Really well constructed argument and guide. I have long felt very strongly that strategic thinking on how to policy entrepreneur on this topic is lacking.
I would suspect that the multiple quoted for donations to individual politicians will diminish as broader AI concerns become more salient to the average voter and coalitions form. Some intrepid political scientist should track that.
AI safety organizations are capacity constrained by a lack of senior researchers who can mentor and supervise junior talent. This bottleneck drives hyper-selective hiring, blocks promising candidates from entering the field, and is being exacerbated by AI automation, which is raising the bar for human contribution rather than lowering it. Based on 23 interviews with hiring managers, research leads, and funders across the ecosystem, we identify six concrete actions field-builders can take to improve talent pipelines, including standardized mentor evaluations, direct policy embedding, and production codebase access.
The thing is that even if this was the case, and money was infinite, you can give money to junior researchers with reasonable proposals and some will figure it out by themselves and become senior without help. After all at some point in history senior alignment researchers didn't exist. This might be a "sharp curve" in the utility/money function, which can make us reconsider what to fund. However you can still in fact make an impact by adding money by e.g funding independent junior researchers. However I'm not 100% that what you're mentioning is the case, because it is my impression that finishing MATS with a good project, that implies a capacity for independent research doesn't guarantee get hired full time due to funding constraints.
I agree that having more junior independent researchers does have some positive impact, especially if they've already done MATS. But as you say, we should reconsider what to fund.
For individual donors, I agree with the piece and especially believe in individual donors funding AI Safety-adjacent political candidates is good for US citizens. This is especially true since
1. you get more favorable rates and can work directly with candidates on initial direct donations compared to going through a Super PAC
2. big donors like Coefficient Giving generally stray away from funding political candidates since they worry reputational risks will hurt their causes.
"But as you say, we should reconsider what to fund.". One thing I forgot to mention: What if we were always on this "limited by senior researchers" cliff, thus there's no cliff at all. I'm not sure.
The cliff would be true if 2023-2024 MATS was about getting junior researchers to make teams around established seniors. And then this theory of change broke due to automation/lack of seniors.
However I agree with the advice for individual donors, nowadays the AI safety funding is probably dominated by corporate interests/interests about donating-to-non-political-organizations-only. which makes the counterfactual impact of individual donors less significant.
Impressively comprehensive
thank you!!
Really well constructed argument and guide. I have long felt very strongly that strategic thinking on how to policy entrepreneur on this topic is lacking.
I would suspect that the multiple quoted for donations to individual politicians will diminish as broader AI concerns become more salient to the average voter and coalitions form. Some intrepid political scientist should track that.
Not reading that essay..
what the hell r u doing here
I think your logic about funding-constrained makes some theoretical sense but I don't think it's right. According to a recent post by a MATS community programs officer, AI Safety is indeed talent constrained specifically of senior researchers (https://forum.effectivealtruism.org/posts/jwwrC4n9H53doRjRH/ai-safety-talent-needs-in-2026-insights-for-field-building). Quoting directly:
"""
TL;DR
AI safety organizations are capacity constrained by a lack of senior researchers who can mentor and supervise junior talent. This bottleneck drives hyper-selective hiring, blocks promising candidates from entering the field, and is being exacerbated by AI automation, which is raising the bar for human contribution rather than lowering it. Based on 23 interviews with hiring managers, research leads, and funders across the ecosystem, we identify six concrete actions field-builders can take to improve talent pipelines, including standardized mentor evaluations, direct policy embedding, and production codebase access.
"""
The thing is that even if this was the case, and money was infinite, you can give money to junior researchers with reasonable proposals and some will figure it out by themselves and become senior without help. After all at some point in history senior alignment researchers didn't exist. This might be a "sharp curve" in the utility/money function, which can make us reconsider what to fund. However you can still in fact make an impact by adding money by e.g funding independent junior researchers. However I'm not 100% that what you're mentioning is the case, because it is my impression that finishing MATS with a good project, that implies a capacity for independent research doesn't guarantee get hired full time due to funding constraints.
I agree that having more junior independent researchers does have some positive impact, especially if they've already done MATS. But as you say, we should reconsider what to fund.
For individual donors, I agree with the piece and especially believe in individual donors funding AI Safety-adjacent political candidates is good for US citizens. This is especially true since
1. you get more favorable rates and can work directly with candidates on initial direct donations compared to going through a Super PAC
2. big donors like Coefficient Giving generally stray away from funding political candidates since they worry reputational risks will hurt their causes.
"But as you say, we should reconsider what to fund.". One thing I forgot to mention: What if we were always on this "limited by senior researchers" cliff, thus there's no cliff at all. I'm not sure.
The cliff would be true if 2023-2024 MATS was about getting junior researchers to make teams around established seniors. And then this theory of change broke due to automation/lack of seniors.
However I agree with the advice for individual donors, nowadays the AI safety funding is probably dominated by corporate interests/interests about donating-to-non-political-organizations-only. which makes the counterfactual impact of individual donors less significant.