology we used isn’t precise enough to make a stronger statement than ‘we are about the same.’”
I don’t think one of these comparison points is the “right metric” – they all have varying degrees of usefulness, and you and I might disagree a bit about their relative value, but, given their contradictory conclusions, I don’t think you can draw strong conclusions other than “we are about the same”.
- ^
Sometimes exceptional candidates are hired straight from an EOI, the example I give is specific to that role. I think in retrospect we should have just left EOIs off, as the data was more confusing than helpful.
- ^
The thing I have most changed my mind about since writing the post of mine you cite is adjacent to the “disvalue through evolution” category: basically, I’ve become more worried that disvalue is instrumentally useful. E.g. maybe the most efficient paperclip maximizer is one that’s really sad about the lack of paperclips.
There’s some old writing on this by Carl Shulman and Brian Tomasik; I would be excited for someone to do a more thorough write up/literature review for the red teaming contest (or just in general).
My anecdata is also that most people have thought about it somewhat, and “maybe it’s okay if everyone dies” is one of the more common initial responses I’ve heard to existential risk.
But I agree with OP that I more regularly hear “people are worried about negative outcomes just because they themselves are depressed” than “people assume positive outcomes just because they themselves are manic” (or some other cognitive bias).
We currently employ over 60 staff and we expect that to grow to 110 in 2022.
The “we” in the sentence is CEA, not the ops team, right? I might suggest clarifying that in the post, as the rest of the post uses “we” to refer to the ops team.
Thanks! I spend a fair amount of time reading technical papers, including the things you mentioned, mostly because I spend a lot of time on airplanes and this is a vaguely productive thing I can do on an airplane, but honestly this just mostly results in me being better able to make TikToks about obscure theorems.
Maybe my confusion is: when you say “participate in object level discussions” you mean less “be able to find the flaw in the proof of some theorem” and more “be able to state what’s holding us back from having more/better theorems”? That seems more compelling to me.
Cool, yeah that split makes sense to me. I had originally assumed that “talking to people about models of whether ELK helps anything” would fall into a “community building track,” but upon rereading your post more closely I don’t think that was the intended interpretation.[1]
FWIW the “only one track” model doesn’t perfectly map to my intuition here. E.g. the founders of doordash spent time using their own app as delivery drivers, and that experience was probably quite useful for them, but I still think it’s fair to describe them as being on the “create a delivery app” track rather than the “be a delivery driver” track.
I read you as making an analogous suggestion for EA community builders, and I would describe that as being “super customer focused” or something, rather than having only one “track”.
- ^
You say “obsessing over the details of what’s needed in direct work,” and talking to experts definitely seems like an activity that falls in that category.
- ^
I would find it helpful to have more precision about what it means to “participate more in object level discussion”.
For example: did you think that I/the forum was more impactful after I spent a week doing ELK? If the answer is “no,” is that because I need to be at the level of winning an ELK prize to see returns in my community building work? Or is it about the amount of time spent rather than my skill level (e.g. I would need to have spent a month rather than a week in order to see a return)?
Minor technical comment: the links to subsections in the topmost table link to the Google docs version of the article, and I think it would be slightly nicer if they linked to the forum post version.
Thanks for posting this! I thought the points were interesting, and I would have missed the conversation since I’m not very active on twitter.
Thanks! It’s a new feature we are piloting – authors can turn it on in their settings:
We will have an announcement post about it coming out soon
Thanks! Lesswrong is currently experimenting with multidimensional voting; if you haven’t already, I would suggest trying that out and giving them feedback.
This is cool, thanks for doing this