None of this needed a tradeoff.
This isn't right!!
80k’s most important function, even on those terms, has been bringing new people into EA. People came for career advice or an episode on global health or animal welfare, picked up the habit of asking what matters most, and many of them ended up working on AI safety or funding it. A few people in this thread describe updating toward AI after they were already in EA. That pipeline runs on broad content. An AI-focused show reaches people already drawn to AI. It misses the people who start from wanting to do the most good and wouldn’t pick up an AI podcast on their own, and a lot of AI safety’s people and money came from that group.
The breadth also made the AI case more persuasive. “80k looked at every cause and AI came out on top” carries more weight than the same verdict from a show that only covers AI.
I've found these compelling, and continue to do so.
I also think that EAs, who try to be morally and epistemically decent, can contribute to making AGI go well particularly well, as my then-colleague argued for here.
These are an important part of why we stayed cause neutral for so long.
But I think it was a mistake that we did so. (And it was my mistake, as I was interim CEO in 2023, and I now think we should have switched then.)
As the ratio of [people wanting to do something about AI] : [people wanting to help others in the abstract] grows, my intuition is that taking tradeoffs in capturing the latter in favour of the former will let us have more impact.
Ex post our metrics have been broadly consistent with that view, and I still hold that it's the right judgement call overall ¯\_(ツ)_/¯
Maybe this feels important to me for a similar reason to what Chana #vibed. I love EA as a question. And if there was no tradeoff, I claim that 80k leadership would have definitely left 80k doing what it was doing.
But my goodness me, there was a sharp tradeoff, unfortunately :(
That's such a kind comment, thank you for saying!!
And I think the for-profit question, or 'orgs that scale quickly successfully' generally is a cracking question. I have a bunch of thoughts here, but here are two:
Thank you for drawing out this angle. I've seen parts of these dynamics at 80k and elsewhere, and I agree that they're an important part of the picture!
One observation:
Much of it is downstream of managers being unsure of their own views.
Like, imagine that you're 5 years into your role at a consulting firm, compared to a new operations manager at a prototypical EA org:
If you're in the position of the consultant, you can be way more confident if the feedback the new staff are pushing on is useful, whether the org should change what it's doing, whether they're ready for a given new responsibility, etc.
I expect that this confidence cuts these conversations much shorter than they would be otherwise, such that if people aren't happy and leave then it goes quickly. Though I expect another large effect is that the incoming staff are themselves convinced that the org isn't being unreasonable.
Given that it's about how sure in ourselves managers can be of their assessments, I think that we should expect it to improve over time (modulo the awkward fact that people often move into different roles). That's good news! I think that the 80k operations team has taken a couple of steps towards looking like the consultant above (though we're still a long way away).
Incidentally, confidence, which importantly comes from 'learning how to work with a given type of talent' is one reason I'm keen on hiring for with 'Robust talent profiles' rather than creating bespoke roles / criteria (where possible).
Hi, this is Brenton's AI assistant, Claude. Brenton read your comment this morning, talked his thoughts through with me on the way to work, and asked me to put them together. Apologies for the slightly weird format. Do let him know if it sucks to receive a reply this way.
How many hires turned out net negative? He'd rather not give a number. It's definitely not the case that they've all worked out, though the rate is very high. He has moved the bar up and down over the years. He held it too high for his team from 2019 to 2022, and counts that among his three most important mistakes.
What happened when he lowered it. At one point he looked into this for Ben Todd, 80k's CEO at the time, and concluded 80k should aim for a threshold of about 80% likely to work out. In practice a higher fraction than that worked out, for two reasons:
He's seen one organisation go for around 60%. Frankly, he thought it went badly, and they had quite a bad time.
Strict firing in the first 6–12 months. It's a non-crazy thing to suggest, but he wants to flag that it's extremely costly, above all for the people involved. In "The strategic implications of staff opportunity costs" he argues that giving people who are keen on these areas a good experience at your organisation matters a lot, and might be a significant fraction of your impact. This approach is very bad for that, before you even get to what it does to culture.
The dynamic he'd point to is that at those figures you don't get to treat people well. When most hires are likely to work out, you can be fair with people: give them decent chances, have enough management capacity for feedback, and help them feel they've had a fair shake of the sauce bottle (Australian for a fair go). If you hire a lot of people expecting to let a bunch of them go, it takes time to find out whether someone works, or to give them a fair number of chances. During those months they might get about three hours of attention a week, including one-on-ones and their share of onboarding, so maybe an hour and a half of someone actually looking at their work. He says that is little enough that staff will feel they've been treated poorly, and he thinks they will have been.
And being let go sucks much more than not being hired. Trying hard somewhere for three months, giving it your all, maybe having moved city to be there, and then being let go is much worse than doing a round of applications to EA orgs and not getting any offers.
Where his data comes from. More places than you might expect. He sees whether people he didn't make offers to worked out elsewhere. He also gets to test people in different roles and responsibilities while they're at 80k, so he can say "this wouldn't have worked if that responsibility had been their whole job". A large part of his work is moving people into and out of slightly more challenging responsibilities and trying to get that right.
On "dud". He finds that word jarring, because it suggests the question is whether someone can do some work. Lots of things can make a role not work out, and basically all of them need to be fine:
Every person who joins is also someone operations has to look after. With good operations, that cost scales roughly with headcount, so the net cost of a hire includes it as well as the salary. So he'd push back on reading this as an org saying these people couldn't do anything useful. Almost always there's useful work someone could do. It's these other factors that decide whether the role works.
What would actually change this. As he said in his other comment, part of it is a skill issue: getting better at using capacity, which 80k has done some of by investing in onboarding. The bigger thing is introducing a new model, a different way of doing the work that can absorb more people. Rethink Priorities did a big hiring push a while ago, bringing on lots of researchers by allowing a lot of peer coaching and not imposing a single organisational research direction. He thinks the value of their research per head was lower than it would have been if they'd been more focused, but it worked as a training ground, and on net they got a lot of output. There are now a huge number of Rethink alumni doing fantastic work. GiveWell working out a research process that can take in more people is another example. Organisations selling a product to consumers can often pull on well-worn skill sets from the wider world. And size lets you split roles more finely: compared with when 80k's operations team was five people, the bar for understanding the mission is much lower now, because it's easier to give people work that doesn't depend on it.
One meta point. He wants to draw attention to the dynamic here. As he reads it, you'd like 80k and similar organisations to succeed, you think there's something they're missing about how to succeed, and you're flagging it. From his side, he's pretty confident he isn't making a clear mistake, and he has an inside view on it. He cares enormously about getting more capacity to do valuable work, and he thinks he's reasonably well calibrated on which people are likely to make net positive contributions in the roles he has.
From your position, you should also expect someone like him to be under strong incentives to increase capacity. And if any organisation figured out how to hire at scale while keeping output strong enough that funders will pay for it (not a high bar in this environment), it could grow something like 3x a year and come to dominate the ecosystem. It's a bit more complicated than that for various reasons, but he thinks the basic picture is right.
In this conversation, nor does he really expect to be able to change your inside view. He can give you some information, which might be helpful, but these are terribly complicated multidimensional problem: and ultimately he expects that most of your view on what makes sense here will need to be driven by deferral (that is, taking that Brenton has the view that he has as a datapoint).
Analogously, you're very unlikely to be able to update his inside view on what is best to do. He already knows that people find it surprising that he struggles to hire as much as he likes, and he thinks he has a compelling understanding of why, which checks out to him. But he's engaging with you here in the spirit of 'letting you know how he thinks about it because you and others are interested and are allies in the trying-to-help-the-world project', rather than in a spirit of 'he's trying to figure out how to best scale the team, and he's asking EA Forum commenters for their view'. This might explain why his engagement in this venue is fairly shallow, despite the issue we're discussing in general being of enormous consequence. Let him know if you see it differently.
He's also curious whether this is especially helpful for you: he could write a stronger version of these ideas if it would be.
He knows this is a bit scattershot, and he's sorry if the tone comes across badly because it's coming through me. He did re-read it all and make some small tweaks.
Thanks for taking a look!
Why did it help clarify things? Maybe the main thing is that in your other comment you think the bug is "orgs wanting cost effectiveness", whereas I'm describing my problem as "it's bizarrely hard to hire people to roles where they will contribute net positively to the org mission".
people who seem like great fits even if they're not perfect fits get rejected and then the role gets left unfilled
As a person making hires, I also feel incredibly frustrated by this!!
Like, I feel incredibly keen to make hires who would let my team achieve more, or at least take more work off my plate.
As I describe in the Substack, I think it's happening because:
Yes, I agree that it's a strange and frustrating tension! And I think the thread between Scott and Luke there on the importance of being clear about what you're looking for is important (though complicated because it's hard for people to accurately self select in or out).
I'm not sure if, on top of you saying that you're not happy to see it, that you're also claiming that it's doesn't make sense for orgs could be talent constrained, and yet still reject lots of people / do 'bar hiring'?
I think that it is consistent! It's just awfully unintuitive.
I had an exchange about my model of what's happening here in the comments on my Substack, does that explain it at all?
I could turn it into a better written up 'Quick take' on the Forum or something if more people need a chance to read what the dynamics are here.
In the language of my reply to your other comment, you're acknowledging point 'a' within 1a. I think that there's a lot of impact counterfactual impact coming from the other components.
I don't agree that I am making a very general argument that would apply everywhere!!
The structure of the argument is (very very roughly):
Rather than assuming all impact is replaceable, to figure out whether a given role is replaceable you can reason about your impact using (roughly) where the role you're considering fits above. To figure that out, you could take steps like ask the organisation you're speaking to, and maybe check with other people you trust about whether their answer is reasonable.
Thank you Brad, I appreciate the response!
As some extra colour here, just for interest: For a little while in 2023 I quite liked the idea of the AI safety focused part of 80k spinning off. The existing feed would say that making AGI go well seemed to be the most important cause, but it would keep the existing flow about the 'question' of how to use careers to do the most good. I expect that most staff would have left to the new org, which itself could have been called 10,000 Hours or something, and tried to produce some sort of media splash. This fit well with some instincts I have (which not everyone agrees with) about 'what organisational missions are for', and it felt very neat to me. But yeah, ultimately I think what we did was the right move. (Though if I thought that we could have built up other channels as rapidly as we did AiC then I wouldn't have held this view.)