Agreed. I’d extend the claim past ideas and say that EA is very non-elitist (or at least better than most professional fields) at any point of evaluation. 

Maybe because of that, it seems more elitist elsewhere. 

Like the idea but it might be at odds with the recent AI governance shift. In general, policy folks of all stripes, especially the more senior and insider ones, practice a lot of selective disclosure.

Having done a lot of this advice in my 20s, I'd recommend just getting started with an online training program you find interesting, seems career relevant, and also not too pie-in-the-sky as a near-term plan. Throughout my life, I think there were one or two that felt unusually good or bad all-things-considered. Even then, training programs are short (~6 weeks) and have no stakes if you stop them.

(The exception is if the training somehow includes hands-on training from someone actively trying to progress in one of your desired career paths. Good mentorship is a scarce resource and you should prioritize it above a lot of other things.)

It's dramatically more important what you do after the online training program. It's extremely rare that these programs set people up to to do the "impressive project" that hiring managers want from less prestigious candidates. If they did, everyone would be doing them. 

As for the program, if you feel like you're at least passing the course (whatever than means) and it seems promising, then I'd pair that with some informational interviews. You can ask "Hey I've been doing X training course and feel like it might be a good career path. Would you be willing to chat about how you got to where you were?".

That will help you identify directions to take for your "career ladder", which I put in quotes since it's really more of a fog-of-war. Unfortunately, it's usually the things between "Step 1" and "desired job" where steps are the least clear and the most consequential. So I would save your energy for when you get there.

I've done a lot of partially blind hiring processes both within EA and outside it ^[1]. And as much as I like them (and feel like I've benefited from them), I think there's good reasons why they aren't done more.

• It seems really hard to create a good blind hiring process. Most of the ones I felt good about were constructed with an immense amount of care to balance not rejecting good candidates but still having enough subtlety to distinguish candidates that would pass a final stage. Even then I still felt like there would be exceptional candidates that would have fallen through the cracks because they were bad at the specific skills being tested in the blind stage. I still think the benefits are overall positive but I'm not super confident in that given the risk of mistakenly rejecting good people.
• There were always at least two blind stages before the final un-blinded one: the first being a fast filter (either an automated quiz or a short task that could be graded in 1-2 minutes) and the second being a very in-depth task. Granted, this isn't experimental evidence but it does suggest that one fast blind part isn't enough.
• The second in-depth task seems very time-consuming to maintain. Having graded anonymous work tests for my current role where there was a big "we want to hire overlooked internationals" motivation, I felt like I needed at least 5 minutes per task before I felt even semi-confident what my final feeling was going to be and usually quite a bit more to be sure and fulfill the spirit of a blind evaluation.
• Many roles have downside asymmetry. I've mostly seen blind tests in highly technical fields where the organization can benefit if the candidate is a superstar but also isolate the damage if they turn out to be highly toxic. With operational roles, the downside is much larger while the benefits are smaller.
• Anecdotally, blind hiring these days doesn't seem guaranteed to increase demographic diversity and may even decrease it a tiny bit. I feel the most confident on this with initiatives to increase women in software engineering via hiring practices. But I'm a lot less confident on country-of-origin. My hunch is it would backfire a bit in the non-profit world, especially in effective animal advocacy building where there seems to be some attempt to build multinational capacity.

1. ^{^}

   Applied seriously to a software engineer role at a prestigious tech firm and had a final stage interview that far exceeded my abilities, which was painful for everyone involved. Applied on a whim to the Charity Entrepreneurship incubation program and got rejected after the second (third?) stage. Accepted the job offer to my current credentialed non-EA job after two blind stages and a final in-person. Applied seriously to Charity Entrepreneurship Research Training program and got rejected after second stage. Applied as a long shot to a GiveWell Senior Researcher role and got rejected after second stage.

A quick drive-by comment on "4. Missed RCT Opportunity": The sample size seems way too small for a RCT to be worth it. There's not much statistical power to work with when researchers are studying a messy intervention with only 6 countries. And I imagine they'd struggle to attribute changes to the Technical Support Units unless it was something truly transformative (at least within the framework of the RCT).^[1]

More broadly, I'm not aware of any commonly accepted way to do "small n" impact evaluation yet, especially with something as customized as Technical Assistance. This blog post from 3ie, a NGO to promote evidence-based policy making, talked about the issue 13 years ago and I think it's still broadly true. The impact evaluation toolkit works best with (1) a precisely defined intervention, (2) a decent sample size (say n > 100), and (3) a very homogeneous sample. This grant, so far, looks to be the opposite of all 3.

1. ^{^}

   I also recall the math for statistical inference gets strange when using very small sample sizes (say n<15) and may require assumptions that most people consider unrealistic. But I could be wrong here.

Yeah this was what I found too when I looked into private US long-term disability insurance a while back. My recollection was:

•  there's a surprising number of health exclusions, even for things that happened in your childhood or adolescence
• it's a lot more expensive in percentage terms if you're at a lower income
• many disability cases are ambiguous so the insurance company may have you jump through a lot of hoops and paperwork (a strange role-reversal in which the bureaucracy wants to affirm your agency)

I had the impression that it was a great product for some people, meaning those with high income, clean medical history, and a support network to wrestle with the insurance company. But at the time I looked into it, it didn't seem like a great option for me even given my risk-adverse preferences.

Planning to look again soon so could change my mind.

I find searching for in-depth content on the EA Forum vastly better than Reddit. This isn't just relating to EA topics. There are a few academic-ish subreddits that I like and will search when I'm interested in what the amateur experts think on a given topic. Finding relevant posts is about the same on Reddit but finding in-depth comments + related posts is very hard. I usually have to do some Google magic to make that happen.

Also on rare occasion, I end up liking a person's writing style or thinking methods and want to deep dive into what else they've written about. On the EA Forum, about 100% of what I find will be tangential to things I care about. On Reddit, it's more likely I'll have to sift through lots of hobbyist content like about sports since it's more of a "bring your whole self" platform.

I'm loving this series so far. I got two questions if you've got some time to answer them.

What categories do you use for time-tracking? I find research tasks unusually hard to categorize.

Do you find that earlier stages in the Ideation -> Exploration -> Understanding -> Distillation pipeline take more time to get good at? My experience is that I improve at later stages far earlier and far faster than earlier stages (passable at Distillation before Understanding, passable at Understanding before Exploration, passable at Exploration before Ideation). And anecdotally, I heard people can take a very long time to come up with a good research idea.

I found it very valuable but (barring any major changes like there being 5 new organizations of similar impact) I wouldn’t find an expanded or updated version that useful.