Cullen_OKeefe

• Cullen_OKeefe 15 Jan 2020 5:26 UTC

  1 point

  in reply to: evelynciara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I’d imag­ine there’s an au­di­ence for it!

• Cullen_OKeefe 15 Jan 2020 5:25 UTC

  3 points

  in reply to: Wei_Dai's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Thanks Wei! This is a very thought­ful com­ment.

  I com­pletely agree that we should be wary of those as­pects of SJ as well. I’m not sure that I’m “less” wor­ried about it than you; I do worry about it. How­ever, I have not seen much of this be­hav­ior in the EA com­mu­nity so I am not im­me­di­ately wor­ried and have some rea­sons to be fairly op­ti­mistic in the long run:

  1. Founder effects and strong com­mu­nal norms to­wards open dis­cus­sion in the EA com­mu­nity to which I think most new­com­ers get pretty heav­ily in­cul­cated.

  2. Cause pri­ori­ti­za­tion and con­se­quen­tial­ism are some­what in­con­gru­ous with these things, since many of the things that can get peo­ple to be un­fairly “can­celed” are quite small from an EA per­spec­tive.

  3. Heavy in­fluence of and con­nec­tion to philos­o­phy se­lects for open­ness norms as well.

  4. Abil­ity and mo­ti­va­tion to se­lec­tively adopt the best SJ po­si­tions with­out adopt­ing some of its most harm­ful prac­tices.

  To restate, I would definitely be pretty wary of any at­tempt to re­form EA in a way that se­ri­ously en­dan­gered norms of ci­vil­ity, open de­bate, in­tel­lec­tual in­quiry, etc. as they cur­rently are prac­ticed. I ac­tu­ally think we do a very good job as a move­ment of bal­anc­ing these goals. This is part of why I cur­rently spend more time in EA than SJ.

• Cullen_OKeefe 13 Jan 2020 4:49 UTC

  13 points

  in reply to: jtm's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Beyond move­ment build­ing & in­clu­sivity, I think it makes sense for EA as a move­ment to keep their cur­rent ap­proach be­cause it’s been work­ing pretty well IMO.

  I think the thing EAs as peo­ple (with a wor­ld­view that in­cludes things be­yond EA) might want to con­sider—and which SJ could in­form—is the de­mands that his­tor­i­cal in­jus­tices of, e.g., colo­nial­ism, racism, etc. make on us. I think those de­mands are plau­si­bly quite large and failure to satisfy them could con­sti­tute a on­go­ing moral catas­tro­phe. Since they’re not welfarist, they’re out­side the scope of EA as it cur­rently ex­ists. But for moral un­cer­tainty rea­sons I think many peo­ple should think about them.

• Cullen_OKeefe 13 Jan 2020 4:47 UTC

  2 points

  in reply to: evelynciara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I don’t! Would be in­ter­est­ing to see! From an EA per­spec­tive, though, flowthrough effects on long-term stuff might dom­i­nate the con­sid­er­a­tions.

• Cullen_OKeefe 13 Jan 2020 4:47 UTC

  2 points

  in reply to: Aidan O'Gara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Hard to imag­ine it ever be­ing too much TBH. I and most of my col­leagues con­tinue to in­vest in AI up­skil­ling. How­ever, lots of other skills are worth hav­ing too. Ba­si­cally, I view it as a pro­cess of con­tinual im­prove­ment: I will prob­a­bly never have “enough” ML skill be­cause the field moves faster than I can keep up with it, and there are ap­prox­i­mately lin­ear re­turns on it (and a bunch of other skills that I’ve men­tioned in these com­ments).

• Cullen_OKeefe 13 Jan 2020 4:45 UTC

  2 points

  in reply to: Aidan O'Gara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I would lean pretty heav­ily to­wards ML. Tak­ing an in­tro to CS class is good back­ground, but spe­cial­ize other than that. Some ad­ja­cent ar­eas, like cy­ber­se­cu­rity, are good too.

  (You could help AI de­vel­op­ment with­out spe­cial­iz­ing in AI, but this is speci­fi­cally for AI Policy ca­reers.)

• Cullen_OKeefe 13 Jan 2020 4:41 UTC

  9 points

  in reply to: Aidan O'Gara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Yeah, I think I’m pretty bullish on JDs more than the av­er­age EA be­cause it’s very use­ful for a ton of ca­reers. Like, a JD is an as­set for pretty much any ca­reer in gov­ern­ment, where you can work on a lot of EA prob­lems, like:

  • Risks from WMDs or pan­demics (or gen­eral emer­gency pre­pared­ness)

  • An­i­mal welfare

  • In­ter­na­tional co­op­er­a­tion and coordination

  • Space law and policy

  • En­vi­ron­men­tal policy

  • Land use regulation

  • Gen­eral eco­nomic growth

  • Im­prov­ing vot­ing and in­sti­tu­tion design

  (Of course, lawyers can use­fully work on these out­side of gov­ern­ment as well.)

  I think EA-rele­vant skills in eco­nomics might be par­tic­u­larly valuable in some fields, like gov­ern­men­tal cost-benefit anal­y­sis.

  Of course, peo­ple in gov­ern­ment can have a lot of im­pact on prob­lems that most EAs don’t work on due to the amount of in­fluence they have.

  I also think that there might be op­por­tu­ni­ties for lawyers to help grow/​struc­ture/​im­prove the EA move­ment, like:

  • Es­tate plan­ning (e.g., help ev­ery EA who wants one get a will or trust that will give a lot of their es­tate to EA char­i­ties)

  • Set­ting up non­prof­its and other organizations

  • Tax help

  • Im­mi­gra­tion help for EA employers

  • Creat­ing weird en­tities or fi­nan­cial in­stru­ments that help EAs achieve their goals (e.g., 1, 2, 3)

  If I was not do­ing AI policy, I might write up a grant pro­posal to spend my time do­ing these things in CA.

• Cullen_OKeefe 12 Jan 2020 22:25 UTC

  1 point

  in reply to: Aidan O'Gara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Limit­ing the dis­cus­sion to the most im­pact­ful jobs from an EA per­spec­tive, I think it can be pretty hard for rea­sons I lay out here. I got lucky in many many ways, in­clud­ing that I was ac­cepted to 80K coach­ing, turned out to be good at this line of work (which I eas­ily could not have), and was in law school dur­ing the time when FHI was just spin­ning up its GovAI in­tern­ship pro­gram.

  My guess is that gen­eral cre­den­tials are prob­a­bly in­suffi­cient with­out ac­com­pa­ny­ing work that shows your abil­ity to ad­dress the very unique is­sues of AGI policy well. So op­por­tu­ni­ties to try your hand at that are pretty valuable if you can find them.

  That said, op­por­tu­ni­ties to show gen­eral AI policy ca­pa­bil­ities—even on “short-term” is­sues—are good sig­nals and can lead to a good ca­reer in this area!

• Cullen_OKeefe 12 Jan 2020 22:18 UTC

  13 points

  in reply to: evelynciara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I think there’s a lot of is­sues that have ap­pli­ca­bil­ity to both short- and long-term con­cerns, like:

  • In­ter­pretabil­ity/​transparency

  • Fore­sight/​Pre­dict­ing AI trends

  • Fu­ture of work/​eco­nomics of AI/​la­bor economics

  • Align­ing recom­mender systems

  • Disinformation

  • Cybersecurity

  • Au­toma­tion of gov­ern­men­tal systems

  Re­lat­edly, this is why peo­ple who can’t im­me­di­ately work for an EA-al­igned org on “long-term” AI is­sues can build both use­ful ca­reer cap­i­tal and do use­ful work by work­ing in more gen­eral AI policy.

  For the sec­ond ques­tion, a pretty bor­ing EA an­swer: I would like to see more peo­ple in near-term AI policy en­gage in ex­plicit and quan­tifi­able cause pri­ori­ti­za­tion for their work. I think that, as EAs gen­er­ally rec­og­nize, the im­pact be­tween these things prob­a­bly varies quite a lot. That should guide which ques­tions peo­ple work on.

  What links here?

  • What are the most press­ing is­sues in short-term AI policy? by evelynciara (14 Jan 2020 22:05 UTC; 7 points)
  • Cullen_OKeefe's comment on I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA by Cullen_OKeefe (12 Jan 2020 22:25 UTC; 1 point)

• Cullen_OKeefe 12 Jan 2020 22:09 UTC

  2 points

  in reply to: Stefan_Schubert's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  The bor­ing an­swer is that there’s a va­ri­ety of re­la­tion­ships that need to be man­aged well in or­der for AGI de­ploy­ment to go op­ti­mally. Com­par­a­tive ad­van­tage and op­por­tu­nity are prob­a­bly good in­di­ca­tors of where the most fruit­ful work for any given in­di­vi­d­ual is. That said, I think work­ing with in­dus­try can be pretty highly lev­er­aged since it’s more nim­ble and eas­ier to per­suade than gov­ern­ment IMO.

• Cullen_OKeefe 12 Jan 2020 22:05 UTC

  3 points

  in reply to: Aidan O'Gara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  ML knowl­edge is good and im­por­tant; I gen­er­ally wish I had more of it and use many of my Learn­ing Days to im­prove it. That link also shows some of the other, non-law sub­jects I’ve been study­ing.

  In law school, I stud­ied a lot of differ­ent sub­jects that have been use­ful, like:

  • Ad­minis­tra­tive law

  • Na­tional Se­cu­rity law

  • Con­sti­tu­tional law

  • Cor­po­rate law

  • Compliance

  • Con­tract law

  • Prop­erty law

  • Pa­tent law

  • In­ter­na­tional law

  • Negotiations

  • An­titrust law

  I am pretty bullish on most of the spe­cific stuff you men­tioned. I think macro­his­tory, his­tory of tech­nol­ogy, gen­eral pub­lic policy, fore­cast­ing, and eco­nomics are pretty use­ful. Un­for­tu­nately, it’s such a weird and idiosyn­cratic field that there’s not re­ally a one-size-fits-all cur­ricu­lum for get­ting into it, though this also means there’s many pro­duc­tive ways to spend one’s time prepar­ing for it.

• Cullen_OKeefe 12 Jan 2020 20:09 UTC

  14 points

  in reply to: Aidan O'Gara's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Lots of good stuff is in SF (OpenAI, PAI, Open Phil). How­ever, there are also very good op­tions for EAs in DC (CSET, US gov­ern­ment stuff) and UK (FHI, Deep­Mind, CSER, CFI).

  You can also build good ca­reer ex­pe­rience in gen­eral AI Policy work (i.e., not AGI- or LTF-fo­cused) in a pretty big num­ber of ar­eas, like Bos­ton (Berk­man-Klein, MIT, FLI) or NYC (AI Now). I don’t know of AI-spe­cific stuff in Chicago or LA, but of course they both have good uni­ver­si­ties where you could prob­a­bly do AI policy re­search.

  See also my replies to this com­ment. :-)

• Cullen_OKeefe 12 Jan 2020 4:39 UTC

  2 points

  in reply to: acrlr's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Thanks! See if this post helps an­swer that! If not, feel free to fol­low-up!

• Cullen_OKeefe 12 Jan 2020 4:12 UTC

  3 points

  in reply to: acrlr's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I also wanted to pass along this ex­pla­na­tion from our team man­ager, Jack Clark:

  OpenAI’s policy team looks for can­di­dates that dis­play an ‘idiosyn­cratic spe­cial­ism’ along with ver­ifi­able in­ter­est and in­tu­itions re­gard­ing tech­ni­cal as­pects of AI tech­nol­ogy; mem­bers of OpenAI’s team cur­rently have spe­cial­isms rang­ing from long-term TAI-ori­ented ethics, to geopoli­tics of com­pute, to is­sues of rep­re­sen­ta­tion in gen­er­a­tive mod­els, to ‘red team­ing’ tech­ni­cal sys­tems from a se­cu­rity per­spec­tive, and so on. OpenAI hires peo­ple with a mix­ture of qual­ifi­ca­tions, and is equally happy hiring some­one with no de­grees and ver­ifi­able in­dus­try ex­pe­rience, as well as some­one with a PHD. At OpenAI, tech­ni­cal fa­mil­iar­ity is a pre­req­ui­site for suc­cess­ful policy work, as our policy team does a lot of work that in­volves em­bed­ding alongside tech­ni­cal teams on pro­jects (see: our work through­out 2019 on GPT2).

  What links here?

  • Cullen_OKeefe's comment on I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA by Cullen_OKeefe (12 Jan 2020 22:05 UTC; 3 points)

• Cullen_OKeefe 12 Jan 2020 4:11 UTC

  3 points

  in reply to: acrlr's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I’m not in­volved in hiring at OpenAI, so I’m go­ing to an­swer more in the spirit of “ad­vice I would give for peo­ple in­ter­ested in pur­su­ing a ca­reer in EA AI policy gen­er­ally.”

  In short, I think ac­tu­ally try­ing your hand at the re­search is prob­a­bly more valuable on the mar­gin, es­pe­cially if it yields high-qual­ity re­search. (And if you dis­cover it’s not a good fit, that’s valuable in­for­ma­tion as well.) This is ba­si­cally what hap­pened to me dur­ing my FHI in­tern­ship: I found out that I was a good fit for this work, so I con­tinued on in this path. There are a lot of very cre­den­tialed EAs, but (for bet­ter or worse), many EA AI policy ca­reers take a com­bi­na­tion of hard-to-de­scribe and hard-to-mea­sure skills that are best mea­sured by ac­tu­ally try­ing to do it. Fur­ther­more, there is un­for­tu­nately a man­age­rial bot­tle­neck in this space: there are far more peo­ple in­ter­ested in en­ter­ing it than peo­ple that can su­per­vise po­ten­tial en­trants. I think it can be a frus­trat­ing space to en­ter; I got very lucky in many ways dur­ing my path here.

  So, if you can’t ac­tu­ally try the re­search in a su­per­vised set­ting, cul­ti­vat­ing gen­eral skills or do­ing ad­ja­cent re­search (e.g., gen­eral AI policy) is a good step too. There are always skills I wish I had (and which I am for­tu­nate to get to cul­ti­vate at OpenAI dur­ing Learn­ing Day). Some of the stuff I stud­ied dur­ing Learn­ing Day which might guide your own skill cul­ti­va­tion in­clude:

  • Ma­chine Learning

  • Eco­nomic Development

  • Fi­nan­cial Accounting

  • Ba­sic Economics

  What links here?

  • Cullen_OKeefe's comment on I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA by Cullen_OKeefe (12 Jan 2020 22:05 UTC; 3 points)
  • Cullen_OKeefe's comment on I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA by Cullen_OKeefe (12 Jan 2020 22:25 UTC; 1 point)

• Cullen_OKeefe 12 Jan 2020 4:05 UTC

  5 points

  in reply to: Linch's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I think it might be some­what more com­pli­cated. As far as I know, the LSAT+GPA mea­sure is ac­tu­ally a pretty strong pre­dic­tor of law school perfor­mance as far as stan­dard­ized tests go. But there’s some con­tro­versy in the liter­a­ture about how much law school grades mat­ter for suc­cess. Good law schools also have much higher bar pas­sage rates, though there could be con­found­ing fac­tors there too.

  In gen­eral, the le­gal mar­ket seems some­what weird to me. E.g., it’s pretty easy for T3 stu­dents to get a BigLaw job, but of­ten hard for stu­dents near the bot­tom of the T14 too. I do not un­der­stand why firms don’t just hire such stu­dents and thereby lower wages, which are very high. My best guess is, Han­so­ni­anly, that there’s a lot of sig­nal­ing go­ing on, where firms try to sig­nal their qual­ity by hiring only T6 law stu­dents. Also, I imag­ine the T6 cre­den­tial is im­por­tant for re­cruit­ing clients, which is very im­por­tant to BigLaw suc­cess.

  But query how much of this mat­ters if you want to do a non-ETG law path.

• Cullen_OKeefe 12 Jan 2020 3:38 UTC

  8 points

  in reply to: Khorton's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  It’s use­ful to think of the OpenAI Policy Team’s work as fal­ling into a few buck­ets:

  1. In­ter­nal ad­vo­cacy to help OpenAI meet its policy-rele­vant goals.

  2. Ex­ter­nal-fac­ing re­search on is­sues in AI policy (e.g., 1 2 3 4)

  3. Public and pri­vate ad­vo­cacy on is­sues in AI policy with a va­ri­ety of pub­lic and in­dus­try actors

  Most of my work so far has been fo­cused on 1 and 2, and less on 3. That work is largely what you would ex­pect it to look like: a lot of time spent re­view­ing aca­demic liter­a­ture or pri­mary sources; draft­ing pa­pers; so­lic­it­ing com­ments and feed­back; and dis­cussing with col­leagues.They also in­volve meet­ing with other in­ter­nal stake­hold­ers, which is an es­pe­cially ex­cit­ing part of my job be­cause the re­sult­ing out­puts will be very tech­ni­cally-in­formed.

  I plan to do some more of 3 this year, which will gen­er­ally be helping co­or­di­nate dis­cus­sion on some is­sues in AI eco­nomics. Stay tuned for more info on that!

  What links here?

  • Cullen_OKeefe's comment on I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA by Cullen_OKeefe (12 Jan 2020 4:39 UTC; 2 points)

• Cullen_OKeefe 12 Jan 2020 3:35 UTC

  9 points

  in reply to: Milan_Griffes's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I should also men­tion that I am gen­er­ally ex­cited to chat with EAs con­sid­er­ing law school. PM me if in­ter­ested, or join the Effec­tive Altru­ism & Law Face­book Group.

• Cullen_OKeefe 12 Jan 2020 3:20 UTC

  1 point

  in reply to: acrlr's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  I’m not ac­tu­ally sure what differ­ence you’re refer­ring to. Could you please elab­o­rate? :-)

• Cullen_OKeefe 12 Jan 2020 3:15 UTC

  7 points

  in reply to: JimmyJ's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Hm, I haven’t thought about this par­tic­u­lar is­sue a lot. I am more fo­cused on re­search and in­dus­try ad­vo­cacy right now than gov­ern­ment work.

  I sup­pose one nice thing would be to have an ex­plicit area of an­titrust le­niency carved out for co­op­er­a­tions on AI safety.