Comment by lapsedacademic | Hacker News Reader

Comment by lapsedacademic | Hacker News Reader

Parent Full thread

> It's hard to pass up making my 2x or more in industry.

Austin's salary is public data, and I have a really good sample of what people like him make in industry.

Try 5x. Before RSU appreciation.

Maybe 3x. But 3x would imply a VERY chill industry position. "Name your interest space and we'll let you work on that" type of gig. Probably way more chill than academia.

> The freedom and unstructured time to explore fuzzy problems is unmatched.

Industry also provides this. Actually, I find much more freedom in industry than in academia. And, if we're honest, my industry position is probably at least as secure as a faculty position at a public university in a red state.

> I don't know any other job that gives you 6-figure salary and lets you basically do whatever you want for 42% of the year.

In industry research, you can't do whatever you want. But you can usually do what you want.

In fact, the caveat of "prove your worth" is a huge tell. In industry, you REALLY CAN do what you want! Real freedom; just be useful. In academia you have "freedom". By which we mean "raise money".

But even if not... look at it like this: an industry position that pays 5x allows you to work for a year and then take the next 4 off. 2/5 is bigger than 1/5. And that is time you can actually take off. To do whatever. Not "off" in the "better be writing grant proposals!" sense. Actually off.

> Is a 2x salary increase worth giving up that kind of time?

Probably not. But if you can land an R1 CS position in SE, you should be making closer to 5x.

Do you know anyone who made $500k for a year and then took 4 years off to work on their research? I'd like to see what they managed to do with their time.

To be fair, no. (People do switch between "low key pure-R" jobs and "higher stress r&D" jobs. Perhaps similar.)

But... perhaps this is because after you've made $industry_money for 10 years and hit your late 30s, you realize you'd rather spend all of your time training your biological AGI agents and don't really give a shit about the kinda unpleasant industry rat race, much less the totally brutal higher ed rat race.

Or skiing. Lots of multi-millionaire ski bums in my grad cohort. No idea what they do in the summers. Not my cup of tea, but probably better than writing papers no one will read.

In early retirement, there's time for research. And time to share that research. Just... no incentive to do the rat race thing that's necessary to convince an academic you're "accomplished something". But who cares? You finished the race. You don't have to get paper awards or NSF grants. You just have to do work that matters. Or ski. Or change diapers.

So, I guess, to be even more fair: yes. CS PhDs work through their 30s and then retire. Totally. All the time.

A friend of mine did! He's been pretty productive, both in terms of academic things and living a good life.

CS is nice in that you don't (initially) need access to a lot of resources. I think this would be much harder to do in a wet-lab field, which is one of the reasons I'm still here.

That's not too different than what I did. I left Google fall 2018, at the time thinking I'd write a video game. That morphed into doing research on 2D rendering, because building the infrastructure for that was more interesting and exciting than the actual game. For a while (not the whole time) it was pure unstructured time, no outside responsibilities. I rejoined Google a year ago with a research software engineer title, mostly because it was a better place to pursue this research than just continuing on my own.

Partly I'm posting this to support 'lapsedacademic's assertion, you can do good research in industry. It takes a certain element of luck to land a spot, but it also helps to be very passionate about the research itself.

I'm now writing a paper on an exciting GPU algorithm, but it's pure love, funding doesn't depend on it, it just makes a nice point on my list of accomplishments during perf time.

I'm curious, what are your options for starting a company based on the research you're doing at Google? What kind of ownership do they get (I assume it's not nothing)?

I thought about starting a company, but that's not where my energy is, it's doing the science. Everything I do is permissively licensed open source (though if I had a chance to work on awesome rendering in a proprietary product I would not turn it down).

I'll also add, I looked into academia and talked to some people about steps in that direction, for example seriously considering taking on teaching a class at UC Berkeley. But I decided that the daily grind of academia was not for me. I think I get to spend a larger fraction of my time actually working on my problems at Google than I would at a university.

What genre of game did you want to build, out of curiosity?

A Zach-like with music synthesis as the core mechanic. I still think it's a good idea, but I don't really know how to write fun games (I think you have to write one or two to get enough experience) and I do know how to write 2D renderers, and increasingly feel able to claim I know how to program GPUs as well. I think it was Ivan Sutherland who said a good plan for research is to do something you think is easy and other people think is hard.

That's a pretty neat idea! (Though probably hard as hell to design...)

>you REALLY CAN do what you want! Real freedom; just be useful.

exactly. high-level ICs at FAANG set their own schedules/roadmaps and salaries as long as they pushing expanding the forefront of an area that brings value. the truth is academia is a shelter for a lot of people who can't do useful things.

> the truth is academia is a shelter for a lot of people who can't do useful things.

And that's a good thing, because often times it's not obvious immediately how a particular thing is useful or brings value. Sometimes a particular area of inquiry doesn't appear promising until a subsequent invention many years later.

For example, driverless cars were incubated in academia. Now corporations are using those same techniques. A lot of researchers crashed a lot of cars before a winning combo was discovered. You don't get positive results without a ton of failures and not-useful ideas. It's worth noting that industry was not the one footing the bill here to get all of these negative results. There were no high-level ICs at Google doing this research when it was being done at CMU. Google only started doing it only once the the techniques were proven at the DARPA grand challenges.

>Sometimes a particular area of inquiry doesn't appear promising until a subsequent invention many years later.

this could be the tagline of every single application to NSF. i'm not sure if you're in academia but there are hordes of projects that are funded that are clearly useless.

>For example, driverless cars were incubated in academia.

are you really claiming that driverless cars weren't immediately obvious as valuable? just because CMU got to it first doesn't mean industry wouldn't have gotten to it.

> hordes of projects that are funded that are clearly useless

I don't understand what your point is. It's not like industry is immune to funding ideas that are clearly useless (Juicero) or even fraudulent (Theranos). If you have a way to fund only good ideas, I think that would be quite a breakthrough. I mean, look around you: this community is set up around a good-idea selection engine that itself struggles to consistently identify good ideas. Bad ideas are a part of getting to good ideas. If you really had a 100% fool-proof way if identifying bad/good ideas at the pre-funding stage you wouldn't be here talking about it -- you'd be using it to make a ton of money.

Also, I would challenge you to point to a NSF funded idea that is 100% clearly, objectively useless. Something so clearly has no plausible utility whatsoever. I think this would be very hard to do.

> are you really claiming that driverless cars weren't immediately obvious as valuable?

It's true that industry could have but they didn't, and that's the point. It only became attractive to industry after the public sector dumped a massive amount of R&D into it and basically proved how to do it to industry, and that's not a coincidence. Before the the DARPA grand challenges it was absolutely a question as to how viable the idea of driverless cars were. This story is not an uncommon thing.

>I don't understand what your point is

No it's I that don't understand your implicit point that we should let academics fiddle. Is it just because there's some grand narrative around science being objectively and absolutely for the betterment of mankind? The ideological devotion to science as some kind of monastic pursuit blinds you to the very real facts (on the ground) about how much funding is wasted on one off papers, one off projects, whole conferences devoted to areas that will never improve anyone's life or generate absolutely any return on investment.

>Also, I would challenge you to point to a NSF funded idea that is 100% clearly, objectively useless

It's very easy: pick any nsf grant going back as far as you'd like (such that there's enough lead time) and look at the number of citations on the papers generated from the grant. I don't even have to do this because everyone knows that probably 50% of all papers get absolutely zero citations (https://www.science.org/content/blog-post/not-so-many-uncite...).

>It's true that industry could have but they didn't, and that's the point

Again I have no clue what you're claiming here. You're trying to make some kind of case for academic science shouldering the burden of some fraction of fundamental (pie in the sky) research by citing self-driving cars which is positively laughable as an example of such research.

> whole conferences devoted to areas... pick any nsf grant... and look at the number of citations

Do you see the contradiction here? I don't think bibliometrics are useful for measuring utility.

> I don't understand your implicit point that we should let academics fiddle

I'm actually totally okay with letting academics fiddle. They are super cheap and you're mostly just paying them for small amounts of their time.

But, it should be the faculty who are allowed to fiddle. They shouldn't be given resources to direct other people's fiddling time. In particular, I think we should massively reform the graduate student system to invert the power relationship between faculty and students on any highly exploratory projects.

Specifically, for any grant whose purpose is "fundamental science" and/or training (e.g., ALL NSF money as a starter):

1. the agency funds students instead of faculty. So 100% of the money that goes to graduate students on NSF grants should be redirected to a GRFP-like funding model. This means that NSF grants to faculty should only fund PI summer salaries & shared department resources. Never students. Want a student? Recruit them to collaborate with you.

2. NSF should put a hard upper bound on the number of funded teaching hours permissible or required and funded through any sort of stipend.

NB: students can still teach more hours! But then they will be normal W2 employees who are paid prevailing rates, are included in faculty+staff retirement/pension/benefits, get FICA benefits, etc. The point: if your uni takes a single dollar from NSF, then student stipends can only be actual stipends, not back doors for tax-advantaged ad junct labor that excludes universities from paying FICA taxes on behalf of their teaching staff (who happen to be grad students).

MIT is one step ahead of you, they already ding faculty with NSF GRF students extra overhead because ostensibly the NSF fellow is there for MIT's brand and not the professor's research.

Yeah, places like MIT are where I got the model. Seeing how my peers were treating at places that are not like MIT is how I got the motivation to care.

Tippy-top programs in any particular field already operate in a way that treats grad students more like students than itinerant labor.

It's the other 99% of institutions that are gutting scientific human capital in the name of empowering incompetent middle management. (Professors are middle management. Most of them are incompetent at that job.)

>Do you see the contradiction here? I don't think bibliometrics are useful for measuring utility.

that's not a contradiction that's like just your opinion man. it's also the no true scotsman fallacy on your part.

>we should massively

>NSF grants should be redirected to a GRFP-like funding model

>NSF should put a hard upper bound

cool when you replace panchanathan you can institute all of the policies and only then will academia be more egalitarian and possibly more useful to society as a whole. until then you're just playing bait and switch wrt the current state of things.

> that's not a contradiction that's like just your opinion man. it's also the no true scotsman fallacy on your part.

I don't follow.

Useless work gets cited all the time. I'm not suggesting that there should be less accountability to be useful to society. I'm just pointing out that bibliometrics is a particularly bad way of measuring utility.

> until then you're just playing bait and switch wrt the current state of things.

Again, I don't follow. How am I playing bait and switch? I don't work in academia.

> replace panchanathan

Fortunately, that's not necessary. You don't put pressure on agencies like NSF by joining the civil service. Even high-ranking civil servants are... well, servants. You put pressure on federal agencies by having tons of money and free time. Which many lapsed academics in CS have in spades ;-)

Reply on news.ycombinator.com