Feedback from STM on my previous attempt on Career decision

Mr. Agent,

I didn’t see the lifetime impact of the data science path. What has been the growth rate of “impact” for the two career paths as people got promoted or started new ventures? What are you assuming it will be?

Related: How many years of career are you assuming? That along with the growth rate would point to the lifetime “impact”. That total would be a better indicator of the expected value of some gamble like Master’s degree and H1B visa. For example, 8 years of a grind even starting at 30 may be worth the remaining 27 years of high pay compared to 35 years of what seems less risky but lower-paying.

For how many years will you be allowed to work in a given country? Can you go back to Europe or is it back to India after that?

The probabilities are sub 50% for success in either cases. In case I fail to make it to the US, I am stuck with very low ETG (5-10k).

Also look at other options that may have a higher acceptance rate for permanent residency and work options, even if they have lower overall compensation. Canada, Australia, other European countries, etc. Also India.

Then there’s the cost of living in different places. For example, Seattle has pretty high salaries and lower cost of living than the Bay Area.

In other words, when I hear “lifetime impact”, I expect to see lifetime calculations. Doesn’t have to be perfect, but has to consider the total career. A very crude one would be the total given some starting salary and some growth rate over X years of career. A more detailed one would be an average salary (or “impact”) of X1 for 5 years and X2 for the next 10 years and so on, and then calculating the sum over the career.

My plan is to roughly first move to DS ASAFP (3-6 months) and start “clocking” DS hours on the side (9-5 job). I will spend the next few years (0-5 years) focusing on gaining R skills, research skills needed for EA and try my luck at breaking into EA “in a way that creates a lot of impact”. (It took 5 years for Peter Hurford to start his own org.)

What is the opportunity cost of doing the above mix vs data science full-time?

Side note: Please avoid using rare abbreviations everywhere (such as GR or ETG).

Entry question

What should I do next? Should I go for a Master’s? Should I upskill for EA Orgs? Introducing Lifetime impact calculations.

EA jobs have high impact in Research

I am going to try making a lifetime impact calculation (very rough) based on Milan Griffes attempt here. People seem hesitant to make such calculations for some reason. And somehow they are able to make decisions without these. All I want to do is just see if it is even worth spending time to be a researcher in the already crowded space (OPPs recent hiring round).

GiveWell moved 110m$ in 2015. An entry level researcher’s contribution is expected to be 10% of a co-founder. Over a lifetime I find it hard to believe that one will be stuck at 10% of the co-founder contribution. With time one would move up the ladder and on average I assume one is at 30% the co-founder contribution. GiveWell employed 9 co-founders that year. So we get a multiplying factor 0.3/9.

Just like we divide the impact of the org among the different people, it seems appropriate that we should also divide the impact of the 110m that GiveWell moves, among the following: the Donors (e.g., Good Ventures), the middle men (GiveWell) and the intervention Orgs (AMF). The impact from the 110m$ seems to belong to each of these Orgs. I assume all Orgs have equal weight in this, for lack of any other better estimate. So we get a multiplying factor 0.33.

Replaceability of org

Now we get into counterfactual estimates of the org itself. What is the utility to the world if GiveWell existed, minus when they didn’t exist. I don’t know how to estimate this but it appears 80000hours uses a factor <=0.3 to estimate their counterfactual impact. We shall roll with this, for lack of any better estimate. So we assume 0.3 again.

So far the Impact for a researcher at GiveWell is estimated to be 0.3/9x0.33x0.3x110m=363k. This final value is TOO sensitive to any changes in the factors listed above. For example, if the ‘Contribution of the Org’, changed from 0.3 to 0.4, then the outcome increases by about 100k$ to 484k$.

Replaceability of person

Counterfactual estimates for the impact created by a person in an EA job, is the difference in impact between, when he takes the job and when he doesn’t. This is quite hard to predict. It not only depends on the next candidate who didn’t get this job, but what that next candidate did (spillover-effects). “Did he end up displacing other people in EA?”, “What happened in the rest of the displacement chain?” etc…

Depending on the situation this can vary from 0 to 363k$. Imagine a job opening where anyone who meets a “very high bar” can get that research position as there are enough funds for it. Such a person seems to take credit for all of the 363k. However, imagine a case where there are 2 similarly skilled people for one job that was open. A gets the job and B can do ETG but can donate only a few 100$ a year. This makes the counterfactual impact of A = 100$.

In reality there are many people and it is hard to trace the actual impact. Attempting to make estimates seems to need a lot more assumptions which I am unable to back up with data or examples. Moving on!

Conclusion

It appears that EA work has a lot of potential to do good ranging from 0k$ to even >363k$. My problem is there is a lot of uncertainty. And this number could be extremely high (even in millions) depending on the type of work one chooses and how much better he is than others.

Hard to get EA jobs. Needs credentials.

In the essay on “Talent Constraint”, this table below was presented. It shows how hard getting an EA job is.

Year Org position AR
2018 OPP GR <5%
2018 EAF GR & Operations 3%
2019 CE CE entrepreneurship program 11.7%
2020 CE CE Entrepreneurship program <1%
2019 CE Internship Mental Health <2.5%
2019 FWI Research analyst Welfare specialist 2.4%
       
2017 Y-comb startups 1.6%
2018 Harvard (#2) student 5.2%
2018 Yale (#3) student 6.9%
2018 Stanford (#6) student 4.7%
       
2018 California LA (#20) student 16%
2018 Florida (#34) student 42%

It takes a few years to break into EA. For Peter, apparently it took him 5 years of hardwork to get to founding his own org. MSJ, to get an internship in CE, took 2-4 years and quite some hardwork (writing your own research, showing up at EA conferences, criticizing research etc…). For Saulius, it took about 2 years and 2 internships.

I recently apped for an internship and didn’t get it. I probably need to work on it for a few years and copy actions of these examples. In addition I really suck at understanding claims, so add a few more years!

Conclusion: It could take about 2-5 years and “a lot of hard work” before I get into an EA org. The counterfactual impact “depends”. The chances for success are unknown, and its getting quite competitive as expressed in the acceptance rates.


DS for ETG

I am going to focus on DS instead of CS, for DS involves “critical thinking”, working with statistics, working with claims, hypothesis testing etc.–things that seem to help become a better EA researcher (in case I need them in the future). Also Earning potential wise, it appears there is not “much difference” (looking at the average salaries). So we stick with DS.

Earning potential is quite high in the US. Earnings are substantially larger than here in Netherlands (or Europe). I know DS people with 4yoe in Apple California earning 300k$ TC and others in Walmart with 300k$ TC. While here (Netherlands) people seem to be earning 78-103k as per Glassdoor. With such salaries as in the US there are people like (Jeff Kaufman) who are able to donate 50%. This amounts to about 150k$ per year with a 300k$ TC.

There are 2 ways many people I know take, to get to the US. People take the masters route, and then work towards an H1B visa which allows you to work there. The other route is the L1 route. In this case you work in say Amazon in India (for e.g.,) and then try to move to the US on an L1.

Chances are low to make it in the US

There are random factors associated with working in US. Getting an H1B is based on lottery. Whether you are doing an L1 or Masters in the US, you end up having to deal with H1B at some point–either after Master’s or before your L1 expires (<5years). There seems to be a 60% chance that a person like me will clear this hurdle 1. This is including atleast 3 tries at H1B (i.e., 1-3 years of applying).

In addition, the US is not the most friendly place for workers rights. Once you loose a job you have 60 days to find another job. In times of recession (like in 2008 or during the Corona period), the chance of loosing the job seems to be about 15%2 and might mean a one way ticket back to home. This comes to a multiplying factor of 89%.

For Master’s route, I think I have “good” chances of getting a DS job in FAANG with 2-3 yoe (not including the H1B chances). I have seen examples of people in FAANG whom I know well, and thus think it is definitely within reach comparing myself to them. Worst case it should be possible to atleast get a DS job in California, from where I can grow into FAANG. People are able to do bootcamps and join DS from different industries such as Electrical Engineering. This suggests to me that it is definitely not out of reach. Let’s put this at 80%. So in total, 60%x89%x80%=42% chance of making it.

For the L1 route, the biggest hurdle would be to find a company that is L1-able. There are examples I have come across, across fields and countries. There are people who went all the way to JP Morgan in New York on an L1 after a few years in India and SWe engineers from India who are now in FAANG. People demanding a job in US because they are good and pushing for a job there. Atleast for mechanical background, people from ASML Netherlands regularly work in the US for a few years on L1. It seems accessible. I have to look into the possibilities more for DS people in Netherlands. Say 60-80% chance. So in total, 32% to 42%.

DS Costs (time and money)

With the Master’s route based on examples I have seen, 4 to 6 years to a FAANG 300k$ TC job (2-4 yoe + 2y for masters). With the L1 route 2-3 yoe + 2-5 years before getting H1B transfer. So, 4-8 years from now.

With the regular masters route, I would have to spend 50-70k$ at a public university and mostly can’t work during that time–becoming a total of 70-90k$ (including the opportunity costs). With the L1 route, I think I have to do an online masters (estimated at 20k for 2 years) atleast to be able to defend an RFE3 and also to improve chances with H1B4. But I will be working during that time and can cover my costs as it will be part time.

Forget spending 2 hrs a day on things related to EA

With the Master’s route, combined with my actual 9-5 job I think there is not going to be much time until I reach FAANG. Starting with a DS Masters implies that for the next 4-7 years I will be investing all my resources on making it into FAANG. First I start with GRE, get a job here (NL) in DS, work out the universities, apply, send documents, talk to people about where to go etc. During Master’s I think I will tend to spend all my time to get better grades and work on DS, and apply for jobs. I would want to do everything in my power to learn skills needed and get good grades to maximize my chances. Until I reach FAANG, I would be doing things that can improve my chances.

With the L1 route, I think the same holds too. If I end up working in Netherlands and doing a masters part-time, there is going to be no time. Forget spending 2 hrs a day. Once I get to the US on L1 I will be working on getting an H1B type job. That would be the focus. So both cases I guess that spending time to build “EA related skills” is gonna be hard during the time leading to FAANG.

When I fail

The probabilities are sub 50% for success in either cases. In case I fail to make it to the US, I am stuck with very low ETG (5-10k). And furthermore it is going to be much harder to start my journey into EA. I will be in my 33-36’s and I am afraid its gonna be “harder” than now as I would have passed several “crucial years” (4-7) where I could have spent learning “skills related to an EA”.

In the Master’s case this seems to imply that I could be in debt of 50-70k$ in case things don’t work out (worst-case). This means that I have to pay my debt for another 5-7 years before I can potentially ETG again. For the L1 case, atleast I am not in going to be in debt.

Considering the debt part alone and the reduced anxiety due to that, I would prefer the L1 route. It is still unclear what the chances are with meeting an L1-able company. L1 on the other hand also allows you to gain a few years (2-4y) experience before applying to FAANG (increasing the chances). However, in theory if someone pays for my Master’s (my brother perhaps) and I had “experience” (of 2-3 years), then both options are “almost” the same.


Conclusion

DS route

The Impact with DS seems to be high in the order of 150k (and more maybe) donations. Getting a job in the US seems to be coming in with a probability of less than 50% (most of which are not in my control) be it Master’s or be it the L1 route .

If this path is chosen, there is a “good chance” that I am unable to sink my teeth into up-skilling myself in other things such as EA (during the 4-8 years of pursuing this path).

I would be in my 33-36’s if I crash and burn. I am afraid its gonna be “much harder than now” to try EA as I would have passed several crucial years (4-7) where I could have spent learning skills related to an EA.

EA route

So far we have seen that breaking into EA can take 2-5 years and a “lot of hard work”. The probabilities for success are unclear, but the impact seems to have a lot of potential i.e., >350k$ (the lower limit could be as low as few k$; very hard to estimate currently).

If this path is chosen, it means that I still need a day job (25-40 hrs). Continuing to work on my usual MechE job is possible until a time that I am able to move to EA full-time.

In case I crash and burn, then all I have is some skills that possibly wont help me in ETG. MechE is a low paying job. And moving to DS will be “much harder than now”. So I am delayed by 2-5 years in getting started with DS.

The path that has it all?

Having to restart after 4-8 years “sounds like a bad idea”. I will be older in my 34s to 38s, and am not sure about how it is perceived and if I can get a master’s at that time, if I can move jobs etc. Also learning skills is supposed to be harder in some cases as you move to a middle age (37-48)5.

However there seems to be a “better” route. Instead of uselessly clocking time on MechE, and not having back-up options, I could first quickly change to a DS job (3-6 months). From there on I could pursue EA route in the evenings and weekends, while I still clock DS time. Granted I wont become great at DS but I will have yoe atleast, and when I crash and burn, I wont be starting anything from scratch.


Footnotes

  1. Probability calculation

    The number of times a person can apply for H1B on a Master’s is around 3 times on a STEM OPT (answer by Carlos Cueva). With an L1, we assume we are ready from the 3rd year on to apply for an H1B (having secured another job by then). So we will have the same 3 chances to apply for H1b.

    Note: Some companies do not hire you in your OPT if you don’t have atleast 2 attempts left. Also there is a possibility of applying with many H1B sponsors for the same year with different employers. So lets assume I have 3 tries for now.

    Based on a 50% split between the general pool and the advanced pool (which is what the DHS have assumed), we see that 1 H1B attempt has a 41.9% chance of success for students with Masters.

    After you are selected in the lottery, you are either approved or rejected or requested for an RFE. The cases for RFE have become “much higher” (33% in 2019 from 10% in 2016) over the trump era. One of the things they seem to be scrutinizing the people is for not having the right background/experience for the work. This comes under “Beneficiary qualifications”. Naturally, this scares me as I have a mechanical Master’s and Bachelor’s. Another common reason issued by the USCIS is called “Speciality Occupation”. The position needs to qualify as a speciality occupation. Apparently USCIS does not usually think a job is “special enough”, if they don’t have the need for a relevant bachelor’s. If you look on trackitt, you do see that “most” of the RFEs are cited based on this “Specialty Occupation” or “lack of relevant degree” (Beneficiary qualifications I presume). I seem like a prime suspect for atleast lack of “beneficiary qualifications”. Thus, I take a 100% chance of getting an RFE.

    Whether the RFE gets approved or not seems to be random. There are cases of people with Bachelor’s and Master’s in Mechanical who have made it to software job after their RFEs (ronaldo7!, tiger12) post 2018. And then there are people with Master’s in Financial Mathematics who are denied after their RFE to a CS/DS job (9mineHwang). A Hiring manager at Amazon sums up what a giant cluster fuck of a mess it is: “One of my report’s visa was rejected saying the job is not a specialty occupation and doesn’t need Bachelor’s degree which is ridiculous as why would Amazon pay that person 300K$ per year for a non-specialty job. The decision on visa application has become so arbitrary that the outcome is not known. One employee has one application rejected and the same application was approved the second time.”—hiring manager Amazon.

    According to Fragoman the success of RFE approval amounts to 60.4%. So, 41.9% x 100% x 60.4%=25.3% for 1 attempt of H1B. Over three attempts the chance becomes 1-(1-25.3%)^3=58%.

     

  2. Probability of layoffs

    US has “had” about 8 recessions in a span of 70 years. Probability of recession is 11.4%. It is assumed that those are the only real times when your job is in danger.

    % Layoffs during a recession for H1B people

    In 2009 it is estimated that 650k people were on H1B. Adding 230k H1B grants every year, this brings the sum to a maximum of 3m H1B people in the US.

    The logged layoffs in this current “recession” due to the pandemic fucking things up for H1B users in tech startups, is logged at layoffs.fyi. It is currently 60k. If we assume all layoffs to be a factor 3 greater i.e., ~200k, we get to 6% layoffs. And mind you this number also has a “portion” of Americans and GC holders and even layoffs in other countries, and hence appears to be conservative.

    The chance that you don’t get laid off is:

    (1-8/70)+8/70x6% = 89% 

  3. Why an online masters from US “accredited university” is important to defend RFE?

    If you look on trackitt, you do see that “most” of the rejects are cited based on this “Specialty Occupation” or “lack of relevant degree” (Beneficiary qualifications I presume). I seem like a prime suspect for atleast lack of “beneficiary qualifications”. Defending a Data science related RFE of Beneficiary Qualifications with Mechanical degree (Bachelor’s and Master’s) might turn out to be hard or atleast incentivizing people at USCIS to reject me. People with Electronics backgrounds with such a “Beneficiary Qualifications” RFE, for a CS job, seem to show how each subject is related to computer science in their rebuttal to USCIS.

    Despite the fact that the RFE rejections are looking random–as they even reject CS students for SWe jobs–it appears that pushing for a degree might help an RFE with “Beneficiary Qualifications” atleast.

    Later I can get more info by contacting firms like FrogoMan. 

  4. There is a difference of 15% in making the lottery, depending on if you are part of the Master’s cap or regular cap. I have a masters from a non-US university which does not seem to cut it, to be part of the Masters pool in H1B. To be part of this Master’s pool, an online masters from the US seems to cut it, according to Murthy Law firm:

    The advanced degree exemption requires a degree from an accredited, nonprofit U.S. college or university. It does not matter if the program was part-time, evening, and/or online. Just check the nonprofit / for-profit status of the school and the accreditation…

    Other links related to this are: Quora, IOBerkeley 

  5. “Acquisition of Word-Processing Skills by Younger, Middle-Age, and Older Adults”

    The study included three age groups composed of 15 female subjects each: younger (M age - 21. 5 years. SD - 3.5. range 18-28): middle age (M age - 42.2 years. SD - 3.1, range = 37-48; and older (M age

    • 60.7 years SD = 3.6, range = 55-67)

    On Item 5 (differentiating between regular and align tabs) both the middle-age and the older group performed less well than the younger group, but did not differ between themselves (p < .05).