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).
Vasco Grilošø
Karma: 12,067
Hi Wladi. Great post.
I very much agree that posts should be assessed on the merits of their content, not on how much they were written by AI. In addition, I agree labels indicating how much posts are written by AI may decrease the time readers spend engaging with them. Here is a related literature review from Claude Opus 5.5 Max.
I guess the visibility of posts on the EA Forum only directly depends on their karma, and when they were posted. In this case, a post being more written by AI is penalised only if readers decide to vote on it less favourably as a result. @Toby Tremlettš¹, is this correct?
Hi Seth.
If something promising (re: novel) and/āor evidence-backed came up, I think people would be into it.
Here is a related post from Tom Billington and Henning Peters.
Folks, stop it!
I found this funny.
Hello K. Kirdan.
Iāve been thinking about the same problem over the past few months. (Thanks to Brian Tomasik and Vasco Grilo.)
Thanks for linking these posts too. You may find interesting my reply to Stijnās post.
Iām generally slightly in favour of reducing meat consumption because of the potential positive side effects, such as shifting peopleās values in a more animal-friendly direction. I hope these benefits will outweigh the harm, but the situation is far from satisfactory.
Advocating for lower meat consumption based on animal right may encourage wildeness preservation, and the same may apply to advocating for improving the conditions of farmed animals.
The only statistically significant difference we observed between groups shown different meat reduction appeals was quite counterintuitive: participants in the animal cruelty group reported greater āpreferences for more habitatā than participants in the environmental group, t(375.5)=3.52, p=0.0005.
I am sceptical that advocating for lower meat consumption is a cost-effective way of helping wild animals.
Hi Stijn.
Most wild animals die prematurely, which can indicate that they have a negative welfare overall.[1] Hence, by freeing up agricultural land, vegans cause harm (to wild animals).
I think it is very unclear whether decreasing agricultural land increases or decreases the welfare/āsuffering/āhappiness of soil invertebrates. For moral weight proportional to āindividual number of neuronsā^n, and n from 0 to 0.95, I estimate the Good Food Institute (GFI), which decreases agricultural land, increases animal welfare with a probability of 50.7 % to 50.8 %.
For n = 0.188, the log of moral weight explains 89.3% of the variance of the log of the mean sentience-adjusted welfare ranges in Bob Fischerās book about comparing animal welfare across species (see the regression below).
It is also unclear to me whether decreasing agricultural land increases or decreases the population of soil invertebrates. Below are my estimates for the change in organism-years per $ donated to GFI. The population of soil ants, and bacteria and archaea, and trees increases, but it is unclear to me whether the population of soil termites, springtails, mites, and nematodes increases or decreases.
Hi Lance. Thanks for sharing this story. I liked that you shared your thoughts very transparently.
Hi Brenton. Thanks for the honesty in the rants.
Say 80,000 Hours uses marginal funding from Coefficient Giving (CG) to hire someone who costs 100 k$/āyear, and is x times more productive than the alternative hire. CG should fund 80,000 Hours until they are indifferent between granting 80,000 Hours 1 $ more, and saving 1 $. So the annual impact of 80,000 Hoursā new hire should be similar to that resulting from donating 100*x k$/āyear to 80,000 Hours or CG?
There are a couple of well-known answers to that. The person hired in your place probably cares about the world too, and would go on to do something useful with their time instead (Ben Todd wrote about this in 2015). And in a hiring round, the top candidate is often a lot stronger than the next-best one (Rob Wiblin worked this through in 2019).
I liked Robās analysis, but I put very little weight on it. Rob guessed the distributions for productivity. They are not based on any concrete empirical data.
DigiĀtal Mind Castes
The Tomasik post is interesting, but he assumes a theory of consciousness which rules out lots of theories many people (including myself) find plausible like psychophysical laws, type B physicalism, and more.
Do you think lack of consciousness implies a moral weight of exactly 0? If so, how are you defining consciousness? For any sharp definition, infinitesimal changes in the state of an organism would make them go from not conscious to conscious. Here are related discussions.
On animals, my main view is that the vast majority of the effect of any near-term intervention is on animals rather than humans. Iāve come around to the idea that there are many more cost-effective animal interventions than THL, having donated to SWP and the Insect Institute.
Do you endorse moral weights in the ballpark of the sentience-adjusted welfare ranges presented in Bobās book? This would be compatible with the above. However, for such moral weights, it is unclear to me whether electrically stunning farmed shrimps increases or decreases animal welfare. Do you disagree?
GFI is a strong choice as well.
Is this because of the effects on farmed animals or soil invertebrates?
Thanks, Laura. I will link to your reply if I end up publishing a post. I am currently not planning to publish a post because it looks like you are going to make changes before you make more grants.
I do not know how much CCFās recommended allocation would change as a result of the marginal cost-effectiveness of AGF being updated to GiveWellās bar. However, the update seems very worth it to me due to taking little time (only one variable has to be updated), and GiveWellās goal of having their marginal cost-effectiveness match their bar.
Here is some context about what it means for an intervention to have a cost-effectiveness equal to 16 times GiveWellās benchark (your current assumption for the marginal cost-effectiveness of AGF). GiveWellās benchmark corresponds to paying 300 2024-$ for 1 unit of value (doubling the consumption of one of their target beneficiaries for 1 year). So GiveWellās current bar of 6 times their benchmark means they are willing to pay at most 50 2024-$ (= 300ā6) for 1 unit of value. GiveWell assumes saving the life of someone with age 1 to 4 is worth 127 units of value. So GiveWellās current bar corresponds to saving someone with age 1 to 4 with 6.35 k$ (= 50*127). A cost-effectiveness of 16 times GiveWellās benchmark, 2.67 (= 16ā6) times GiveWellās bar, corresponds to saving someone with age 1 to 4 with 2.38 k$ (= 6.35*10^3/ā2.67). GiveWell reports higher costs to save a life for the grants they made to their top charities in 2022-2024 (and the cost to save a life would be even higher for the last $ of such grants):
āIn 2022-2024, we directed funding to the Malaria Consortium to support this program at an estimated average cost-effectiveness of $4,000 per life savedā.
āIn 2022-2024, we directed funding to the Against Malaria Foundation to support this program at an estimated average cost-effectiveness of $5,500 per life savedā.
āIn 2022-2024, we directed funding to Helen Keller Intl to support this program at an estimated average cost-effectiveness of $3,500 per life savedā.
āIn 2022-2024, we directed funding to New Incentives to support this program at an estimated average cost-effectiveness of $4,500 per life savedā.
I have listened to at least 2 episodes of these podcasts (I keep updated the list in the doc I just linked):
GiveWell says āWe maintain a cost-effectiveness threshold [bar] at a level where we expect to be able to fully fund all the opportunities we find above itā, as I have been assuming. So I am confident it would be better to set the marginal cost-effectiveness of GiveWellās All Grants Fund (AGF) to GiveWellās bar, 6 times their benchmark in May 2026, instead of the cost-effectiveness of all the grants they made in a year.
You are currently assuming AGF has a marginal cost-effectiveness of around 16 times GiveWellās benchmark (see my 1st comment in this thread), 2.67 (= 16ā6) times GiveWellās bar in May 2026. This is a significant difference in the context of global health interventions, whose cost-effectiveness is less uncertain compared to that of interventions targeting animals, and decreasing catastrophic risks neglecting effects on non-target individuals. So I suggest you update your assumption about the marginal cost-effectiveness of AGF before CCF makes more grants.
I am considering publishing a post about this on October 28. Let me know if you update your assumptions.
Hi Ariel. You may be interested in Brian Tomasikās post Dissolving Confusion about Consciousness.
It doesnāt make sense to ask questions like, Does a computer program of a mind really instantiate consciousness? That question reifies consciousness as a thing that may or may not be produced by a program. Rather, the particles constituting the computer just moveāand thatās it. The question of whether a given physical operation is āconsciousā is not a factual dispute but a definitional one: Do we want to define consciousness as including those sorts of physical operations?
Relatedly, you may like this linkpost for Ethics without sentience. Facing up to the probable insignificance of phenomenal consciousness by FranƧois Kammerer.
The relevant question is how much animalsā experiences are discounted relative to the global poorās experiences. If theyāre discounted the same amount, animal welfare remains ~1000x as cost-effective [ratio between the cost-effectiveness of corporate campaigns for broiler welfare and GiveWellās top charities neglecting effects on non-target individuals].
I got this estimate based on the sentience-adjusted welfare ranges Rethink Priorities (RP) initially presented, and neglecting effects on soil invertebrates. However, I have little idea about whether chicken welfare reforms increase or decrease welfare accounting for uncertainty in welfare comparisons across species, and effects on soil invertebrates. For the sentience-adjusted welfare ranges presented in Bob Fischerās book, which are similar to the ones RP initially presented, I think effects on soil invertebrates would be much larger than those on chickens. So, trusting the values in the book, and excluding interventions decreasing uncertainty, I would focus on increasing the welfare of soil invertebrates by increasing or decreasing agricultural land. I believe this can be achieved more cost-effectively by the Good Food Institute (GFI) or Giving What We Can (GWWC) than interventions improving the conditions of farmed animals. In reality, I favour interventions decreasing uncertainty about welfare comparisons across species, and effects on soil invertebrates.
Hello. You may be interested in David Thorstadās series Instrumental convergence and power seeking.
I see. I wonder if any of the patients who has benefitted from DMT would be interested in appearing on podcasts. I suspect founders of orgs fighting cluster headaches, people leading the research, and people with cluster headaches who have personally benefitted would more easily be accepted as guests than someone hired for public-facing work who is not leading the research, and has not personally beneffited.
Hello Jim. I had understood you used data from GiveWell (this is mentioned in the section I linked above with your methodology). However, as far as I know, GiveWell has never said the cost-effectiveness of their marginal spending in 2025 matched the cost-effectiveness of all their grants considered together in 2025, which is what you are assuming (see the quote after my question in the comment above). Here is what they say in the post you linked.
In 2025, we generally considered funding opportunities that our research estimated to be at least 8x our cost-effectiveness benchmark. Overall, we estimate that the average cost-effectiveness of the money spent across all our grants was approximately double that: 16x our benchmark.
So GiveWell estimates the cost-effectiveness of all their grants considered together in 2025 was 16 times their benchmark. The cost-effectiveness of GiveWellās marginal spending in 2025 was lower than this. GiveWell prioritises the most cost-effective grants, and some grants had a cost-effectiveness lower than 16 times their benchmark.
The marginal cost-effectiveness of GiveWellās marginal spending in 2025 may have exceeded their bar. In principle, it could have been anything from 8 to 16 times their benchmark. In practice, I think it is close to 8 times their benchmark. Assuming they funded all grants with a cost-effectiveness higher than their bar of 8 times their benchark (considering all factors, not just the ones covered in the explicit cost-effectiveness analysis (CEA)), GiveWell granting more in 2025 would not have led to more grants with a cost-effectiveness higher than 8 times their benchmark (which would have been made anyway). It would have led to grants with a cost-effectiveness of around 8 times their benchmark (slightly below or above 8 times their benchmark because the distribution of the cost-effectiveness of the grants is not perfectly continuous).
Hi Brenton.
Salary
Opportunity cost
Best quantified from the point of view of aligned funding: roughly, how much an aligned funder would be willing to pay for the impact they expect this person could produce if they werenāt working at their current org.
Capacity-building value
Working at the organisation affects the expected value of the rest of that staff memberās career: they gain skills and context, and based on their experience there, the job increases/ādecreases their inclination to remain in the ecosystem.[1]
Should the opportunity cost as defined above already cover capacity-building value? Funders should account for capacity-building value when assessing how much they would be willing to pay for the impact of hiring someone.
Hi Toby. Great post.
Hello Alfredo. Thanks for this update. I am glad you have many ideas to make progress.
Our project ClusterInfo is one of our attempts to reach such patients. Itās a website with free resources on how to diagnose and treat cluster headaches, currently available in eight languages. Since launching in late May, weāve already reached over 11,000 unique visitors, and weāre thinking hard about how to make sure that anyone, anywhere searching for information on cluster headaches can quickly find the information they need. We work with patients, clinicians, and other advocates to ensure the content is accurate, comprehensive, and easy to understand. You can find some web analytics here, in case youāre curious.
Have you considered trying to spread the word about ClusterInfo on podcasts? Claude says Joe Rogan has had at least 3 episodes centred on DMT:
Rick Strassman, #1854 (August 2022). Strassman is the psychiatrist who ran the 1990s University of New Mexico DMT trials. He wrote āDMT: The Spirit Moleculeā, a book about his findings and experiences researching DMT, and Rogan hosted the documentary version of it. sportskeeda
Rick Strassman again, #2385 (26 September 2025). He discussed his 1990s DMT studies, and the two compared DMT experiences with ancient religious texts, especially the Hebrew Bible and the Book of Enoch. He also promoted a new book, which another summary describes as being about trauma, psychedelics and spiritual growth. shortformpodwise
Andrew Gallimore, #2403 (30 October 2025). Gallimore is a chemical pharmacologist and neurobiologist. The interview was on DMT and the nature of reality. It covered his book, subtitled āConfronting the mystery of the worldās strangest drugā. singjupost
Hi Siebe. Nice point. Relatedly, I think people often overestimate the importance of the probability of sentience (in the EA community too).
If you trust there is as little variation in the probability of sentience as suggested by the values used by Ambitious Impact (AIM) and Animal Charity Evaluators (ACE), or presented in Bob Fischerās book about comparing welfare across species, I believe there are other factors which may be more important for the probability of a small donation increasing animal welfare:
Donating to incremental instead of hits-based interventions.
Donating to smaller organisations.
Donating to organisations in lower income countries
I wonder to what extent people donate to interventions targeting animals which are more likely to be sentient to boost the probability of increasing welfare. People routinely take actions which are super unlikely to actually matter:
I calculate driving a car for 10 km in Great Britain without a seatbelt leads to 1 additional death with a probability of 1 in 73.0 M. AIM uses a probability of sentience of shrimps which is 34.2 M times as high.
Andrew Gelman found the probability of a voter in a small United States (US) state polling around 50ā50 in a close election nationally changing the outcome of the national election could get as high as 1 in 3 million. AIM uses a probability of sentience of shrimps which is 1.40 M times as high.
I do not have a strong view. However, I would drop the labels saying how much posts were AI-written, and let readers decide whether they want to disclaim it. The labels create an incentive to assess posts partly based on how much they were AI-written, and I do not think this matters per se. I would also not want labels about the IQ of the author, and their ethnicity, even assuming they were reliable. I guess posts of people with IQ about 130 are usually higher quality than posts of people of IQ below 70, but I would still not want to incentivise assessing posts based on IQ.
Nice.