I like this advice a lot but want to add a quick asterisk when transitioning to a new field.
It’s really really hard to tell what an expensive signal is without feedback. If you’re experienced in a field or you hang out with folks who work in a field, then you’ve probably internalized what counts as an “impressive project” to some degree.
In policy land, this cashes out as advice to take a job you don’t want in the organization you do want. Because that’s how you’ll learn what’s valuable and what’s not. Or taking low paid internships and skilled volunteering roles. Or dropping a lot of money to attend a conference for your target field.
It’s also really hard to know the steps to executing the “impressive project” (which is why the signal is so expensive!). With internships and skilled volunteering, you’ll get supervision. And even a light touch can prevent you from investing a ton of time in something that doesn’t matter. Or get reassurance that task X really does take everyone a long time so don’t feel bad about the time sink.
But with grants or independent work, you’ll have to seek out the feedback, brief them on project and hope you’ve given enough context for useful feedback, and also hope you picked someone who knows your area well enough. (I haven’t had success here and I’m not sure how realistic it is for most people.)
Work tests are awesome here since they’re mini-projects. But feedback is often noisy and hard to interpret since there aren’t good incentives for orgs to specialize in concrete feedback. I’ve interpreted this feedback wrong in both directions (first being too optimistic about a generic but lengthy “there were many strong candidates” and then too pessimistic about the terse but personalized rejections encouraging me to still consider research as a career)
The point I’m trying to make is that the idea of “cheap tests, expensive signal” is probably a lot easier for mid-career folks to apply independently. But for people without any experience, the advice depends on whether you can get supervision from an organization. Without that, it may be better to just “get your foot in the door” in any way possible. Maybe a “good enough cool sounding project” helps to demonstrate interest, but it’s tough for people to perform at 1-year of experience level before they have that 1 year of experience.
I imagine you can get a lot of the value here more cheaply by reaching out to people in the field and asking them a bunch of questions about what signals do and do not impress them?
Doing internships etc is valuable to get the supervision to DO the impressive projects, of course.
EDIT: Speaking as someone who does hiring of interpretability researchers, I think there’s a bunch of signals I look for and ones I don’t care about, and sometimes people new to the field have very inaccurate guesses here
I do think this can run into Goodhart’s Law problems. For example, in the early 2010s, back when being a self-taught software engineer was much more doable, and it was a strong sign when someone had a GitHub profile with some side projects with a few months of work behind each of them. GitHub profile correlated with a lot of other desirable things. But then everyone got that advice (including me) and that signal got rapidly discounted.
So I guess I’d qualify that with: press really hard on why the signal is impressive and also ask people explicitly if they agree with the signals you heard from others (ex. I heard from people in field that signal X is good / bad, do you agree with that?)
I think this is a valid long term concern but takes at least a few months to properly propagate—if someone qualified tells you that when hiring they look at a github profile, that’s probably pretty good for the duration of your job search
This advice sounds right to me if you already have the signal in hand and deciding whether to job search.
But if you’re don’t have the signal, then you need to spend time getting it. And then I think the advice hinges on how long it takes to get the signal. Short time-capped projects are great (like OP’s support on 80,000 hours CEO search). But for learning and then demonstrating new skills on your own, it’s not always clear how long you’ll need.
I like this advice a lot but want to add a quick asterisk when transitioning to a new field.
It’s really really hard to tell what an expensive signal is without feedback. If you’re experienced in a field or you hang out with folks who work in a field, then you’ve probably internalized what counts as an “impressive project” to some degree.
In policy land, this cashes out as advice to take a job you don’t want in the organization you do want. Because that’s how you’ll learn what’s valuable and what’s not. Or taking low paid internships and skilled volunteering roles. Or dropping a lot of money to attend a conference for your target field.
It’s also really hard to know the steps to executing the “impressive project” (which is why the signal is so expensive!). With internships and skilled volunteering, you’ll get supervision. And even a light touch can prevent you from investing a ton of time in something that doesn’t matter. Or get reassurance that task X really does take everyone a long time so don’t feel bad about the time sink.
But with grants or independent work, you’ll have to seek out the feedback, brief them on project and hope you’ve given enough context for useful feedback, and also hope you picked someone who knows your area well enough. (I haven’t had success here and I’m not sure how realistic it is for most people.)
Work tests are awesome here since they’re mini-projects. But feedback is often noisy and hard to interpret since there aren’t good incentives for orgs to specialize in concrete feedback. I’ve interpreted this feedback wrong in both directions (first being too optimistic about a generic but lengthy “there were many strong candidates” and then too pessimistic about the terse but personalized rejections encouraging me to still consider research as a career)
The point I’m trying to make is that the idea of “cheap tests, expensive signal” is probably a lot easier for mid-career folks to apply independently. But for people without any experience, the advice depends on whether you can get supervision from an organization. Without that, it may be better to just “get your foot in the door” in any way possible. Maybe a “good enough cool sounding project” helps to demonstrate interest, but it’s tough for people to perform at 1-year of experience level before they have that 1 year of experience.
I imagine you can get a lot of the value here more cheaply by reaching out to people in the field and asking them a bunch of questions about what signals do and do not impress them?
Doing internships etc is valuable to get the supervision to DO the impressive projects, of course.
EDIT: Speaking as someone who does hiring of interpretability researchers, I think there’s a bunch of signals I look for and ones I don’t care about, and sometimes people new to the field have very inaccurate guesses here
Ooh good idea. I should do more of that.
I do think this can run into Goodhart’s Law problems. For example, in the early 2010s, back when being a self-taught software engineer was much more doable, and it was a strong sign when someone had a GitHub profile with some side projects with a few months of work behind each of them. GitHub profile correlated with a lot of other desirable things. But then everyone got that advice (including me) and that signal got rapidly discounted.
So I guess I’d qualify that with: press really hard on why the signal is impressive and also ask people explicitly if they agree with the signals you heard from others (ex. I heard from people in field that signal X is good / bad, do you agree with that?)
I think this is a valid long term concern but takes at least a few months to properly propagate—if someone qualified tells you that when hiring they look at a github profile, that’s probably pretty good for the duration of your job search
This advice sounds right to me if you already have the signal in hand and deciding whether to job search.
But if you’re don’t have the signal, then you need to spend time getting it. And then I think the advice hinges on how long it takes to get the signal. Short time-capped projects are great (like OP’s support on 80,000 hours CEO search). But for learning and then demonstrating new skills on your own, it’s not always clear how long you’ll need.