They look at people who switch into entrepreneurship, and then consider the whole sample, so I don't think survivorship bias is a big issue.

I agree it would be good to go more fine grained in the categories though.

Yes - they try to address that in the paper, but it's hard.

We totally agree - we go into that in the post e.g. see the conclusion section.

We discuss this somewhat in the post - the line at which you say something "works" is arbitrary. We explore how the choice of definition changes the stat.

Some of it is that, but much less so for the meta-analyses. We have a bunch more comments about this in the doc.

Hi all, I'm the author of the post and happy to take questions.

We think even if AGI is only invented in 50 years, then it's still worth doing some work on the control problem today, so our view isn't that dependent on whether we're currently in an AI investment bubble or not.

However, if you did really believe the hype, and think that ML will lead to major breakthroughs in the next decade or so, then that would make it even more urgent.

We're going to work on the causes that we think our most pressing, so to be consistent, we're going to end up recommending the same things we work on.

At 80,000 Hours, our focus is also not on giving money. We're trying to help people find jobs in these areas, often in academia or policy and many other sectors.

There's also real research progress being made. Here are some of the issues people are working on:

https://arxiv.org/abs/1606.06565

More info on the pros and cons of water:

http://www.copenhagenconsensus.com/publication/post-2015-con...

http://blog.givewell.org/2016/05/03/reservations-water-quali...