I am a generalist quantitative researcher. I am open to volunteering and paid work. I welcome suggestions for posts. You can give me feedback here (anonymously or not).
It accounts forkey variables at play in determining and comparing the cost-effectiveness of different interventions, including:
How to weigh the welfare of humans against that of non-human animals and that of different non-human animals between themselves (moral weights)
How to weigh different time periods, like the near future (0-5 years) versus the very distant future (over 500 years), e.g., in light of deep uncertainty about far future effects (Discount factors across time periods)
Attitudes towards risk
Diminishing returns
The first attempt to include the possibility of AI transforming the world so radically that any interventions not aimed at altering the trajectory of AI would be rendered ineffective (under improvement)
On diminishing returns, why do you set the (marginal) cost-effectiveness of funding GiveWell's All Grants Fund to the cost-effectiveness of the grants GiveWell made in 2025 instead of GiveWell's cost-effectiveness bar?
We used this information [about the grants GiveWell made in 2025] to construct a distribution for the average cost-effectiveness of GiveWell’s entire portfolio of spending in 2025 in Python, which we’ll assume is roughly equal to the marginal cost-effectiveness of GiveWell spending.
I think GiveWell's current cost-effectiveness bar is 6 times their benchmark. From the page I just linked which was last updated in September:
As of May 2026, our bar for funding programs is 6x our benchmark
You assumed a significantly higher cost-effectiveness.
Hi Dan. I expect terrorist attacks to be super unlikely to cause human extinction in the absence of very strong evidence suggesting otherwise. For the attacks killing more than 10 people among the 202 k listed on the Global Terrorism Database (GTB), the fraction killing more than 10*N people is around 1 % of the fraction killing more than N people. This suggests a Pareto distribution with a tail index of 2. Around 0.001 % of attacks killed more than 1 k people. Assuming the same Pareto distribution (straight line fitted to the section of the graph below with more than 10 deaths), more than 1 billion deaths would be 10^-12 (= 0.01^(9 - 3)) times as likely as more than 1 k deaths. So the Pareto distribution implies only 10^-17 (= 10^(-5 - 12)) of terrorist attacks cause more than 1 billion deaths. I think the actual risk is way higher. At the same time, I believe a significant risk of a terrorist attack causing human extinction, like 1 % over the next 20 years, would depart a lot from the most natural uninformed ways of extrapolating past data. A pandemic causing human extinction looks much more plausible based on past data (although I still guess the risk of human extinction in the next 20 years is much lower than 1 %).
Hi Liam. Nice post. You may be interested in this linkpost for the article "The Conscious Nematode: Exploring Hallmarks of Minimal Phenomenal Consciousness in Caenorhabditis Elegans".
And given the sheer number of soil nematodes — some researchers estimate there may be over 10^20 of them, or more than a million million million nematodes! — that means their aggregate welfare could be extremely large, possibly larger than all other animals on Earth.
In this post, "welfare range as a fraction of that of humans" means "individual welfare per fully-healthy-animal-year as a fraction of the welfare per fully-healthy-human-year".
I don't think you can extrapolate standard population models past the invention of artificial wombs and robo-nannies. The "Recovery" premise is much more plausible if creating new lives is less personally costly.
Why would artificial wombs and robo-nannies have very different effects from other innovations that have decreased the time cost of having children (for example, washing machines), or increased the income of prospective parents (for example, increasing income of women)? A lower cost of having children as a fraction of available income would increase fertility if the cost of everything else remained the same, but alternatives would become cheaper too relative to available income. It is unclear to me whether your point strengthens or weakens the conclusions of Alexandrie and Eden (2025) under the population models they analysed.
Drone replacements for pollination services. Bring down the cost of micro-drones that can pollinate almond trees down to be cost-competitive with bee-based pollination services, potentially ending an industry that harms billions of bees each year.
Are you assuming the bees would prefer not existing? I suppose this is very uncertain.
Of course, one could always try to argue on the merits that standard EA views on both global poverty and AI safety are empirically mistaken. That’s fine. All I’m saying here is that one had better not endorse higher-level heuristics, applied prior to any engagement with the specific details, that rule out both rigorously-supported global health charities (simply for being insufficiently ambitious) and speculative longtermist longshots (simply for being insufficiently rigorous). After all, if you prioritize rigor, the prima facie case for global health charities is going to be hard to beat. And if you prioritize reach (potential impact), nobody else is even in the same ballpark as the longtermists.
Are you confident the longtermist interventions you have in mind, like work on AI safety, are more cost-effective accounting for longterm effects than global health interventions? Decreasing the risk of human extinction may be less cost-effective than global health interventions accounting for longterm effects hold under standard population models. So it is unclear to me whether there is the rigor-reach dilemma you describe.
I can even see global health interventions decreasing nearterm extinction risk more cost-effectively than AI safety interventions. I am not aware of any quantitative modelling comparing both options.
On diminishing returns, why do you set the (marginal) cost-effectiveness of funding GiveWell's All Grants Fund to the cost-effectiveness of the grants GiveWell made in 2025 instead of GiveWell's cost-effectiveness bar?
I think GiveWell's current cost-effectiveness bar is 6 times their benchmark. From the page I just linked which was last updated in September:
You assumed a significantly higher cost-effectiveness.
Hi Dan. I expect terrorist attacks to be super unlikely to cause human extinction in the absence of very strong evidence suggesting otherwise. For the attacks killing more than 10 people among the 202 k listed on the Global Terrorism Database (GTB), the fraction killing more than 10*N people is around 1 % of the fraction killing more than N people. This suggests a Pareto distribution with a tail index of 2. Around 0.001 % of attacks killed more than 1 k people. Assuming the same Pareto distribution (straight line fitted to the section of the graph below with more than 10 deaths), more than 1 billion deaths would be 10^-12 (= 0.01^(9 - 3)) times as likely as more than 1 k deaths. So the Pareto distribution implies only 10^-17 (= 10^(-5 - 12)) of terrorist attacks cause more than 1 billion deaths. I think the actual risk is way higher. At the same time, I believe a significant risk of a terrorist attack causing human extinction, like 1 % over the next 20 years, would depart a lot from the most natural uninformed ways of extrapolating past data. A pandemic causing human extinction looks much more plausible based on past data (although I still guess the risk of human extinction in the next 20 years is much lower than 1 %).
Hi Marcus and Jim. Great post. I agree. I also liked the direct writing style.
Hi Liam. Nice post. You may be interested in this linkpost for the article "The Conscious Nematode: Exploring Hallmarks of Minimal Phenomenal Consciousness in Caenorhabditis Elegans".
Here are illustrations of the above.
In this post, "welfare range as a fraction of that of humans" means "individual welfare per fully-healthy-animal-year as a fraction of the welfare per fully-healthy-human-year".
Why would artificial wombs and robo-nannies have very different effects from other innovations that have decreased the time cost of having children (for example, washing machines), or increased the income of prospective parents (for example, increasing income of women)? A lower cost of having children as a fraction of available income would increase fertility if the cost of everything else remained the same, but alternatives would become cheaper too relative to available income. It is unclear to me whether your point strengthens or weakens the conclusions of Alexandrie and Eden (2025) under the population models they analysed.
That makes sense.
Hi Mal.
Are you assuming the bees would prefer not existing? I suppose this is very uncertain.
Hi Richard.
Are you confident the longtermist interventions you have in mind, like work on AI safety, are more cost-effective accounting for longterm effects than global health interventions? Decreasing the risk of human extinction may be less cost-effective than global health interventions accounting for longterm effects hold under standard population models. So it is unclear to me whether there is the rigor-reach dilemma you describe.
I can even see global health interventions decreasing nearterm extinction risk more cost-effectively than AI safety interventions. I am not aware of any quantitative modelling comparing both options.
Relatedly, I liked the post by 50 things I know by Cate Hall.