I meant that if EtG is important to you, then why not just go all in on EtG? You might say that AI-related career advice assigns higher priority to research, but I think that assumes you only pick one of these and not split your attention. I think 80k would encourage focusing on EtG if it is important to you for some reason:

Overall, earning to give typically ends up seeming most attractive in the following cases: ... There’s a particular job you really want to do for other reasons that’s higher-earning than average, or that can be adapted into. This might look like Jeff working as a software engineer, or someone who really always wanted to be a doctor, or a teacher who does private tutoring and donates the extra. ...

I think I'd only encourage seriously doing two of these in parallel if both are important to you for some reason other than impact.

How do you expect to spread your effort across the two? What major sacrifices are you making (e.g., not attend a CS degree, work +N hours a week compared to average) that enable this, and what is your estimate on how it impacts your productivity in the long run?

I’m still sceptical about your plan on pursuing two very different careers in parallel. It’s not an issue if EtG is important to you but I’m not seeing what justifies doing two in parallel? Why is it better than one person doing EtG and another doing CS?

On the original question, I think it’s very difficult to tell from the outside. You want to do projects to gain experience that lets you enter certain roles (e.g. entry AI safety roles). The people hiring for such roles are the ones who can tell what they are looking for and how people usually get there.
 

On immigration, I think the EA orgs are terrible entry examples for counterfactual impact. There is already large interest to enter them despite the small size of the field so you'd probably replace a similarly good candidate who already had US work permit.

For AI companies, you could argue that you'd likely replace someone who has potentially never thought about safety before. But the AI entry roles are also incredibly competitive so it's not that you directly replace someone, it's more like you change the candidate distribution from 1000:10 to 1000:11 (non-safety vs safety focused).

I think it's worth considering upskilling outside these bubles and then transfer as your profile becomes more scarse.

I considered "doing work better" to be included in "doing work differently," but probably better to name these explicitly, thanks! Either the direction could be different (different goals) or the productivity (difference in effort/fit/etc).

AFAIK research suggests that people are terrible at predicting what they would enjoy. (Check 80k for this). But personal fit is a crucial thing, once you actually tried (seriously and with being as open as you can) and it does not feel right, you should move on and try something else. After a few diverse “samples” you should get a rough idea of your options and decide what to stick with.

Hey! Great to hear you are considering impact for your career decision!

EE definitely has an overlap with 80k priorities, e.g. https://80000hours.org/career-reviews/ai-hardware/ (which includes security/cryptography). And certain ML work can also be high-impact.

Regarding ME, I’d suggest reaching out to https://high-impact-engineers.github.io/

In general, people say that people with certain backgrounds are more of a bottleneck in priority areas than funding. (This does not mean that everyone is definitely a good fit for some high-impact direct work but rather,  should be viewed as one factor when prioritizng what to explore next). 

On your example, yes you could use money to fund more marginal candidates but if your fit is much better than marginal, then you’ll have much better chances of success so it’s worth running some cheap tests first unless entry to the field is already highly competitive.

Also, it’s interesting that you are comparing the researcher path with ~$100k income vs ML with 400-100=$300k. Does the ~3x factor feel about right to compensate for taking the EtG path instead of your default? (You may end up similarly enyojing the EtG work and happy to live on $100k or slightly dislike it and require more reward to compensate for that). 

I guess an alternative (probably more fortunate) framing is how much you'd be willing to pay to save your own life using money that you'd otherwise donate to high-impact causes. (assuming you'd have that much money available to allocate)

I'd argue that if they would pay their employees a competitive for-profit salary in this case, the employee's share of the input is 0. Maybe they aren't motivated by impact, that's why they took the full salary. They just "do their job" and produce value for their employer.

I’d push back on this. Maybe the career choice (decision to do the job at the given market rate instead of doing something else) of N employees on the market allows achieving the current total production at $2X total cost instead of, say $3X, if it was fully up to the market. (The decisions shifting the supply curve upwards so the market can buy more production at a given price)

In that case, the total funding required is $X less, so on average $X/N for each impact focused employee. For comparison, if the market consists of, say, another N impact agnostic employees (2N employees total), then the individual market rate salary was also $X/N. (Average funding need reduction comparable to the market salary.)

Now, the above numbers are obviously arbitrary, but I think this illustrates the effect that I (we?) hope to have from direct work.

I think location constraints (either visa or personal) should be an important factor for career choice. 80k recommends moving to a hub in general, but I think the importance varies across career paths. As a starting point, I'd try to factor such constraints into my exploration priority list. I.e., if I'm roughly equally excited about two paths overall and one fits my location constraints much better, then explore that path first.

I would definitely advise against doing a medical degree, just to then become an engineer in the US. I think it's not realistic to keep your motivation level sufficiently high.

However, it seems it's difficult to make impact to reducing AI risks outside of the USA/UK. All of the frontier AI companies and EA organzations(like CLR, MIRI) are mostly in the USA(or the UK), and it's hard to get a remote job opportunity.

Frontier companies are highly competitive; they can afford not to bother with remote working. For EA/AIS, I think most entry roles are similarly highly competitive, and for some roles, it's simply mandatory to be close to Tech/Policy hubs. But there are also some remote-first organizations, mostly ones that require scarce experienced profiles.

I think it's important to distinguish between the following two theories of change categories:

1. Doing important work that not enough people want to do
2. Doing important work differently from how others would do it

For #1, if the nature of the work itself does not require in-person presence, and there are no remote options, then maybe the current supply/demand gap is not as big as you originally thought, so the expected counterfactual impact might also be much lower. For #2, there is no such gap by definition, and filtering applicants for in-person presence is very likely.

Maybe we need to flip this around. Instead of tracking how much funding was allocated to a certain cause area, we should be tracking the expected marginal opportunity in each and comparing those. I.e., what was the expected result of a marginal $1M donated to each cause area on average in, say, a given year?

This does not incentivize for making the allocations secret since the decisions are made based on the current state of the "market" irrespective of any previous allocations.

Going back to the 100 fund managers example, I think I'd much prefer them to individually recognize that people preferring the alternative allocation are just as competent in the decision as they are (in an ideal case), and as a result, apply the uncertainty to their own preference (making it 75-25 instead of 100-0/50-50) rather than relying on an external mechanism.