Thanks for the update! It is always good to get a better sense of the decisions behind the grants my (tiny) donations have been funding. Some questions:
Do you have a sense of which fraction of the grant evaluations involve some sort of estimation (namely a back-of-the-envelope-calculation)?
What is the break-down of funding and grants by cause area? I see most is going to AI risk, but it would be nice to have specific numbers.
Relatedly, which fraction of applications not focussing on AI nor bio are accepted?
Do you have a sense of which fraction of the grant evaluations involve some sort of estimation (namely a back-of-the-envelope-calculation)?
I don’t have a great sense of this. It’s down to the individual fund manager evaluating the grant to decide whether they should do these kinds of calculations. If I had to guess, I would say fund managers do explicit BOTECs on 5-20% of grants. Note that a fund manager is often assigned similar grants, so it may not make sense for a fund manager to repeat a BOTEC at the individual grant level (but they will analyse the grant using other methods).
What is the breakdown of funding and grants by cause area? I see most is going to AI risk, but it would be nice to have specific numbers.
I don’t have a good breakdown of this right now, but I am hoping to release so more information on the breakdown in future. I am also not sure what fraction of grants not related to AI or bio are accepted.
I’ve done BOTEC’s for a lot of classes of grants (independent researchers, various community-building interventions, websites, etc.), which informed a lot of my broad thinking about the relevant type of intervention, and then I mostly apply rough heuristics for individual grants.
Note that a fund manager is often assigned similar grants, so it may not make sense for a fund manager to repeat a BOTEC at the individual grant level (but they will analyse the grant using other methods).
This agrees with my experience. Speaking personally, I do (often implicit) head-to-head comparisons more often than BOTEC. I have a rough sense of what the “breakeven” grant is, so the algorithm I most often run is more-or-less “if this grant costs X times the breakeven grant, would it be better in expectation for there to be X•C grants similar to the breakeven grant, or C grants similar to this one?” (where C is chosen so that I’m dealing with whole numbers).
Thanks for the update! It is always good to get a better sense of the decisions behind the grants my (tiny) donations have been funding. Some questions:
Do you have a sense of which fraction of the grant evaluations involve some sort of estimation (namely a back-of-the-envelope-calculation)?
What is the break-down of funding and grants by cause area? I see most is going to AI risk, but it would be nice to have specific numbers.
Relatedly, which fraction of applications not focussing on AI nor bio are accepted?
I don’t have a great sense of this. It’s down to the individual fund manager evaluating the grant to decide whether they should do these kinds of calculations. If I had to guess, I would say fund managers do explicit BOTECs on 5-20% of grants. Note that a fund manager is often assigned similar grants, so it may not make sense for a fund manager to repeat a BOTEC at the individual grant level (but they will analyse the grant using other methods).
I don’t have a good breakdown of this right now, but I am hoping to release so more information on the breakdown in future. I am also not sure what fraction of grants not related to AI or bio are accepted.
I’ve done BOTEC’s for a lot of classes of grants (independent researchers, various community-building interventions, websites, etc.), which informed a lot of my broad thinking about the relevant type of intervention, and then I mostly apply rough heuristics for individual grants.
This agrees with my experience. Speaking personally, I do (often implicit) head-to-head comparisons more often than BOTEC. I have a rough sense of what the “breakeven” grant is, so the algorithm I most often run is more-or-less “if this grant costs X times the breakeven grant, would it be better in expectation for there to be X•C grants similar to the breakeven grant, or C grants similar to this one?” (where C is chosen so that I’m dealing with whole numbers).