Thanks for writing up this piece, Jack! I think this is one of the main questions we need to answer if we want all the money that might come online during the next few years to create real impact.
On the one hand, obviously CG is an exceptional place - the fact that they were able to identify future AI risks and engineered pandemics as top priorities more than a decade ago, when ~no other major philanthropic players were covering those topics, looks near-prophetic in hindsight. Animal welfare is another unusual bet where it looks like others might follow them in years to come, and even within more established fields like global health, they have often outperformed peers (hence the meme "Open Philanthropy strikes again"). So there's a clear case why new donors might defer to them as the default experts across many of these domains.
Even so, I agree with you that I think these philanthropic "markets" can benefit from more diversity of philanthropic intermediaries (I'm using this to refer to anyone sitting between the donor and the grantee, including both philanthropic advisors and also fund managers), and I agree with many of your reasons for it. Some considerations that are decisive to me:
- It looks like *founders* are going to be the binding constraint on many of the relevant philanthropic cause areas (e.g., AI governance, biosecurity, and nuclear stability - areas where I am currently working). So we should design our ecosystem by what would attract more and better founders. A more founder-friendly ecosystem would give top founders more leverage, translating in better terms (e.g., faster grant decisions, longer runways, clearer pathways to scaling up through future grants, and fewer strings attached with grants). A simple way to do this is by creating multiple separate funding sources which would then compete over the best founders.
- I think the simplest way to think of this is, if you believe something is good and that there should be a lot of it (with "it" e.g., being biodefense projects like PPE stockpiling and air purification technologies, or some other public good that the new donors are interested in) , then many independent decision makers should have the power to make it happen. There's a reason why the Renaissance happened among competing Italian city states, why the Industrial Revolution happened among competing UK merchants, and why most current frontier technologies are built among competing VCs in the Bay Area; that's just reflecting where founders have the easiest time finding a patron for their project.
- I think I emphasize speed and agility a bit more than you do. The world is going to change fast in the coming years, and grantmaker strategies will likely need to change with it. An ecosystem of independent decision makers can evolve at the speed of the fastest, most foresightful funder (though who claims that mantle will change over time). A monolithic ecosystem moves at the speed of the slowest bureaucratic procedure in the monolith.
- On the unilateralist curse, this is something I've also heard mentioned as an objection against more diversity in the field. As you say, many of the candidates for "Coefficient Giving 2.0" or 3.0 are existing players in the field that are well known to CG, and I don't think they have a track record of being significantly more reckless, nor that CG would have a hard time convincing them to be cautious in the most risky domains of their portfolio. This seems like an entirely solvable problem. I also think that as a philanthropic grantmaker, most of your portfolio should ideally not be so dual use/ knife-edge that a small mistake risks large harm; if that's the case, I think the portfolio is emphasizing the wrong things, and you should look for more robustly benign projects.
- On donation timing, I agree that it is a hard question. But I think it is bad if the question is decided by operational constraints, instead of value tradeoffs between current donations and future donations. *If* donors want to give now, there should be a way to deploy the resources now. The default deployment pathways look like they will saturate, which means now is a time to get creative about that. And personally I lean in favor of haste over patience - there's almost no historical issue (e.g.,, slavery) where people tend to think "that was solved too fast, the world should have waited a bit longer". That's probably a good heuristic today, as well. And for some issues, like AI safety, it looks like we are racing against a clock that's ticking very quickly.
- Additionally, more diversity in intermediaries would also give the donors themselves more "consumer choice". A broader class of intermediaries with a credible track record of scope-sensitive prioritization within these areas would let donors pick the intermediaries that are the best fit with their own values and theories of change. As such it reduces the market power of intermediaries in favor of donors.
Finally, a note on CoIs: I am currently a program manager at a smaller philanthropic advisory shop in this space, so my views are clearly colored by this and should be discounted accordingly (though the flip side of this is that I would also stand to lose market power from a more competitive intermediary market).
Your point on "seek robustly benign things" is pretty interesting -- certainly if we think there are lots of impactful and robustly benign areas, then we should let a thousand flowers bloom. But I'm not sure we're that lucky.
Biodefense tech like PPP or UVC is a great example of an area where I'm generally quite happy for whoever to go out and build it, and the downside is mostly limited to "the funding was useless" or "people think the field is a little dumb", rather than having potential for a large negative impact.
But I can think of intervention areas that seem important but pretty necessarily fraught -- e.g. most things in AI policy, some things in AI security (like oversight tech), and safety. Overall I think I'd be happy to pull a switch that increased funding & distributed decisionmaking in those areas, because I think the EV is fairly robustly positive, but I do think I would expect some real backfires among the top 100 projects I'd fund.
Other arguments that push harder against the robustness heuristic in various directions, and that I put _some_ weight on:
- We're in a losing situation by default, so it's worth taking high-variance bets
- We should be OK doing some things that have a serous risk of negative consequences, but are clearly the right thing from a deontological/higher-order-consequentialism POV
I think this is pretty much correct (also your mild pushback). And if CG was worried about that and thought they had exceptionally good procedures for handling such risks, they should just double down on those areas and free up space for others to help with the more benign stuff. But I also feel very unsure about the level of risk management that would be needed in the dual use areas and if CG really is set up to deliver on that.
FWIW, biosecurity used to be the #1 example of "this is super dual use, it's dangerous to bring in more people into the inner circle" and then that just entirely flipped once good roadmaps were in place and it became possible to bring in operators who didn't need the full context. So this is also an argument in favor of bringing other cause areas to a higher level of maturity so that it's safe to bring in more collaborators (it doesn't seem like "level of dual use risk" is a fundamental property of a cause area, rather it's a function of multiple components, including the maturity of the field)
Hey Jack, that's an excellent read thanks for writing it! I've actually been trying to quantify this from the AI-safety side (i.e. it's not a forecast of the whole Funding Anthropalypse).
I built a scenario model that separates a few things:
The result supports your thesis around capacity-mismatch, but with some nuance. In the base case, the annual funding target goes from $0.22B today to $3.05B in 2028, while near-term responsible deployment capacity reaches only $0.92B.
That means it creates a $2.13B mismatch in 2028, but the model’s capacity curve catches up by 2031. Then the main point may be "capital could arrive several years before the infrastructure needed to allocate and deploy it well".
The absorption layer is definitely the least certain part of the model. If you've made some thinking here, would love to get your views. Does a roughly $0.9B near-term ceiling look directionally correct, or is the model bottlenecking the wrong layer?
Really enjoyed this. One observation from the science side. In translational research, what often seems to hold things back isn't the number of funders so much as how few projects are actually ready to fund. Grants tend to pay for research without turning it into the kind of de-risked project a next funder can pick up. So I wonder whether more funders, on their own, would end up putting more money on the same small pile of ready projects rather than creating new ones. Might just be that too few funders and too few fundable projects are two separate problems. Curious whether that matches what others are seeing.
Thanks Joe. I do worry about this, yes. It doesn't necessarily create diversity of thought if a second foundation hires from the same pool as the first.
That being said, it does help some of the issues of power and funding concentration, and in time it should lead to more diversity, as two foundations would have a higher chance of hiring people from outside the movement, making different hiring decisions, targeting different talent, developing different reputations etc.
Thanks for writing up this piece, Jack! I think this is one of the main questions we need to answer if we want all the money that might come online during the next few years to create real impact.
On the one hand, obviously CG is an exceptional place - the fact that they were able to identify future AI risks and engineered pandemics as top priorities more than a decade ago, when ~no other major philanthropic players were covering those topics, looks near-prophetic in hindsight. Animal welfare is another unusual bet where it looks like others might follow them in years to come, and even within more established fields like global health, they have often outperformed peers (hence the meme "Open Philanthropy strikes again"). So there's a clear case why new donors might defer to them as the default experts across many of these domains.
Even so, I agree with you that I think these philanthropic "markets" can benefit from more diversity of philanthropic intermediaries (I'm using this to refer to anyone sitting between the donor and the grantee, including both philanthropic advisors and also fund managers), and I agree with many of your reasons for it. Some considerations that are decisive to me:
- It looks like *founders* are going to be the binding constraint on many of the relevant philanthropic cause areas (e.g., AI governance, biosecurity, and nuclear stability - areas where I am currently working). So we should design our ecosystem by what would attract more and better founders. A more founder-friendly ecosystem would give top founders more leverage, translating in better terms (e.g., faster grant decisions, longer runways, clearer pathways to scaling up through future grants, and fewer strings attached with grants). A simple way to do this is by creating multiple separate funding sources which would then compete over the best founders.
- I think the simplest way to think of this is, if you believe something is good and that there should be a lot of it (with "it" e.g., being biodefense projects like PPE stockpiling and air purification technologies, or some other public good that the new donors are interested in) , then many independent decision makers should have the power to make it happen. There's a reason why the Renaissance happened among competing Italian city states, why the Industrial Revolution happened among competing UK merchants, and why most current frontier technologies are built among competing VCs in the Bay Area; that's just reflecting where founders have the easiest time finding a patron for their project.
- I think I emphasize speed and agility a bit more than you do. The world is going to change fast in the coming years, and grantmaker strategies will likely need to change with it. An ecosystem of independent decision makers can evolve at the speed of the fastest, most foresightful funder (though who claims that mantle will change over time). A monolithic ecosystem moves at the speed of the slowest bureaucratic procedure in the monolith.
- On the unilateralist curse, this is something I've also heard mentioned as an objection against more diversity in the field. As you say, many of the candidates for "Coefficient Giving 2.0" or 3.0 are existing players in the field that are well known to CG, and I don't think they have a track record of being significantly more reckless, nor that CG would have a hard time convincing them to be cautious in the most risky domains of their portfolio. This seems like an entirely solvable problem. I also think that as a philanthropic grantmaker, most of your portfolio should ideally not be so dual use/ knife-edge that a small mistake risks large harm; if that's the case, I think the portfolio is emphasizing the wrong things, and you should look for more robustly benign projects.
- On donation timing, I agree that it is a hard question. But I think it is bad if the question is decided by operational constraints, instead of value tradeoffs between current donations and future donations. *If* donors want to give now, there should be a way to deploy the resources now. The default deployment pathways look like they will saturate, which means now is a time to get creative about that. And personally I lean in favor of haste over patience - there's almost no historical issue (e.g.,, slavery) where people tend to think "that was solved too fast, the world should have waited a bit longer". That's probably a good heuristic today, as well. And for some issues, like AI safety, it looks like we are racing against a clock that's ticking very quickly.
- Additionally, more diversity in intermediaries would also give the donors themselves more "consumer choice". A broader class of intermediaries with a credible track record of scope-sensitive prioritization within these areas would let donors pick the intermediaries that are the best fit with their own values and theories of change. As such it reduces the market power of intermediaries in favor of donors.
Finally, a note on CoIs: I am currently a program manager at a smaller philanthropic advisory shop in this space, so my views are clearly colored by this and should be discounted accordingly (though the flip side of this is that I would also stand to lose market power from a more competitive intermediary market).
Your point on "seek robustly benign things" is pretty interesting -- certainly if we think there are lots of impactful and robustly benign areas, then we should let a thousand flowers bloom. But I'm not sure we're that lucky.
Biodefense tech like PPP or UVC is a great example of an area where I'm generally quite happy for whoever to go out and build it, and the downside is mostly limited to "the funding was useless" or "people think the field is a little dumb", rather than having potential for a large negative impact.
But I can think of intervention areas that seem important but pretty necessarily fraught -- e.g. most things in AI policy, some things in AI security (like oversight tech), and safety. Overall I think I'd be happy to pull a switch that increased funding & distributed decisionmaking in those areas, because I think the EV is fairly robustly positive, but I do think I would expect some real backfires among the top 100 projects I'd fund.
Other arguments that push harder against the robustness heuristic in various directions, and that I put _some_ weight on:
- We're in a losing situation by default, so it's worth taking high-variance bets
- We should be OK doing some things that have a serous risk of negative consequences, but are clearly the right thing from a deontological/higher-order-consequentialism POV
I think this is pretty much correct (also your mild pushback). And if CG was worried about that and thought they had exceptionally good procedures for handling such risks, they should just double down on those areas and free up space for others to help with the more benign stuff. But I also feel very unsure about the level of risk management that would be needed in the dual use areas and if CG really is set up to deliver on that.
FWIW, biosecurity used to be the #1 example of "this is super dual use, it's dangerous to bring in more people into the inner circle" and then that just entirely flipped once good roadmaps were in place and it became possible to bring in operators who didn't need the full context. So this is also an argument in favor of bringing other cause areas to a higher level of maturity so that it's safe to bring in more collaborators (it doesn't seem like "level of dual use risk" is a fundamental property of a cause area, rather it's a function of multiple components, including the maturity of the field)
Two Gates Foundations at half the size would have probably been way way better than what we have now.
It’s awesome to see how Founder’s Pledge often does CEAs in a slight different way than GiveWell—and I think that’s such a good thing.
An ecosystem of big thinkers is so much better than just 1-2.
I’m so glad that Ren Phil exists.
The tension between the Big Bang/Mulago/Skoll world and the Charity Entrepreneurship/Givewell world has been really productive.
I think this is great.
Hey Jack, that's an excellent read thanks for writing it! I've actually been trying to quantify this from the AI-safety side (i.e. it's not a forecast of the whole Funding Anthropalypse).
I built a scenario model that separates a few things:
paper wealth -> liquidity -> philanthropic-vehicle conversion -> annual payout -> strict AI-safety allocation -> responsible absorption capacity
The result supports your thesis around capacity-mismatch, but with some nuance. In the base case, the annual funding target goes from $0.22B today to $3.05B in 2028, while near-term responsible deployment capacity reaches only $0.92B.
That means it creates a $2.13B mismatch in 2028, but the model’s capacity curve catches up by 2031. Then the main point may be "capital could arrive several years before the infrastructure needed to allocate and deploy it well".
Model is here in case that's of interest: https://docs.google.com/spreadsheets/d/1-OIRMdtdApgawftL2xT4QbhBqMuETkf1k1p9Mte4v98/edit?gid=929545455#gid=929545455
The absorption layer is definitely the least certain part of the model. If you've made some thinking here, would love to get your views. Does a roughly $0.9B near-term ceiling look directionally correct, or is the model bottlenecking the wrong layer?
Really enjoyed this. One observation from the science side. In translational research, what often seems to hold things back isn't the number of funders so much as how few projects are actually ready to fund. Grants tend to pay for research without turning it into the kind of de-risked project a next funder can pick up. So I wonder whether more funders, on their own, would end up putting more money on the same small pile of ready projects rather than creating new ones. Might just be that too few funders and too few fundable projects are two separate problems. Curious whether that matches what others are seeing.
Really great piece. Appreciate your writing it and giving voice to this train of thought.
Along these same lines, do you worry about ideological and personnel overlap of this second hypothetical CG equivalent?
Thanks Joe. I do worry about this, yes. It doesn't necessarily create diversity of thought if a second foundation hires from the same pool as the first.
That being said, it does help some of the issues of power and funding concentration, and in time it should lead to more diversity, as two foundations would have a higher chance of hiring people from outside the movement, making different hiring decisions, targeting different talent, developing different reputations etc.