This seems like an extremely easy avenue for improvement.
First, if you think this because you have experience recruiting people for projects like this (i.e. it “seems easy” because you have an affordance for it, not because you don’t know what the obstacles are), please reach out to someone at MIRI (probably Malo, maybe Nate), and also PM me so I can bug them until they interview you.
As for why it hasn’t happened yet:
I no longer work for MIRI, and so am not speaking for current-MIRI, but when I did work at MIRI I was one of the three sort-of-responsible-for-recruitment employees. (Buck Shlegeris was the main recruiter, and primarily responsible for engineering hires; I was in charge of maintaining the agent foundations pipeline and later trying to hire someone for the Machine Learning Living Library role, tho I think I mostly added value by staffing AIRCS workshops with Buck; Colm O’Riain took over maintaining that pipeline, among doing other things. Several MIRI board members were involved with hiring efforts, but I’m not quite sure how to compare that.)
I think the primary reason this got half-hearted investment was that we didn’t have strong management capacity, and so wanted to invest primarily in people who could self-orient towards things that would be useful for existential risk. (If someone is at MIRI because MIRI is willing to pay them $1M/yr, they might focus on turning MIRI into the sort of place that is good at extracting money from OpenPhil/FTX instead of the sort of place that is good at decreasing the probability of x-risk, which seems like a betrayal of OpenPhil/FTX/the whole Earth. One could imagine instead having an org that was very good at figuring out how to make use of mercenaries, but that org was not MIRI.)
For the MLLL specifically, we had the problem that we were basically offering to pay someone to skill up into being hirable for a very high-paid ML position somewhere else, and so part of the interview was “can you gain the skill?” and the other part of the interview was “do we expect that you won’t just leave?”. I couldn’t figure out how to do this successfully, and we didn’t end up hiring anyone for that role (altho we did trial more than one person).
I think the secondary reason was that we didn’t have people who were great recruiters, and so we didn’t really get into a hiring spiral. [Like, I think it was basically luck that MIRI managed to hire Colm in the first place, and I’m not sure how we would have deliberately found a recruiter that got us more of the candidates we wanted.] I think Buck was our best person for that sort of thing—extroverted, good at thinking about AI safety issues, and good at interviewing people to figure out their level of technical skill—but also Buck is good at doing direct work, and so I think didn’t put as much time into it as a full-time recruiter would have, and maybe being slow on that sort of thing lost us some candidates. [And, in retrospect, I’m really not sure this was the wrong call—I think we spent something like one Buck-year to get something like one Buck-year back in terms of other employees working for us.]
As an example of an element of what a recruiting spiral might have looked like (which I’m summarizing based off memory, since I can’t review the relevant emails anymore): at one point Tanya Singh read Superintelligence, emailed MIRI interested in doing something about AI alignment, ended up in the “idk maybe” part of the application inbox, I scanned thru it and said “whoa this person looks actually exciting, let’s try to get them a job at FHI or something”, Malo responded with “idk why you’re so excited, but this is the right contact person at FHI”, and then after the introduction it ended up working out. (I’m not sure how much Tanya ended up increasing their recruitment capacity, but if she did then some of those hires are downstream of us doing something with our inbox, which was downstream of me reading it, which was downstream of efforts to hire me.)
[Actually, I wasn’t around for this part, but I remember hearing that in the very early days MIRI got some mileage out of getting someone else to read Eliezer’s inbox, because there were a bunch of quite valuable messages buried in the big pile of all the emails. But it mattered that it was someone with lots of context like Michael Vassar reading the emails, instead of some random secretary.]
I think a tertiary reason is that the funding growth ramped up relatively late in MIRI’s history. When I started working for them in 2016, the plan for paying employees was “you tell us how much you need for money to not be an issue, and we’ll decide whether or not to pay you that”, which made sense for a non-profit that had some money but had to work for it. [I told them the “this is the number that would have my net cash flow be $1k/mo, and this is the number that would match my net income as a data scientist in Austin after correcting for Bay Area cost of living”, and the latter number was 3X the former number; you can guess which they wanted to pay me ;).] When MIRI started hiring engineers (2018?), funding had reached the point that it was easy to offer people high-but-not-fully-competitive salaries (like, I think engineers were paid $100-200k, which was probably ~60% of what they could have gotten at Facebook or w/e?), and I think now funding has reached the point where it’s reasonable to just drop big piles of money on people doing anything at all useful, and announce things like the Visible Thoughts Project. [As another example, compare ARC’s approach to funding the ELK prize in 2022 to Paul’s approach to a similar thing in 2018.]
if you think this because you have experience recruiting people for projects like this (i.e. it “seems easy” because you have an affordance for it, not because you don’t know what the obstacles are), please reach out to someone at MIRI (probably Malo, maybe Nate), and also PM me so I can bug them until they interview you.
I don’t have any experience with direct recruiting (and would expect myself to be rather bad at it tbh), but I do have significant experience with PR, particularly media relations. If an EA-aligned organization could use more people on that, and would be willing to hire someone who can’t work full-time for medical reasons, let me know.
Could someone from MIRI step in here to explain why this is not being done? This seems like an extremely easy avenue for improvement.
First, if you think this because you have experience recruiting people for projects like this (i.e. it “seems easy” because you have an affordance for it, not because you don’t know what the obstacles are), please reach out to someone at MIRI (probably Malo, maybe Nate), and also PM me so I can bug them until they interview you.
As for why it hasn’t happened yet:
I no longer work for MIRI, and so am not speaking for current-MIRI, but when I did work at MIRI I was one of the three sort-of-responsible-for-recruitment employees. (Buck Shlegeris was the main recruiter, and primarily responsible for engineering hires; I was in charge of maintaining the agent foundations pipeline and later trying to hire someone for the Machine Learning Living Library role, tho I think I mostly added value by staffing AIRCS workshops with Buck; Colm O’Riain took over maintaining that pipeline, among doing other things. Several MIRI board members were involved with hiring efforts, but I’m not quite sure how to compare that.)
I think the primary reason this got half-hearted investment was that we didn’t have strong management capacity, and so wanted to invest primarily in people who could self-orient towards things that would be useful for existential risk. (If someone is at MIRI because MIRI is willing to pay them $1M/yr, they might focus on turning MIRI into the sort of place that is good at extracting money from OpenPhil/FTX instead of the sort of place that is good at decreasing the probability of x-risk, which seems like a betrayal of OpenPhil/FTX/the whole Earth. One could imagine instead having an org that was very good at figuring out how to make use of mercenaries, but that org was not MIRI.)
For the MLLL specifically, we had the problem that we were basically offering to pay someone to skill up into being hirable for a very high-paid ML position somewhere else, and so part of the interview was “can you gain the skill?” and the other part of the interview was “do we expect that you won’t just leave?”. I couldn’t figure out how to do this successfully, and we didn’t end up hiring anyone for that role (altho we did trial more than one person).
I think the secondary reason was that we didn’t have people who were great recruiters, and so we didn’t really get into a hiring spiral. [Like, I think it was basically luck that MIRI managed to hire Colm in the first place, and I’m not sure how we would have deliberately found a recruiter that got us more of the candidates we wanted.] I think Buck was our best person for that sort of thing—extroverted, good at thinking about AI safety issues, and good at interviewing people to figure out their level of technical skill—but also Buck is good at doing direct work, and so I think didn’t put as much time into it as a full-time recruiter would have, and maybe being slow on that sort of thing lost us some candidates. [And, in retrospect, I’m really not sure this was the wrong call—I think we spent something like one Buck-year to get something like one Buck-year back in terms of other employees working for us.]
As an example of an element of what a recruiting spiral might have looked like (which I’m summarizing based off memory, since I can’t review the relevant emails anymore): at one point Tanya Singh read Superintelligence, emailed MIRI interested in doing something about AI alignment, ended up in the “idk maybe” part of the application inbox, I scanned thru it and said “whoa this person looks actually exciting, let’s try to get them a job at FHI or something”, Malo responded with “idk why you’re so excited, but this is the right contact person at FHI”, and then after the introduction it ended up working out. (I’m not sure how much Tanya ended up increasing their recruitment capacity, but if she did then some of those hires are downstream of us doing something with our inbox, which was downstream of me reading it, which was downstream of efforts to hire me.)
[Actually, I wasn’t around for this part, but I remember hearing that in the very early days MIRI got some mileage out of getting someone else to read Eliezer’s inbox, because there were a bunch of quite valuable messages buried in the big pile of all the emails. But it mattered that it was someone with lots of context like Michael Vassar reading the emails, instead of some random secretary.]
I think a tertiary reason is that the funding growth ramped up relatively late in MIRI’s history. When I started working for them in 2016, the plan for paying employees was “you tell us how much you need for money to not be an issue, and we’ll decide whether or not to pay you that”, which made sense for a non-profit that had some money but had to work for it. [I told them the “this is the number that would have my net cash flow be $1k/mo, and this is the number that would match my net income as a data scientist in Austin after correcting for Bay Area cost of living”, and the latter number was 3X the former number; you can guess which they wanted to pay me ;).] When MIRI started hiring engineers (2018?), funding had reached the point that it was easy to offer people high-but-not-fully-competitive salaries (like, I think engineers were paid $100-200k, which was probably ~60% of what they could have gotten at Facebook or w/e?), and I think now funding has reached the point where it’s reasonable to just drop big piles of money on people doing anything at all useful, and announce things like the Visible Thoughts Project. [As another example, compare ARC’s approach to funding the ELK prize in 2022 to Paul’s approach to a similar thing in 2018.]
I don’t have any experience with direct recruiting (and would expect myself to be rather bad at it tbh), but I do have significant experience with PR, particularly media relations. If an EA-aligned organization could use more people on that, and would be willing to hire someone who can’t work full-time for medical reasons, let me know.