It looks like Open Phil’s approach to this is to evaluate all programs (neartermist and longtermist) against cash transfers (they use an internal ‘unit of impact’, I think, to try to compare all programs). As I understand it, any program funded by Open Phil needs to beat this standardised bar—it isn’t that they actually believe longtermist projects are much more impactful but still grant to neartermist causes for political reasons or similar. Or, to put it another way, the bar in neartermist funding isn’t lower than it is in longtermist funding.
Accordingly, they wouldn’t necessarily reallocate funds from one place to another in advance, but if the new longtermist applications seem to be more impactful in expectation that the neartermist ones, they might choose to fund more longtermist programs than neartermist ones going forward.
For what it’s worth, things like the GiveWell charities actually perform extraordinarily well in this analysis, so my prior is that FTX-funded projects won’t outperform them significantly in Open Phil’s evaluation, and so won’t lead to a reallocation of funding.
A) If we call the quality threshold for projects T, T_G = T_L = T. In addition, project-funding is allocated by whether, for project p, p > T... ...and G vs L allocations are determined by how many projects are (anticipated to) pass this threshold from each category.
B) With OP = all projects p such that p was not funded by FTX and is likely to seek Open Philanthropy funding, and L_OP being the longtermist subset of OP, and L_FTX = all L such that L was funded by FTX and now seeks OP funding:
sum_of_funding(p in L_OP such that ((p > T) and (p > x for all x in L_FTX))) >= current_OP_longtermist_funding.
First of all, OP funding won’t currently be reallocated to FTX projects, because better non-FTX projects crowd them out of OP’s current allocation for this area. And also:
count(p in OP such that ((p > T) and (p > x for all x in G_FTX))) is large -> count(L_OP > T)/(cout(OP > T)) ~= count((L_OP +FTX) > T)/count((OP + FTX) > T)
Second of all, since there are so many projects applying to OPhil which are better than T AND better than the FTX ones... ...the addition of the FTX projects shouldn’t significantly affect the proportion of ‘good enough’ projects which are longtermist... ...meaning that long-term funding allocations between global-health vs longtermist cause-categories also shouldn’t be affected.
Of course, I know very little about this topic, having come in from the LessWrong side of things, so I can’t comment as to the accuracy of the above claims.
It looks like Open Phil’s approach to this is to evaluate all programs (neartermist and longtermist) against cash transfers (they use an internal ‘unit of impact’, I think, to try to compare all programs). As I understand it, any program funded by Open Phil needs to beat this standardised bar—it isn’t that they actually believe longtermist projects are much more impactful but still grant to neartermist causes for political reasons or similar. Or, to put it another way, the bar in neartermist funding isn’t lower than it is in longtermist funding.
Accordingly, they wouldn’t necessarily reallocate funds from one place to another in advance, but if the new longtermist applications seem to be more impactful in expectation that the neartermist ones, they might choose to fund more longtermist programs than neartermist ones going forward.
For what it’s worth, things like the GiveWell charities actually perform extraordinarily well in this analysis, so my prior is that FTX-funded projects won’t outperform them significantly in Open Phil’s evaluation, and so won’t lead to a reallocation of funding.
I’ve read this comment a few times, and I don’t understand what it is saying.
Just to make it more concrete, suppose that the estimated cost-effectiveness for global health (G) and speculative longtermist (L) projects look like:
For example, this could look like:
Then what are you saying applies to this list of marginal impacts?
If I may:
I interpreted Jack as saying that...
A) If we call the quality threshold for projects T, T_G = T_L = T.
In addition, project-funding is allocated by whether, for project p, p > T...
...and G vs L allocations are determined by how many projects are (anticipated to) pass this threshold from each category.
B) With OP = all projects p such that p was not funded by FTX and is likely to seek Open Philanthropy funding, and L_OP being the longtermist subset of OP,
and L_FTX = all L such that L was funded by FTX and now seeks OP funding:
sum_of_funding(p in L_OP such that ((p > T) and (p > x for all x in L_FTX)))
>=
current_OP_longtermist_funding.
First of all, OP funding won’t currently be reallocated to FTX projects, because better non-FTX projects crowd them out of OP’s current allocation for this area. And also:
count(p in OP such that ((p > T) and (p > x for all x in G_FTX))) is large
->
count(L_OP > T)/(cout(OP > T)) ~= count((L_OP +FTX) > T)/count((OP + FTX) > T)
Second of all, since there are so many projects applying to OPhil which are better than T AND better than the FTX ones...
...the addition of the FTX projects shouldn’t significantly affect the proportion of ‘good enough’ projects which are longtermist...
...meaning that long-term funding allocations between global-health vs longtermist cause-categories also shouldn’t be affected.
Of course, I know very little about this topic, having come in from the LessWrong side of things, so I can’t comment as to the accuracy of the above claims.