[Link] 80,000 Hours Nov 2020 annual review

I've heard that in the past 80k has sometimes reached out to employees at EA organizations asking if they'd like to leave and work for 80k, including in instances in which those organizations felt that it would be highly costly to lose that employee.

I now see that Habiba has joined 80k, and I had previously heard that it was a major win for FHI when she was able to get her role at Oxford. Was it costly to FHI for her to leave, and if so, how do you balance poaching employees for 80k with trying to help and fill those same roles at those organizations?

I'm trying to make a rough estimate for the value of doing local career advice (and writing this while I think it through).

The leading metric used by 80k is DIPY (Discounted, Impact-adjusted Peak Years). This table estimates how many DIPYs per employee working on advising (not necessarily people doing the advising, measured as FTE: Full-Time Employee equivalent for a year). 

One DIPY is roughly the (time discounted) value we expect someone to produce each year during the most productive few decades (the ‘peak’) of their career if they’re about as promising as someone taking an effective altruist approach to improving the world who is currently in a position such as: 

• a postdoc focused on a top problem area at a top 30 world university, 
• one of the 20 most important operations roles in our top problem areas, 
• an ML PhD at a top 3 university with the aim of working on AI Safety or policy.

The results of 80k's career advice per year are:

• 2017: 10.5 DIPY/FTE
• 2018: 3.6    DIPY/FTE
• 2019: 4.0    DIPY/FTE

(2016 has 41.5 DIPY/FTE! But they only write 0.2 FTE for that year, compared to 2.5-3 on later years while they had about 100 advise calls rather than about 250 in later years). 

2017 is addressed as an interesting anomaly, with possible explanations given here. They write

Our guesses as to why:

1. Web engagement hours in 2017 were 6x higher than in 2015, which may have let us reach a new audience and uncover low-hanging fruit.
2. Peter McIntyre (predominant 2017 advisor) might be unusually effective.
3. In 2017, advising was more forceful in encouraging AI safety work, which might have driven more plan changes. The downside is that we probably pushed some people too hard, as reported in our 2018 annual review.
4. We wrote the career guide and some of our most popular articles in 2015–2016 but didn’t generate large amounts of traffic until 2017. 2017 website plan changes therefore likely relied particularly heavily on work from earlier years. 
5. The apparent pattern could also be partly due to errors in our estimates.

These numbers are good, but not astoundingly good, so in total it makes sense to me even if less than I'd hope for. Anything greater than 1 DIPY/FTE should be considered a success if 80k employees are valued at 1 DIPY/year (although both Michelle and Habiba are potentially valued more than that). 

So now we need to understand how local career advice compares to 80k's. 

From this section of the report:

The aim of the one-on-one team is to:

1. Identify the most promising new readers who haven’t yet engaged in person.
2. Activate them by providing — over the course of 1–2 calls and follow-up emails — (i) introductions to people and jobs, (ii) encouragement (e.g. making it clear that EA has a place for them, or that they have a good career plan), and (iii) basic checks on their plan (e.g. adding options, recommending further reading).

...

We analysed recent plan changes and concluded that the top three value adds are (in order): (i) introductions to people and jobs, (ii) encouragement (as defined above), and (iii) basic checks on their plan and information (as defined above).

[Ben/Michelle, is that analysis available?]

Let's see how these three metrics make sense for local groups. I expect introductions to be of much lower value (as the available network is likely much smaller), encouragement to potentially be of about an equal value (this can, and should, be practiced and done well), and basic checks to be somewhat less (as this demands skill and a good framework - this probably varies a lot between groups). Also, I expect 80k to appear to the advisees as higher status compared to local groups, which probably gives more weight to their advice.

First, this is prescriptive for local groups. They should 

1. Make effort to expand their network. 
2. Make more effort into making introductions. Even if the advisor is not familiar with anyone relevant, they should help promising people with reaching out to people in the broad EA network (cold emails or sending a direct message on EAHub should work well most of the time).
3. Become more familiar with top local job opportunities.
4. Be more encouraging, and practice doing it well. 
5. Recommend them to also apply for 80k coaching (or AAC / Effective Thesis).

Also, I recommend this guide for conducting career consultation.

Furthermore, in Appendix A 80k offers two "alternative visions of advising":

EA welcoming committee

• Minimal selection and prep, and speak to a larger number of people
• Focus on being encouraging and welcoming; de-emphasise providing specific advice

EA mentoring

• Work with a smaller number of people but have lots of meetings until they’re fully up-to-speed with EA.

For local groups it might make sense to focus on the first when doing career advice. I think it could be harder to recognize the most promising people and target them, and it requires more EA-expertise to do mentoring. More importantly, I think that local groups in general should be welcoming and less elitist, and giving career advice services is a great way to do that (and signal the relevant audience). That said, I do think that doing "EA mentoring" is very important, and in fact we are also doing something similar in EA Israel and it seems to be successful, I just recommend that in addition to the career advice.

Secondly, this makes me expect that overall the career consultation services done locally can perform quite well compared to 80k. Say, have 2-10 times less value with most of the loss from a poor network. However, I also expect local groups to have much more overhead because they aren't as skilled and have poorer infrastructure. Say, they take a lot of time to prepare and write a followup email / 1-1 sessions take longer to be fruitful / more time spent building network to make introductions. So all in all I guess something like 4-30 times less effective.

However, the people that apply for local career consultation are of different demography. They are probably less promising than people who get advised by 80k (at least within longtermism) and might be more closely related to people in the local group (which could help with setting long-term accountability but might have some complicated dynamics). 

I think that this probably lowers the total efficiency substantially (say 2-3 times, but could be much more). So in total, I guess local career advice to be about 10-100 times less effective compared to 80k coaching. This is highly uncertain. I'd like to know a lot more about 80k's process and the analysis of how different factors contribute to its success (and how dependant that is on the advisor). Also, there might be more important indicators of success, like potentially: involvement with the local group / EA community or having a better understanding of EA and feelings toward it.

If that's the case, that still feels like a good use of the local group resources. I expect most people to have a hard time doing work that's 10-100 times as good as Michelle and Habiba can do. Also, giving career advice for some (most?) people is a highly rewarding experience by itself and could be a valuable learning experience for people early on in their careers. At least for me, (and at least at this time in my life where I'm recovering from burnout), it seems to take less energy than doing other work.