My hunch is (as implied elsewhere) âtalent-constraintâ with âtalentâ not further specified is apt to mislead. My impression for longtermist orgs (I understand from Peter and others this may apply less to orgs without this as the predominant focus) is there are two broad classes, which imperfectly line up with âseniorâ versus âjuniorâ.
The âseniorâ class probably does fit (commensensically understood) âtalent-constraintâ, in that orgs or the wider ecosystem want to take everyone who clears a given bar. Yet these bars are high even when conditioned on the already able cohort of (longtermist/â)EAs. It might be things like âready to run a research groupâ, âcan manage operations for an orgâ (cf. Taraâs and Tanyaâs podcasts), âsubject matter expertise/âability/âtrack recordâ.
One common feature is that these people add little further load on current (limited) management capacity, either because they are managing others or are already âup to speedâ to contribute themselves without extensive training or supervision. (Aside: I suspect this is a under-emphasised bonus of âvalue-aligned operations staffâ - their tacit knowledge of the community/âmission/âwider ecosystem may permit looser management than bringing on able professionals âfrom outsideâ.) From the perspective of the archetypal âpluripotent EAâ a few years out from undergrad, these are skills which are hard to develop and harder to demonstrate.
More âjuniorâ roles are those where the criteria are broader (at least in terms of legible ones: âwhat it takesâ to be a good generalist researcher may be similarly rare to âwhat it takesâ to be a good technical AI safety researcher, but more can easily ârule themselves outâ of the latter than the former), where âupskillingâ is a major objective, or where thereâs expectation of extensive âhands-onâ management.
There might be similarly convex returns to getting a slightly better top candidate (e.g. âexcellent versus very goodâ might be 3x rather than 1.3x). Regardless, there will not be enough positions for all the talented candidates available: even if someone at an org decided to spend their time only managing and training junior staff (and haste considerations might lead them to spending more of their time doing work themselves than investing in the ânext generationâ), they canât manage dozens at a time.
I think confusing these two broad classes is an easy way of burning a lot of good people (cf. Deniseâs remarks). If Alice the 23-year-old management consultant might reason on current messaging, âEA jobs are much better for the world than management consultancy, and theyâre after good peopleâI seem to fit the bill, so I should switch career into thisâ. She might then forsake her promising early career for an unedifying and unsuccessful period as âEA perennial applicantâ, ending up worse than she was at the start. EA has a vocational quality to itâkey it does not become a siren song.
There seem a few ways to do this better, as alluded to in prior discussions here and elsewhere:
0) If Iâm right, itâd be worth communicating the âperson specâ for cases where (common-sense) talent constraint applies, and where we really would absorb basically as many as we could get (e.g. âWe want philosophers to contribute to GPR, and weâre after people who either already have a publication record in this area, or have signals of âsuperstarâ ability even conditioned on philosophy academia. If this is you, please get in touch.â).
1) Concurrently, itâd be worth publicising typical applicants:place or similar measures of competition for hiring rounds in more junior roles to allow applicants to be better calibrated/âemphasise the importance of plan B. (e.g. âWe have early-career roles for people thinking of working as GPR researchers, which serves the purpose of talent identification and development. We generally look for XYZ. Applications for this are extremely competitive (~12:1). Other good first steps for people who want to work in this field are theseâ). {MIRIâs research fellows page does a lot of this well}.
2) It would be good for there to be further work addressed to avoiding âEA underemploymentâ, as I would guess growth in strong candidates for EA roles will outstrip intra-EA opportunities. Some possibilities:
2.1) There are some areas Iâd want to add to the longtermist portfolio which might be broadened into useful niches for people with comparative advantage in them (macrohistory, productivity coaching and nearby versions, EA-relevant bits of psychology, etc.) I donât think these are âeasierâ than the existing âhotâ areas, but they are hard in different ways, and so broaden opportunities.
2.2) Another option would be âpre-caching human capitalâ into areas which are plausible candidates for becoming important as time goes on. I imagine something like international relations turning out to be crucial (or, contrariwise, relatively unimportant), but it seems better rather than waiting for this to be figured out for instead people to coordinate and invest themselves across the portfolio of plausible candidates. (Easier said than done from the first person perspective, as such a strategy potentially involves making an uncertain bet with many years of oneâs career, and if it turns out to be a bust ex post the good ex ante EV may not be complete consolation).
2.3) There seem a lot of stakeholders where it would be good for EAs to enter due to the second-order benefits even if their direct work is of limited direct relevance (e.g. having more EAs in tech companies looks promising to me, even if they arenât doing AI safety). (Again, not easy from the first person-perspective).
2.4) A lot of skills for more âseniorâ roles can and have been attained outside of the EA community. Grad school is often a good idea for researchers, and professional/âmanagement aptitude is often a transferable skill. So some of the options above can be seen as a holding-pattern/âbet hedging approach: they hopefully make one a stronger applicant for such roles, but in the meanwhile one is doing useful things (and also potentially earning to give, although I think this should be a minor consideration for longtermist EAs given the field is increasingly flush with cash).
If the framing is changed to something like, âThese positions are very valuable, but very competitiveâit is definitely worth you applying (as you in expectation increase the quality of the appointed candidate, and the returns of a slightly better candidate are very high), donât bet the farm (or quit the day job) on your applicationâand if you donât get in, hereâs things you could do to slant your career to have a bigger impactâ, Iâd hope the burn risk falls dramatically: in many fields there are lots of competitive oversubscribed positions which donât impose huge costs to unsuccessful applicants.
My hunch is (as implied elsewhere) âtalent-constraintâ with âtalentâ not further specified is apt to mislead. My impression for longtermist orgs (I understand from Peter and others this may apply less to orgs without this as the predominant focus) is there are two broad classes, which imperfectly line up with âseniorâ versus âjuniorâ.
The âseniorâ class probably does fit (commensensically understood) âtalent-constraintâ, in that orgs or the wider ecosystem want to take everyone who clears a given bar. Yet these bars are high even when conditioned on the already able cohort of (longtermist/â)EAs. It might be things like âready to run a research groupâ, âcan manage operations for an orgâ (cf. Taraâs and Tanyaâs podcasts), âsubject matter expertise/âability/âtrack recordâ.
One common feature is that these people add little further load on current (limited) management capacity, either because they are managing others or are already âup to speedâ to contribute themselves without extensive training or supervision. (Aside: I suspect this is a under-emphasised bonus of âvalue-aligned operations staffâ - their tacit knowledge of the community/âmission/âwider ecosystem may permit looser management than bringing on able professionals âfrom outsideâ.) From the perspective of the archetypal âpluripotent EAâ a few years out from undergrad, these are skills which are hard to develop and harder to demonstrate.
More âjuniorâ roles are those where the criteria are broader (at least in terms of legible ones: âwhat it takesâ to be a good generalist researcher may be similarly rare to âwhat it takesâ to be a good technical AI safety researcher, but more can easily ârule themselves outâ of the latter than the former), where âupskillingâ is a major objective, or where thereâs expectation of extensive âhands-onâ management.
There might be similarly convex returns to getting a slightly better top candidate (e.g. âexcellent versus very goodâ might be 3x rather than 1.3x). Regardless, there will not be enough positions for all the talented candidates available: even if someone at an org decided to spend their time only managing and training junior staff (and haste considerations might lead them to spending more of their time doing work themselves than investing in the ânext generationâ), they canât manage dozens at a time.
I think confusing these two broad classes is an easy way of burning a lot of good people (cf. Deniseâs remarks). If Alice the 23-year-old management consultant might reason on current messaging, âEA jobs are much better for the world than management consultancy, and theyâre after good peopleâI seem to fit the bill, so I should switch career into thisâ. She might then forsake her promising early career for an unedifying and unsuccessful period as âEA perennial applicantâ, ending up worse than she was at the start. EA has a vocational quality to itâkey it does not become a siren song.
There seem a few ways to do this better, as alluded to in prior discussions here and elsewhere:
0) If Iâm right, itâd be worth communicating the âperson specâ for cases where (common-sense) talent constraint applies, and where we really would absorb basically as many as we could get (e.g. âWe want philosophers to contribute to GPR, and weâre after people who either already have a publication record in this area, or have signals of âsuperstarâ ability even conditioned on philosophy academia. If this is you, please get in touch.â).
1) Concurrently, itâd be worth publicising typical applicants:place or similar measures of competition for hiring rounds in more junior roles to allow applicants to be better calibrated/âemphasise the importance of plan B. (e.g. âWe have early-career roles for people thinking of working as GPR researchers, which serves the purpose of talent identification and development. We generally look for XYZ. Applications for this are extremely competitive (~12:1). Other good first steps for people who want to work in this field are theseâ). {MIRIâs research fellows page does a lot of this well}.
2) It would be good for there to be further work addressed to avoiding âEA underemploymentâ, as I would guess growth in strong candidates for EA roles will outstrip intra-EA opportunities. Some possibilities:
2.1) There are some areas Iâd want to add to the longtermist portfolio which might be broadened into useful niches for people with comparative advantage in them (macrohistory, productivity coaching and nearby versions, EA-relevant bits of psychology, etc.) I donât think these are âeasierâ than the existing âhotâ areas, but they are hard in different ways, and so broaden opportunities.
2.2) Another option would be âpre-caching human capitalâ into areas which are plausible candidates for becoming important as time goes on. I imagine something like international relations turning out to be crucial (or, contrariwise, relatively unimportant), but it seems better rather than waiting for this to be figured out for instead people to coordinate and invest themselves across the portfolio of plausible candidates. (Easier said than done from the first person perspective, as such a strategy potentially involves making an uncertain bet with many years of oneâs career, and if it turns out to be a bust ex post the good ex ante EV may not be complete consolation).
2.3) There seem a lot of stakeholders where it would be good for EAs to enter due to the second-order benefits even if their direct work is of limited direct relevance (e.g. having more EAs in tech companies looks promising to me, even if they arenât doing AI safety). (Again, not easy from the first person-perspective).
2.4) A lot of skills for more âseniorâ roles can and have been attained outside of the EA community. Grad school is often a good idea for researchers, and professional/âmanagement aptitude is often a transferable skill. So some of the options above can be seen as a holding-pattern/âbet hedging approach: they hopefully make one a stronger applicant for such roles, but in the meanwhile one is doing useful things (and also potentially earning to give, although I think this should be a minor consideration for longtermist EAs given the field is increasingly flush with cash).
If the framing is changed to something like, âThese positions are very valuable, but very competitiveâit is definitely worth you applying (as you in expectation increase the quality of the appointed candidate, and the returns of a slightly better candidate are very high), donât bet the farm (or quit the day job) on your applicationâand if you donât get in, hereâs things you could do to slant your career to have a bigger impactâ, Iâd hope the burn risk falls dramatically: in many fields there are lots of competitive oversubscribed positions which donât impose huge costs to unsuccessful applicants.