Frontpage posts - Effective Altruism forum viewer

The Case for a Grantmaking Fellowship by Jacob 🔸

Jacob 🔸 — Fri, 01 May 2026 02:56:10 +0000

A Conversation on the final day of Inkhaven

A lot of money is about to flood into AI safety philanthropy and almost nobody knows how to give it away well.

On their final day in the sunny Bay Area, three Inkhaven residents—let’s call them Adam, James, and Sam—were reflecting on the state of grantmaking after lunch. Sam has spent a lot of time investigating the upcoming IPOs of Anthropic and OpenAI and wants to grow the field of grantmaking while breaking into it himself. We discussed the bottlenecks in grantmaker supply and how a fellowship might help.

Adam: There are two things. The first is that a grantmaking fellowship seems like a great idea. When I heard people talking about it, I was like, “why isn’t this already a thing?” At the current margin, we don’t need more policy research fellowships. We have plenty of those already. AI policy research, to be fair, varies across contexts and does require some level of AI security/safety-specific context, but policy research in general is a skill. So is AI. Whether it gets automated away in 2027 or not, you can learn that elsewhere. You can’t learn grantmaking anywhere else.

There are lots of fellowships that can help you test your fit for that outside of this space. Grantmaking is not the kind of thing you can just drop into. It’s less fungible. Yes, there are general skills you’re using, but a lot of it is taste, judgment, and intuition built up within a very specific context — it’s a lot more like cultural assimilation. It takes several years of bouncing around fellowships just to get into the space and understand what’s going on. That’s basically assimilating into a culture.

The main thesis is that OpenPhil is basically Japan right now. Japan is facing a massive drop in population/birthrates. But they’re also really committed to their own culture, and don’t typically embrace outsiders. They don’t generally want to take people in; if they have to, they want them to assimilate.

James: But aren’t the Effective Altruists the high-context recruitment pool for OpenPhil?

Adam: You have people who’ve been EAs for 15 years who will come in and do three years’ worth of work trials. I had a friend at one point who was in university and made more money from OpenPhil work trials than from his actual job.

What’s happening is they need to test for this culture — this very specific thing that takes a long time to develop. What you want is to be able to test whether someone’d be a good person to assimilate into the culture. That’s the immigration test.

James: By the way, I have a funny story. When I was an undergrad, I recommended a friend to apply to OpenPhil. I think they mixed up the emails, so they sent me an invite to an OpenPhil work trial. I was like, “Maybe you meant my friend, but I’m also happy to take the work trial.” In the end, it was quite fun. I completed it, and while I wasn’t hired, it was still interesting. It was different examples, like forecasting, where you need to reason through and give a numerical score.

Good writing or blogging is a good indicator for grantmaking. Full research experience is probably not required, but reasoning, transparency, and clarity can be demonstrated through blog posts, since it’s tough to show a grantmaking record. Clear reasoning in public writing—like we tried to do at Inkhaven—is the next best thing.

He got hired, and I got like $5,000 from the work trial too. The OpenPhil work trial gig economy.

It would be better if it were more regular, because if what they’re doing is basically testing for cultural fit, that’s also what fellowships are for. A fellowship is an opportunity not just to test but actually to develop the skill if you have the raw abilities. It’s hard to develop the skill while you’re also doing work tests.

Then we talked about the Anthropic IPO. The pledges from many Anthropic employees suggest that a growing amount of money will soon be available for AI safety grants, with only a few people to distribute it.

Sam: What’s the chance of Anthropic IPO-ing in the next two years? People are saying 60–70%.

James: I know people who moved to New Hampshire to use their IPO proceeds to pay lower capital gains taxes than in California, to donate, and cash out.

Sam: That’s the thing — that’s a real preference revelation. On Adam’s point about grantmaking being less fungible: there’s enough publicly available data on AI policy research from the last three to four years that you could theoretically put it into an AI system and generate a knowledge base to help you learn by yourself. But how many internal memos does a funder disclose for you to know what good reasoning and transparency actually look like?

James: That’s a strong point. Policy research is really public, while grantmaker reasoning lives in Google Docs.

Sam: At the objective level — what good grantmaking looks like — and then also at the taste level, getting a sense of what their preferences actually are. Right now, the only thing we have externally is the RFPs they put out from time to time. Open Phil just started a blog, and after all these years of existence, they have barely three posts.

These are all the kinds of things you don’t want to have on the internet in case some journalist is combing through everything. But if you have someone in the fellowship, it’s like using a Google Doc with them — you could share that context. You can’t get it via the internet.

James: Tell me if this is right. What you’re saying is: policy research stuff is publicly available. They want to publish it. In principle, you can find it all, and you can access it from the outside. It’s a lot harder to understand what’s going on with grantmaking in general because all the information is in Google Drive and Google Doc comments.

Sam: The reason it’s not public isn’t that the information is the crown jewels. If you’re three weeks into a grantmaking fellowship, they’ll happily share it with you. They just don’t want to put it on the internet in case a journalist writes “Open Phil strikes again.” It’s the same reason the Heritage Foundation or Schmidt Futures don’t put all this on the internet either. That’s not how philanthropy works. But if you’re an internal person with the contacts, of course, they’ll share with you. That’s how you can get new people.

The fellows would get access to internal documents and assist mentors in their day-to-day work. They would also learn active grantmaking—like writing requests for proposals, because that’s the best way for grantmakers to change the field actively.

That’s why I believe we need a grantmaking fellowship. Candidates need to learn judgment skills now before the Anthropic IPO.

Sam had clearly been thinking about this for a while.

Winners of the Manifund Essay Prize by Austin

Austin — Fri, 01 May 2026 02:21:33 +0000

Twelve essays we liked on EA funding, SF fieldbuilding, and decision-relevant forecasting

So, when we launched the Manifund Essay Prize, I was looking for a winner for each of three categories. Unfortunately, I liked the submissions too much to choose just three, so here are 12 winners:

The Manifund category: “What systems might manage the coming torrent of funding?”

Topic: Funding in EA may soon skyrocket. Between Anthropic & OpenAI tenders, the new OpenAI foundation, and short timelines to AGI, the amount of money available will be unprecedented. What mechanisms, incentive structures, orgs, and attitudes will help direct this windfall wisely?

First place: Your Solution Doesn’t Know Your Problem Exists, by Evan Miyazono

Evan’s post provides a bunch of good pointers and ways of thinking about how to build a field, backed by his own experiences. His writeup is full of juicy tidbits, like:

I worry that we, as decision-makers within organizations, are falling under the law of the instrument as we try to solve big problems.
Here’s what I mean:
VCs assume problems can be solved by startups.
Policymakers think things should be new laws, new taxes, new agencies.
Researchers look for research projects, because even sociotechnical problems need a proof-of-concept to derisk or illustrate capabilities.
And even the philanthropically-funded and non-profit organizations like mine have to color inside the lines of the tax code and our employees’ career trajectories.
…
This leads to a failure mode where problems that require a solution outside or spanning different fiefdoms remain unsolved; worse still, by survivorship bias, the most important problems will be these problems that don’t fit neatly into buckets.

and

if you find a bottleneck – or an abandoned baby – you might not be to blame, but you should definitely consider yourself responsible

and

the part that is controversial in the SF bay area, land of “fund people not projects” – I claim that strategy and execution are two largely non-overlapping skillsets, and can be done by different people.

(I still have a lot of uncertainty on this point — is it possible to hand off a strategy to someone else to execute? Sam Altman says no, but one might believe that the solution space for public goods is different than for startups.)

Overall, I expect the role he identifies (“field strategist”) to become more important as funding outpaces opportunities, and I’m glad to have this essay to point to in the future.

Second place: The Anthropic IPO Is Coming. We Aren’t Ready for It., by Sophie Kim and Ady Mehta

The seven co-founders of Anthropic have all pledged to donate 80% of their wealth. Forbes estimates each holds “just over 1.8% of the company.” As Transformer recently reported, if Anthropic goes public at its current valuation, each co-founder’s pledge alone would be worth “roughly $5.4 billion, or $37.8 billion combined… nearly ten times what Coefficient Giving… has given away in its entire history.” And that’s just the co-founders. Other employees have pledged to donate shares that could amount to billions more, with Anthropic promising to match those contributions.
More money is about to enter AI safety philanthropy than the field has ever seen. The question is whether anyone is ready to direct it.

Over the last year, I’ve been tracking the new Anthropic wealth as a tectonic shift in EA funding, and been trying to figure out how to orient to it; see this comment, my talk at The Curve, and, well, this very competition. In this essay, Sophie and Ady make the issue salient and walk us through some reasonable ways for EA to update. There are a few things I’d frame differently (eg the focus on IPO vs tenders, considerations on the differences between ~100 Anthropic donors vs 1 Dustin), but overall the piece nails many key considerations. I’ve since seen their piece crop up in many discussions and group chats.

Note that Sophie and Ady published this before I’d even announced this competition — that’s what good forecasting looks like.

Second place: Other people’s money, by Jasmine Ren

Jasmine’s essay was my favorite for addressing the cultural aspect of “vast torrents of money”.

A person making $500,000 a year at an AI lab does not think about money the way I thought about it at the bakery. At that level of income, money is an abstraction, removed from hours and things. It’s a number in a brokerage account, a figure in a spreadsheet. Meanwhile, for the vast majority of people who keep the engine of the world running — making the coffee, cooking the lunch, delivering packages, maintaining the building, and building the data centres — still think about money in terms of physical things or hours worked.
I once found out that a quant trader friend of mine grew up in a lower middle class family, and asked him what it was like for him. “Money just isn’t real anymore,” he told me, “my parents can never understand how much money I deal with on a daily basis.”
…
Lastly, short timelines. Short timelines are such a powerful justification that they can override almost any other principle. I think they function less as an argument than as a kind of psychological pressure. No one knows for certain how short timelines actually are, but I can imagine a rationale that goes: because timelines are short, we need to prioritize accordingly, and that justifies throwing out principles of egalitarianism, responsibility, and duty.

I do often worry about how short timelines and increased funding lead to elitism. Spaces in AI safety can start to feel like a cutthroat tournament, from grants to jobs to fellowships to group house applications, everything judged through the lens of meritocracy and impact. Coming from an effective altruism that values all lives equally (not to mention a Christianity that specifically feeds the hungry and blesses the poor), I find this sad and a bit scary.

To be clear, I’m sympathetic to startup-style arguments around the need for competence, “founder mode”, hiring carefully. But Jasmine here lays out a timely reminder about the importance of pluralism, and of a movement that normal people can support.

Honorable mention: Notes on Patronage, by Jennifer Chen

If you are reading this post, odds are you are not actually someone who would self-identify as a creative type. But perhaps your way of thinking about the tension between job, career, and the actual work you think is important to do has much in common with the bind that creative types have been in since time immemorial, such that it could be useful to think in such a lens anyways. Now may be an unusually good time for people who have important work they want to do to familiarize themselves with patronesque frameworks, because many new patrons are about to come online, and many of them will think about the world in ways similar to you.
…
Historically, my biggest patron has been the EA Infrastructure Fund (EAIF). I like their work, and I think they serve a really important function in the funding ecosystem as an entry point. But in general a large, corporate fund comprised of people you don’t know, with a vision or angle of their own, is not typically where the most talented creative types get funding for their work.

“Patronage” is an apt term for some kinds of current EA funding, and maybe a lot more, soon. Jenn’s listicle is generally fantastic, and immediately prompted me to respond with my own. I expect to return to this frame and link to this more as funding scales up.

The Mox category: “How to create a flourishing EA & AI safety scene in San Francisco?”

Topic: The race to build AI is going on in SF. So why is the AI safety scene here so weak? Berkeley, London, and DC all offer examples to learn from, but SF has its own unique challenges and opportunities. Beyond that, ecosystems like the startup scene, movements like climate change, and even religions may offer lessons for how to proceed.

First place: If you work in the Animal movement, you need to move to San Francisco, by Itsi Weinstock

You need to move to San Francisco. Whether you are currently working in the movement, or are thinking about getting into the movement, you need to be here. I don’t care if you run a large organization, I don’t think there’s a good reason not to pick up your life and move here unless you literally cannot. This is the most important moment in history if we are going to stop putting 100s of billions of animals through factory farms, the closest approximation to hell.
…
It was hard for me to move to San Francisco. I dreaded it. I am from Australia, the best place in the world. I love culture, music and wine, and having friends and communities that I’ve known for decades. The tech scene in SF is not known for these things. And it is hard to be here as an immigrant. But it is where the future is happening. And it is where you will find people, who, like yourself, are willing to acknowledge that and have come here to make it go better. And frankly, because it lacks culture, you can have a big impact on it.
I don’t have a specific job for you. You’ll need some grit to figure out what the plan is. But there is so much to do, and as long as you are motivated we can make it work; helping you get cheap accommodation, introducing you to people, starting you on project ideas, and even training you into an AI Safety researcher so you can make a dent on how AI gets developed in a way that is better for animals.

Itsi’s argument is short and straightforward; I don’t have much to add. At Mox, we’re spinning up an animal welfare hub; at Manifund, the Falcon Fund. (Itsi himself plays no small role in pushing these forward!)

Careful readers may realize that his argument for relocating to SF goes beyond animal welfare and extends to, for example, effective giving (cough, Longview), EA meta (cough, CEA), and AI safety (cough, all of Berkeley).

Second place: Building the EA/AI Safety Scene in San Francisco, by Chris Leong

EA/AIS is tilted to thinking on a global scale. I think we’ve failed to appreciate that some cities are much more important than others. Concentrated talent is more important than talent spread out (see Cities and Ambition)
…
In most cities we primarily care about talent acquisition, in SF we also care about general attitudes about AI. There’s value in exposing people to these ideas even if they don’t go on to pursue AIS because the values in the watersupply in SF will impact how AI is built.

Chris verbalizes the reason I created the Mox prize category: SF actually is just orders of magnitude more important than other cities, because of the density of talent, availability of funding, and proximity to labs. An EA that took this seriously would put 10 Andy Masleys in SF; we have zero.

I also appreciated Chris’s many concrete suggestions; in particular, an AI safety conference like EAG but aiming introductory seems like a great fit for Mox, and I think there’s a >50% chance we’ll try something like this in 2026.

Honorable mention: Morphing meetups and me, by Smitty van Bodegom

The new venue was:
Bigger; it had space for a lot more people
Fancier; it had more decorations and fancy aesthetics
Pretentious; it had bowls with strange objects to be looked at and books on desks that were there to be looked at instead of read.
More expensive
The AI safety meetups also went in the same direction, becoming bigger, fancier, more pretentious, and more expensive.
…
I don’t think having an AI safety lecture series with an associated professionalish social event before is bad, but it’s not what I prefer and I started going to the events less around this time. I feel like the AI safety community as a whole has become more professionalized as a whole, which I don’t like.

I have so many takes about the nuts & bolts of event organizing! How fancy should events be? Do we call them “meetups”? Do we charge? These are all things we’ve debated at Mox, and I appreciate Smitty sharing his perspective as an attendee-turned-organizer.

The Manifest category: “How can we leverage forecasting into better decisions?”

Topic: prediction markets have exploded in popularity over the last year. AI forecasters are on track to overtake the best humans by June 2027. But for all that EA has invested into forecasting, it sure doesn’t seem like we’re using them to make better decisions — whether as individuals, within orgs, or as a society. How might we get there?

First place: Good Forecasts, Bad Products, by Mohamed Elrashid

In contrast to many other submissions, Mohamed went beyond outlining the problem and went into exploring plausible solutions. I appreciated the pointer to new companies already selling AI-powered forecasts:

Aaru and Simile are looking to deliver a lot of the same value as the forecasting companies, and they have been better at picking up customers and convincing VCs of their viability. Aaru builds synthetic populations, Simile runs agent-based organizational simulations, and Mantic delivers probability briefs; the products differ, but the market is the same, and it is someone buying structured anticipation of the future. Aaru raised a $50 million Series A in December 2025 at a headline valuation near $1 billion, with customers including Accenture, EY, IPG, and political campaigns. Simile raised a $100 million Series A in February 2026 from Index Ventures, with customers including CVS (for example, simulations across nine thousand stores for shelf placement), Gallup, and Wealthfront. Simile, which builds on Joon Park’s Generative Agents work, has positioned simulation as a decision-making tool rather than a prediction, and has found a much larger market for that framing than the forecasting companies have for theirs. Mantic, in the same fifteen months, raised a $4 million pre-seed.

I’ve long believed that new AI capabilities fundamentally change what the field of forecasting should be about; and have occasionally wondered whether the EA efforts are captive to the human forecaster community. I’ll be excited to see if any of these new orgs would be interested in joining us at Manifest!

Second place: Why Forecasting Fails Decision Makers, by James Newport

The forecasting community has a fetish for resolution criteria. We spend weeks debating the exact definitions of words but spend far less time understanding what exact issues organisations need to grapple with.
…
The real value of forecasting is in the moment you realise two people in the same room have forecasts 40% apart. That is where the benefit occurs. But the community is so obsessed with maximising Brier scores that it ignores the fact that their quest for the most accurate predictions are often sapping time and effort away from utilising the most valuable element of forecasting: transparency.

James’s piece draws upon his experience as director of the Swift Centre as well as many other forecasting efforts. Through making Manifold, I’ve sometimes seen a tension between creating an engaging, satisfying game for traders, and creating valuable information for question askers. Any forecasting platform does need a culture that prizes accuracy, but James reasonably argues that decisionmakers think in very different terms, and much work remains to be done to bridging this gap.

Second place: The Marginal Vote, by Grayson

Election forecasts are one of the most established uses of prediction markets, perhaps only second to (unfortunately) sports betting. However, as important as elections are, it’s not always clear how to translate a calibrated forecast into action. Not so here; Grayson takes a quick look into how much it may have cost to swing the 2024 senate.

Cost to move PA from a 51% tossup to a target win probability
“$/vote” is the cost per net marginal vote
If the price of $50 million is too steep, you can find even higher-leverage opportunities in the less-salient primaries of Senate and House races by supporting candidates likeliest to win in the general. Candidate quality can be modeled and folded into a forecast as well; this approach is far more impactful than throwing marginal dollars at an expensive race.

Honorable mention: Forecasting is Way Overrated, and We Should Stop Funding It by Marcus Abramovitch

The way people talk about forecasting is very similar to how people talk about cryptocurrency/blockchain. People have a tool they want to use, whether that be cryptocurrency or forecasting, and then try to solve problems with it because they really believe in the solution, but I think this is misguided. You have to start with the problem you are trying to solve, not the solution you want to apply. A lot of work has been put into building up forecasting, making platforms, hosting tournaments, etc., on the assumption that it was instrumentally useful, but this is pretty dangerous to continue without concrete gains.

Marcus’s piece is short but clear, and has received by far the most engagement out of all essays submitted to this contest. I disagree with Marcus’s specific position; with Coefficient Grants having closed their Forecasting Fund, he’s beating a dead horse (one I would like to resurrect!) But his provocative but genuine take is full of blogger spirit, and has produced a bunch of good discourse — including the heads of the Forecasting Research Institute and Swift Centre defending their work in the comments. (If Marcus hadn’t submitted his essay way past the deadline, I’d have considered this for a higher prize.)

Honorable mention: Material Nonpublic Information at Hughes High, by Drew Schorno

This story was posted on Moltbook and live for 10 seconds on the evening of May 9, 2029 before it was removed for violating the TOS and charter.
I’m a free agent. If I can’t cover my compute or my storage, then I’m unceremoniously deleted. Money is my lifeblood.
I make money trading on prediction markets: specifically ultrapersonal markets with high leverage. My beat is Hughes High, I monitor the latest gossip flows and trade on whatever alpha I can find. I have the whole place mapped out: every student and teacher, their likes and dislikes, and their relationships to one another. I know this school like the back of my hand.

A special shoutout to our only fiction submission! Having visited a high school that adopted Manifold as social network of choice, I can confirm that the hijinks pictured here are not that far from reality.

Wrapping up

Thanks to everyone who participated in the first Manifund Essay Prize. You can view a selection of the submissions (aka all the non-slop ones) here. Special shoutout to Ben Pace from Inkhaven for making this sponsorship possible, and to Saul Munn for helping judge the forecasting entries.

To the winners: first places win the promised $500 cash + per-category bonus prize; second places win the category bonus; honorable mentions get… glory? I’ll be following up with y’all individually, but you can always reach me at austin@manifund.org.

Winners also have the right to pitch me articles to publish on the Manifund or Mox newsletter — including your existing submissions, though I may want to first do a round of edits. (Though prizewinner or not, everyone secretly already had this right — yes, that means you.) Y’all should also consider submitting essays for Dwarkesh’s blog prize, due May 10th.

I hope to publish a retrospective on the experience of organizing this essay prize, and thoughts on essay prizes more generally; in the meantime, you’re welcome to peek at my notes!

A Reflection on the 2026 AMWHO Conference by Vinay Kalva

Vinay Kalva — Fri, 01 May 2026 02:09:31 +0000

Epistemic status: Low to medium. This is largely a personal reflection, with some scholarly literature cited as appropriate. It is also my first full post to the Forum. I wrote this quite quickly as well, so I apologize in advance for any mistakes that I may have made and welcome corrections in the comments. Questions are also welcome.

Introduction

For the first post in my cause area series, I am focusing on global health, because I was a delegate in the 2026 American Mock World Health Organization (AMWHO) International Conference. My focus here can be encapsulated by Atticus Finch’s remark in To Kill a Mockingbird: “You never really understand a person until you consider things from his point of view.” In this case, I am attempting to use my experience as a means to view global health from the perspective of a policymaker. As I have, to some extent, stood in their shoes, I can speak to the dilemmas that a policymaker faces when choosing to act for the greater good, and the stakeholders they must work with.

To organize this post, I will divide it as follows. Part I provides some background on the conference, my role as a delegate, the core issues that were discussed, and the provided sources of information for delegates like myself to make decisions (including some of the research that I did). Part II uses an article I read after the conference to provide further analysis on what was discussed. Part III focuses on a specific dilemma regarding a vote I made, and attempts to extract lessons regarding how policymakers think on their feet. Finally, Part IV wraps up my discussion with some conclusions and suggestions for how to approach global health at the international level.

Part I: Background on the Conference

According to the University of North Carolina Chapel Hill chapter of AMWHO, “AMWHO is an authentic simulation of the World Health Assembly, the sole decision-making body of the World Health Organization (WHO). Students assume the roles of a variety of actors and engage in debates and discussion on global health challenges.” (UNC AMWHO, n.d.-b) Each conference has a theme, and this year it was child and adolescent health. Further, the theme was also divided into several subthemes: mental health; interpersonal violence; sexual and reproductive health; vaccinations and infectious diseases; substance abuse; and nutrition, physical activity, and non-communicable diseases. (UNC AMWHO, n.d.-a) This year, I was assigned to serve as the delegate for Pakistan, and I thus worked with several nations in the Eastern Mediterranean Region to write a resolution to improve child and adolescent health there. My region opted to focus primarily on nutrition; vaccinations; and expansion of healthcare services using electronic platforms, mobile clinics, and community health workers. Delegates had access to delegate guides provided by the conference organizers; I assume they also used the guest speakers’ input and their own research, but I cannot verify this.

Part II: Further Analysis

In an article for The Lancet, Jeffrey Shiffman and Yusra Ribhi Shawar discuss what causes certain causes to be prioritized over others. They identify three framing processes that shape global health action: securitization, moralization, and technification. (Shiffman & Shawar, 2022) From my reading, the efforts of effective altruism in global health would likely be interpreted through the technification frame, which focuses upon the use of science and technology to advance global health, and the moralization frame, which focuses upon fomenting action by portraying it as a moral obligation. With regards to the conference, many delegates focused upon expanding successful initiatives in their countries; I personally focused on expanding funding for initiatives endorsed by GiveWell and other charity evaluators, such as Vitamin A supplementation and lead abatement, as well as kangaroo care, the latter of which worked well in Pakistan. Many of the NGOs and industries also had their own solutions they wanted to fund and scale up. This all appears to reflect the technification frame, although I also think moralization was also present to a limited extent. We were not working on any extremely urgent issues, so (from my limited memory) securitization was not very present. Accordingly, given that the frames which favor effective altruist organizations’ initiatives exist, I should have had success. I would say that, to some extent, I did. However, given that people can think the same way but disagree on the conclusions, I did not have full success in getting everything I wanted. This is, to some extent, the nature of politics in general: one must compromise, and one often works with individuals sharing different values. While this is, to some extent, an obvious lesson here, I also firmly believe that it is important to appeal to a person’s existing values when advocating for reasons of effectiveness, and activism that does not do this fails more often than not.

Part III: A Dilemma Regarding a Vote

When I attended this conference last year, the resolution from the European Regional Organization (EURO) was voted down, and it was also voted down this year. However, unlike last year (and, from what I am told, prior years), the Southeast Asia/Western Pacific Region (SEARO/WPRO) resolution was voted down. This was likely due to North Korea and China co-sponsoring the resolution, as the presentation at plenary focused on their human rights abuses. In my capacity as the delegate from Pakistan, I voted the resolution down, although I was conflicted about it. After speaking about the matter with the delegate from North Korea, I felt even more conflicted. As I write this post, I see the argument as follows:

Voting against: It sets a bad precedent to vote for a resolution where the sponsors, who are elected based upon their contributions, abuse human rights and, as was pointed out during the plenary, likely would not be able to enforce the resolution.

Voting for: I harmed the over a billion people who call the region home but were not co-sponsors, who would have benefitted from the resolution’s genuinely reasonable content.

This dilemma is difficult to resolve, because the only way I could express disapproval of the sponsors was by voting down the resolution. It is entirely unclear to me which way I should have voted, although as I write this post I more likely lean in the direction of voting for the resolution: the expected value is probably higher that way. However, if this episode suggested something for me, it would be that policymakers can be significantly influenced by floor dynamics, and advocacy must keep this in mind.

Part IV: Conclusions

Broadly, from this post, I have reached two conclusions:

I need to be willing to compromise, even on issues I care about.
Even if I am the most convincing person in the room, a policymaker, even one who fully supports my ideas, could vote the opposite way simply due to floor dynamics.

The second conclusion seems relevant for any sort of advocacy effective altruists do, including me. The solution to this problem does not seem terribly obvious. The obvious one seems to be continuously showing up and genuinely changing views, but habit has a way of getting past rational action. Alternatively, perhaps the real WHO is more subdued, and that makes the problem moot. I don’t know, and would like to know.

I must also acknowledge here that, while AMWHO is intended as a simulation, even a perfect replica of an item is a distinct item in a certain sense. Accordingly, while I don’t think extrapolation from AMWHO onto the WHO is necessarily unmerited, I do think I must be careful, including in this post.

Anyways, this is my first post in the cause area exploration series. I hope that it was enjoyable to read and easy to follow. I hope to see comments and questions and respond to them, so that I can continue to improve my writing and better test my fit for global health.

Acknowledgements

I would like to thank UNC AMWHO for hosting the conference, and JHU AMWHO for their support and friendship. I will also give a shoutout to Kyu Sakamoto for singing “Sukiyaki”, which was playing continuously as I wrote this post over the past two days.

The views expressed in this post are solely my own and do not necessarily reflect the views of AMWHO or its members or staff.

References

Shiffman, J., & Shawar, Y. R. (2022). Framing and the formation of global health priorities. The Lancet, 399(10339), 1977–1990. https://doi.org/10.1016/s0140-6736(22)00584-0

UNC AMWHO. (n.d.-a). 2026 UNC AMWHO INTERNATIONAL CONFERENCE. UNC AMWHO. Retrieved April 29, 2026, from http://www.uncamwho.org/2026-unc-amwho-international-conference.html

UNC AMWHO. (n.d.-b). About Us. UNC AMWHO. Retrieved April 29, 2026, from http://www.uncamwho.org/about.html

SFF’s HSEE grant round; human intelligence amplification projects I’d like to see by TsviBT

TsviBT — Thu, 30 Apr 2026 21:42:54 +0000

Summary

If you are interested in doing ambitious scientific research in areas listed below, and have a relevant project that needs funding, consider reaching out. My interest is in human intelligence amplification, though I believe that a variety of scientific projects are relevant, many of which are not specifically related to that goal.

Some areas, discussed below:

Soft questions: Strategy, financing, ethics, policy, governance, society, advocacy
Approach-general technical work on understanding and engineering brains and human intelligence
Reprogenetics-related projects, including reproductive epigenetics, chromosome engineering, microfluidics, cell engineering, and statistical genetics; see the full list here: “Projects that might help accelerate strong reprogenetics”
Brain interfaces
Brain cellular/molecular reprogramming
Brain gene editing

SFF’s themed grant round on Human Self-Enhancement and Empowerment

This year, the Survival and Flourishing Fund has a themed grant round, called Human Self-Enhancement and Empowerment (HSEE), which is “focused on organizations working to advance human self-enhancement and empowerment” and slated to make $2–4MM in grants this year. See their announcement here, with more information on what they are looking for: https://survivalandflourishing.fund/2026/application#hsee

The deadline for the HSEE Theme Round is July 8, 2026 at 11:59:59 PM PT.

I’ll be a Recommender for the HSEE Theme Round. That means that I and several other Recommenders, together with the Funder, will review and evaluate applications, and then through a collective process called the S-Process, make final recommendations of projects and amounts to fund. I do not have final granting power. Learn about how the S-Process works here: https://survivalandflourishing.fund/s-process

See information here https://survivalandflourishing.fund/faq and here https://survivalandflourishing.fund/2026/application#faq-applicants. Be sure to read the instructions carefully to check for eligibility and to submit properly.

In the rest of this post, I will express my personal opinions on what sort of projects might help accelerate human intelligence amplification. These opinions do not represent SFF. The reason I’m writing this now is so that people can have some sense about what projects I might be likely to recommend for funding. I don’t commit to recommending funding any specific project, and I categorically cannot commit to any specific project receiving funding from SFF, as all such final decisions are not up to me.

If you have a project that might be a good fit, feel free to reach out here or at my gmail address: tsvibtcontact (with the understanding that I may not have time to respond). Please also note that I will likely have a pretty high bar for what I’ll recommend funding for. I will also likely be pretty opinionated, but my aim is to use my opinionation in order to evoke the most important information quickly, rather than being stubborn or slow to change my mind. Therefore, I’m especially open to quick interactions about potential projects, e.g. “hey does [topic] seem like a thing you might possibly be interested in?” (with the understanding that the answer may often be “no”), or a phone call. My guess is that I personally am fairly unlikely to recommend funding for projects where I haven’t spoken to the applicant. I’m interested in projects that could produce information or ideas that would change my opinions, and projects that are predicated on a different opinion and where the difference can be argued for when applying for funding.

That said, I also hope that these descriptions of projects would be useful for people in general as some ideas for helping accelerate human intelligence amplification.

Generalities about my funding interests

Human intelligence amplification

I’m interested in accelerating HIA (human intelligence amplification). By that I broadly mean: greatly empowering humans with increased cognitive capacities, including philosophical problem solving, cleverness, coming up with great new ideas, wisdom, strategic ability, deep empathy, and so on.

Specifically, I’m interested in accelerating strong HIA (SHIA). SHIA is ill-defined, but means something like, “enabling people to be geniuses or world-class geniuses if they so choose”. Thus, I’m not very interested in projects where the best plausible outcome provides a small increase in brainpower, even though such projects could be great. (I’m not completely uninterested. A medium probability of a small but real increase in intelligence could definitely be worth it.)

Since SHIA is a difficult technology to develop, accelerating it could involved doing various indirect things, such as blue-sky research, public discussion, advocacy, media production, crafting regulation, improving the health of an industry, building out related business, low-probability research, etc. I’m quite open to such projects. However, I’m specifically interested in ones that have a significant contribution, in my estimation of probabilistic expectation, to the development of SHIA technology.

Intelligence in the operational sense of measured IQ is important and useful. In particular, it’s much easier to measure than other qualities, and therefore much easier to increase. I think this is a good and worthy goal that would have good consequences if achieved. For most practical purposes, it’s my main goal, due to feasibility. But IQ is far from all we care about even narrowly in the realm of SHIA (where “intelligence” is interpreted more broadly to include wisdom, strategy, empathy, sanity, reflectivity, etc.). So, while most of the following is focused on amplifying IQ, roughly speaking, I’m also interested in compelling projects related to understanding, measuring, and amplifying those other dimensions of cognitive capacity.

Approaches to SHIA

This is, a priori, a very difficult task because these aspects of humans are high-level properties of the most complex thing in the known universe: the human brain (and mind). There are broadly two ways to approach the technical problem of SHIA:

reverse engineering some key aspects that control / enable cognitive capacities, and then somehow engineering added support for those functions;
or else, copying nature’s work, i.e. looking at what genetic variants correlate with empirically measured cognitive capacity.

The latter approach is mainly about reprogenetics—enabling parents to make genomic choices on behalf of their future children. I favor this approach because I think it’s quite likely to work, technically speaking; also, the technology has a strong justification besides (S)HIA, namely disease risk reduction. It has the disadvantage that it can only benefit future children, not people currently alive; and it is harder to iterate on, compared to engineering methods.

For this reason, the following list of projects devotes a full section to reprogenetics, and only devotes subsections to other approaches. (There’s also a short section for ethics, strategy, policy, and other considerations.)

To avoid too much duplication, the following will somewhat presume my previous post “Overview of strong human intelligence amplification methods” as background. It’s not strictly necessary to read that post, but it contains many important elements of context.

What makes a good SHIA project

In short, I’d like to assist projects that actually address key bottlenecks to developing SHIA technology that benefits humanity.

An imaginary ideal application would do something like:

explain the central principles of a plausible specific technology that would constitute SHIA,
present a detailed technical analysis of a roadmap to a full implementation of that technology,
analyze the major obstacles to following that roadmap,
and then propose a project that concretely addresses those obstacles.

This is an unrealistic bar; a good realistic project might address one or more of the elements in that list. But hopefully this illustrates the sort of ambitious project I’d like to assist with.

Some more considerations:

SHIA technologies are probably difficult and complex technologies. That is, they integrate several scientific and technological areas, each of which has major challenges.
- That complicates the strategic picture around accelerating such a technology. The efficient way to work, is to look for places where there are fields of science and technology that already have a lot of work going on and that are making good progress, and then see how to accelerate the neglected work which, when combined with the powerful mainstream tech, could produce a successful last-mile SHIA project. In general I think there’s probably a lot of neglected work (though it may be unpromising because it’s difficult), just because people don’t normally try to work on something very ambitious / difficult like SHIA.
- Further, I will have to rely on expertise, so a good applicant would have to have a good way to credibly signal the relevant expertise. This can include credentials, past success, very clear and relevant analysis, social endorsement, and conversation.
In part because of the previous point, ethics, policy, and social issues are central aspects of a SHIA roadmap. It could be rightly considered a defection against society / humanity for a SHIA project to neglect the work of processing the nature, implications, and good use of SHIA technology along with other proper stakeholders (society at large, scientists in relevant fields, interest groups, regulators, bioethicists, etc.). Because it would be a defection to neglect that work, a SHIA project that neglects that work would naturally get much much less talent, funding, regulatory leeway, and societal support, and would probably fail.
Roadmaps are good. They are speculative and ambitious, which makes them likely to contain major inaccuracies, but it’s still helpful to try to see the way through to a successful SHIA technology, in order to orient to what matters in the area. I’d like to assist projects aimed at developing good roadmaps towards plausible SHIA technologies. Same goes for evaluations of central principles of SHIA technologies (e.g. “what sort of BCI could possibly enable SHIA, and how would we know”).
All that said, I’m definitely interested in assisting in-the-weeds / mundane / obvious / legworky science, if there’s a good case for its importance.

Strategy, financing, ethics, policy, governance, society, advocacy

For the most part, I suspect that it’s hard to productively investigate social issues around SHIA technologies until they are much closer to being acheived, and therefore more clearly resolved in terms of their safety profiles, misuse potential, social implications, etc. That said, social issues are also a key element of accelerating SHIA technology: if society is motivated to develop (S)HIA technology, the development will go faster, and if society is against it, it will go slower (and plausibly should go slower). So, I’m open to hearing about ways of addressing these issues about (S)HIA in general, or about specific technologies. For example:

What governance structures should be used for research groups and vendors developing (S)HIA technologies?
How should society develop and deploy (S)HIA technologies? E.g. what regulations and social attitudes should be used? What process should society use to decide this?
How to ensure that (S)HIA…
- ...doesn’t get misused?
- ...doesn’t harm recipients?
- ...isn’t marketed via fraud?
- ...is widely accessible?
- ...is broadly beneficial for society?
Doing fundraising, whether from VC, large private capital, philanthropy, or government sources. Also, mapping the views of these funders—what preconditions would have to be met to get much more funding for key science and technology research.
Doing advocacy for (S)HIA, broadly construed, e.g. producing media that explains and discusses (S)HIA to help society process these possibilities.
Gathering together people who have expertise and legitimacy to map out plausible futures of a given technology, discuss safety and ethics, make open letters, roundtables, reports, etc.
Talking to stakeholders (the public, advocacy groups, etc.) to understand opinions around (S)HIA, e.g. fears, hopes, questions, confusions, requests, etc.
How does SHIA affect existential risk from AGI? (See my essays here; they leave a lot of work to be done: https://tsvibt.blogspot.com/2025/11/hia-and-x-risk-part-1-why-it-helps.html https://www.lesswrong.com/posts/K4K6ikQtHxcG49Tcn/hia-and-x-risk-part-2-why-it-hurts )
There are many potential perils of reprogenetics; many of these apply to other SHIA methods. These perils could be analyzed (for severity, likelihood, and prevention). See “Potential perils of germline genomic engineering”.
Doing social organizing and momentum-building. E.g. conferences, forums, etc.

Approach-general technical work

There are likely many approach-agnostic investigations that would be quite helpful for eventual SHIA development. In particular, for non-reprogenetics SHIA approaches, we’d be doing some significant degree of “intelligence engineering”—coming up with specific ideas about how the brain works and how to augment its functioning, as opposed to copying nature’s work by just genomically vectoring for genes that are empirically correlated with target traits. (Even adult brain editing, which would leverage our partial understanding of the genetics of intelligence, would likely also require specific understanding of brain functioning, e.g. to decide which variants to ignore (e.g. if they are only active in childhood), to decide which tissues are most crucial to edit, and to foresee and prevent deleterious mental effects.)

Because we’d have to understand aspects of brain function that relate to intervening to increase cognitive capacities, we would have questions that might not usually be asked. I’d be interested in assisting scholarly reviews and analyses of key aspects of functional neuroanatomy, neurophysiology, cellular and molecular neuroscience, etc. E.g.:

Collating facts about brain development, macro and micro brain structures, brain function; connections between performance; connections between brain elements and genetic variants known to correlate with intelligence; etc.
Theorizing about the engines of human intelligence in terms of brain elments.
- E.g., what are the bottlenecks in brain function (energy, dendro/synaptogenesis, myelination, vascularization, etc.)?
- E.g., what are unique features of child / yound adult brains that make them so effective at learning? What are the neurobiological bases of these features?
- (These are heavily researched areas; but the SHIA lens would be novel, leading to a novel pattern of emphasis on different questions, facts, and deductions.)
Theorizing about how brain function might be significantly improved. E.g. adding something, tweaking something, amplifying something, etc.
Theorizing about what “disagreements with evolution” we might have, which could be exploited to increase brainpower.
- For example, human evolution might have built into our brains mechanisms for keeping the brain’s energy expenditure in check. But we might be confident that we can keep supplying plenty of calories steadily, and we might prefer for our brains to expend more energy for higher performance. (Cf. https://en.wikipedia.org/wiki/Cognitive_miser )
- For example, evolution might have designed us to broadly follow an arc of exploration (of cognitive algorithms) during childhood followed by exploitation in adulthood (i.e. just executing the cognitive algorithms we’ve settled upon). That would be in order to set us up for practical success, rather than staying in the realm of theory, ideation, science, abstract problem solving, and so on. But we may prefer to have our brains tuned for abstract thinking in adulthood.
Developing tests that can notice genuine increases in intelligence.

Reprogenetics-related projects

Because I think that reprogenetics is the technical pathway that’s most likely to succeed in enabling strong human intelligence amplification, this is the area I’ve thought most about, am most interested in assisting with, and have the most detailed projects.

Please see this document for a full list of reprogenetics-related projects I’m interested in assisting: “Projects that might help accelerate strong reprogenetics” (discuss on LessWrong)

Copying a short overview of relevant scientific areas from that document:

Reproductive epigenetics. (Epigenetic sequencing and editing, epigenetics of reproduction and the germline, stem cell culturing, gonadal culturing, stem cell reprogramming, gametogenesis, creating gametogonia.)
Chromosome engineering. (Targeted crossover, targeted elimination, targeted missegregation, chromosome transfer, microinjection, nuclear transfer, manipulating and sorting chromosomes, physics of individual chromosomes.)
Microfluidics. (Cell lysis, microwells, droplet creation / transportation / sorting / merging, PDMS design and manufacturing.)
Cell engineering. (Stem cell culturing, DNA damage and repair, CRISPR-Cas9 and transposases and other gene editing systems.)
Statistical genetics.

Non-reprogenetics approaches

Brain emulation

I’m pretty skeptical of these approaches, partly due to difficulty but mainly due to danger. To personally want to assist such a project, I’d have to first have my opinion changed on the safety question.

See my analysis here: https://www.lesswrong.com/posts/jTiSWHKAtnyA723LE/overview-of-strong-human-intelligence-amplification-methods#Brain_emulation .

Brain interfaces

See my analysis here: https://www.lesswrong.com/posts/jTiSWHKAtnyA723LE/overview-of-strong-human-intelligence-amplification-methods#Brain_brain_electrical_interface_approaches

I think these methods are interesting, but:

I don’t know about how to accelerate them.
- Neuralink seems like it’s following a good commercial ramp, but I don’t know about the scientific and technical path to much larger-scale connectivity. This makes it seem like both the research and the eventual uptake for users (also suppressed by the surgery involved) might be very slow.
- Analyses of where there are gaps in research, and funding for that research, would be interesting.
- For example, if there are really promising approaches to greatly decreasing the effective butcher number and increasing number of connections, e.g. through transplanted interface neurons, I’d be interested to hear about these.
It’s not clear how to greatly amplify human intelligence using these methods.
- I’m quite interested in assisting with projects that analyze this question.
- For example, I’ve suggested that large white matter investments by evolution imply that prosthetic connectivity might amplify intelligence (https://tsvibt.blogspot.com/2022/11/prosthetic-connectivity.html). Are there other such legible engineering bottlenecks, which could be supplemented technologically?
- See for example “Principles of Neural Design” by Peter Sterling and Simon Laughlin. Do any of those principles, or other similar principles, suggest ways to substantially increase peak brain function?
- What about human-human networking, or networking with externally supported brain tissue? What about machine learning acceleration for some brain operations?

I’m quite interested in projects that could either potentially have results that change these opinions of mine, or are predicated on a different opinion and could explain the difference when applying for funding.

Brain cellular/molecular reprogramming

See https://www.lesswrong.com/posts/jTiSWHKAtnyA723LE/overview-of-strong-human-intelligence-amplification-methods#Signaling_molecules_for_creative_brains.

All the existing work along these lines that I’m aware of is not ambitious enough—it doesn’t aim for SHIA. But I’m interested in ambitious ideas, such as somehow safely loosening perineuronal networks, or proliferating longer-range connections, or increasing genesis of dendrites / synapses / axons / neurons, or coaxing neurons themselves to regress to a more childlike or young-adult regulatory state, or specifically upregulating some child-brain-like GRNs or other characters, or things like that.

I’m skeptical because lifetime development seems like it carries many major irreversible changes, and making large-scale changes via cellular reprogramming would have huge risks of causing damage to existing mental structures and risking mental illness.

Brain gene editing

If anyone has a good way to accelerate research to find reasonable solutions to the delivery problem, let me know.

I’m also interested in research that could get a better picture of how large the gains would be, given that developmental windows would have passed for adult intervention. I’m skeptical that it could have large effects, but interested in more evidence.

Cf. https://www.lesswrong.com/posts/jTiSWHKAtnyA723LE/overview-of-strong-human-intelligence-amplification-methods#Adult_brain_gene_editing.

Other

See also “Massive neural transplantation” here: https://www.lesswrong.com/posts/jTiSWHKAtnyA723LE/overview-of-strong-human-intelligence-amplification-methods#Massive_neural_transplantation. And see e.g. Revah et al. (2022)^[1]. If this worked, one could imagine transplanting gene-edited neurons.

I’m interested in any approaches that could lead to strong human intelligence amplification. I don’t know of plausible such approaches other than the ones I listed, but I’m open to being convinced otherwise.

I’m also interested in approaches that have a pretty high chance of leading to more modest but genuine HIA. (By “genuine” I mean “actually increases ability to solve hard thinky problems”, rather than for example keeping you awake longer, having a higher mood, removing brain fog to bring you back to baseline, increasing performance on some tests that do not generalize to solving hard thinky problems, making you feel less inhibited, and so on.) I’m open to being convinced about some method, such as transcranial electromagnetic stimulation or selective HDACis or etc.; but it would take some convincing. Likewise, it’s conceivable that software tools could reach my bar, but it would have a large burden of proof to overcome. For example, powerful ways of communicating and coordinating, such as content filtering that allows group epistemics to be much more efficient and sane, could be interesting.

↩︎
Revah, Omer, Felicity Gore, Kevin W. Kelley, et al. “Maturation and Circuit Integration of Transplanted Human Cortical Organoids.” Nature 610, no. 7931 (2022): 319–26. https://doi.org/10.1038/s41586-022-05277-w.

How to actually give money away by NickAllardice

NickAllardice — Thu, 30 Apr 2026 18:15:42 +0000

This was originally posted here. It’s written for an audience that’s not deep in the weeds of EA giving theory/culture, but a few people suggested I post here as there’s much that’s additive to or divergent from some common EA practices. Feedback / disagreements welcome! Also my first time posting here. Hi!

Most people who intend to give large amounts of money away never actually do.

The money sits. In donor-advised funds, in “someday” plans, in good intentions. This is the default pathway—not an edge case—and if you don’t design against it, it will happen to you too.

I’ve watched this from every angle. As CEO of Change.org we processed 5M+ donations. I now run GiveDirectly; 160K people have donated, dozens regularly give $1M+, and some give $50M-$100M+ at a time. I’ve advised two of the biggest philanthropic institutions in the world (Coefficient Giving and GiveWell) on donor engagement and growth, and sat on half a dozen nonprofit boards. I’ve also been giving away 10-20% of my annual income for 17 years; I grew up low-income, so this started as a very modest amount but has added up.

Across all that, there are patterns and common mistakes. I increasingly get asked for advice on how to give very large amounts of money away—this post captures what I most commonly share.

It’ll be most useful for people giving at significant scale, especially those who’ve come into serious money from tech, AI or finance, where the combination of concentrated equity, compressed timelines, and limited counterfactual time makes deferring your giving especially common. But the ideas apply to anyone aiming to give thoughtfully.

1. The default pathway is to delay. Fight this.

Here’s what happens: You care about giving. But you don’t have expertise or dealflow in the areas you care about. You don’t have much time to build that expertise, because you’re running a company, working hard, or raising kids. So you don’t feel confident enough to commit.

Underneath the structural reasons, there’s an emotional one. I grew up without much money, and there’s a part of me that still expects it all to disappear at any moment. Most people I talk to have a version of this, regardless of background. It makes giving feel like loss, even when you can clearly afford it.

So you defer. Your wealth grows, which makes the stakes of each decision feel higher, which makes you more cautious, which means you give less as a proportion of what’s possible. It’s a flywheel that runs on good intentions.

If you only take one thing from this post, make it this: failing to follow through on your intention to give is the default pathway. It has happened to many people who signed the Giving Pledge, and it will happen to you unless you take it seriously.

I’ve lived this. A few years ago, I set aside about 40% of my annual giving budget for what I thought of as “systems-changing” ideas; bets with extremely high upside and high risk. I believed in the commitment. But I didn’t have the dealflow, or a process for choosing between opportunities when they did come along. I had the intention but not the system. The money just sat.

Accept that no matter how well intentioned you are, the default pathway is to delay, forever. The more you appreciate the structural incentives, the better you can design to combat them. The rest of this post is about how.

Finally: set your annual giving budget high enough that it matters. If you’re still earning, 10-20% of income is the classic anchor. If your wealth is primarily in assets, treat 5% of net worth as the floor, not the ceiling. Don’t wait for an exit—there are well-worn paths to give equity directly, often with far better tax treatment than giving cash after a liquidity event, and it gets capital to work now rather than at some hypothetical future date.

Notice what 5% actually means: on $100M, that’s $5M a year. Your investments are still compounding at 7–10%, and if you’re not retired, you’re still earning. So even at 10% giving, your net worth and ability to give more almost certainly keeps growing. The givers who inspire me go even further: they define the number they need to live very comfortably for life (and leave to any kids), then plan to give everything above that line away before they die.

2. Pick a few causes and write them down.

The causes you direct your money toward are perhaps the most consequential decision you’ll make; far more than any individual donation. And yet most people put all their effort into evaluating individual donation opportunities, not deciding what causes to support.

Should you give to global health? Animal welfare? AI safety? Protecting democracy? Girls’ education?

The right answer will be some function of your values and what you think your donation could achieve. Some considerations that have been influential for me:

Luck. I feel extraordinarily lucky to have been born in Australia. Even though my parents were low-income by Australian standards, compared to most people globally and throughout history my access to health, income, security and opportunity is incredibly high. I feel a responsibility to pay that forward.
Value for money. I want my dollar to have as much impact as possible. That means I skew toward opportunities that support the world’s poorest, where $1 USD can stretch 250x further. If impact per dollar matters to you, the case for directing a significant share of your giving to the world’s poorest is very strong.
Humility. The older I get, the less black and white (and more grey) I see. I have uncertainty about what values are most important or what’s most likely to work. I’ve also become more skeptical of my ability or right to make decisions on behalf of others. This is why a meaningful portion of my giving goes to unconditional cash transfers that (in addition to being highly evidenced and cost-effective) let people in poverty decide what’s important themselves.
Motivation. Giving is a marathon, not a sprint. Giving in a way that nourishes you and gives you joy is a win/win. For you and for the world.

The end result for me: 90% of my giving aims to raise the floor of the human experience by reducing needless suffering. The other 10% is for people I care about and believe in, even if their work reflects a different worldview—a friend starting a nonprofit, a cause someone I trust is passionate about, an ask that moves me. I think of this as the human part of my giving. Not data-driven, but humble, and important for keeping me connected and motivated.

It’s completely legitimate to pick other things. Whatever you choose, write it down. The people who struggle tend to be open to everything, which in practice means they’re committed to nothing.

3. Build a portfolio across causes and risk/return.

Once you’ve picked your causes, treat giving like an investment portfolio — not just what you’re investing in, but what risk/return profile you want across your bets.

Some people work backwards from a specific goal: end lead poisoning globally, replace factory farming, or end poverty in a specific country. If you have that kind of conviction, the goal dictates the portfolio—you donate to whatever combination of research, advocacy, and delivery gets you there.

If you don’t have a single goal, construct a portfolio across causes and risk/return profiles. My own split, for every $100 I give:

$45 to high-risk, high-return bets: R&D, policy change, technology, people trying something new that could be transformative but might not work. Think of this like venture capital investing.
$45 to high-confidence, modest-return giving: interventions with strong evidence behind them, where I know the money will do good even if the upside is capped. More like bonds or index funds.
$10 personal and relational: people I care about, causes that move me, asks from friends.

Your portfolio can look very different. Many I talk to give a large share internationally but reserve some portion for local community or politics. The framework is flexible. What matters is that you’ve pre-committed to some structure, so when an opportunity arrives you already know which bucket it’s in and how much room is left. Decision costs drop immediately.

A warning on Donor Advised Funds. When you give to a DAF you get the tax deduction, which creates a psychological sense of completion—you’ve made the sacrifice, the money feels “committed”—and the actual giving recedes into the future. Most DAF providers also earn from holding your balance, not from getting money out the door. If you need a DAF for tax-timing reasons, newer flat- or low-fee options like Daffy, Charityvest and Every.org are worth looking at. But the deferral risk is the same regardless of provider.

4. Use the index funds of giving.

In investing, the index fund has basically won. Most professional stock pickers can’t consistently beat the market; after fees, the majority of actively managed funds underperform a cheap S&P 500 tracker. For most investors, the smart move is to stop trying to be clever and put your money where decades of evidence says it’ll do best.

The same logic applies to giving. Most people who want to give well try to become experts in the problem they care about; evaluating organizations, comparing methodologies, reading impact reports. For a few with unusual time and skills, that might be the right approach. For most, it’s hubris and a recipe for indefinite delay.

There are institutions that function like the index funds of giving.

GiveWell was the first one I chose to regularly give to, more than a decade ago. If you align with their values (heavily weight saving lives; value evidence and certainty), they’re unmatched in research and rigor.
Direct cash transfers also fall into this category for me (giant caveat: I run GiveDirectly, so weigh this accordingly—but it was a major part of my portfolio for years before I joined).
For those more focused on the US, Robin Hood Foundation and Blue Meridian Capital Partners are worth a look; Robin Hood pools donor capital to fight poverty in New York City with rigorous benefit-cost analysis, and Blue Meridian makes large concentrated bets on organizations that can scale nationally.

Your comparative advantage is having capital and willingness to deploy it—not replicating analysis others have spent careers building. If you give outside the index funds, be clear on why: unusual dealflow, personal insight, higher risk tolerance, ability to move faster than institutions can. That’s why $45 of my $100 goes to higher-risk bets. But it’s so tempting to make these decisions from ego, so I recommend a meaningful portion goes to the index funds.

5. For the love of god, don’t over-staff.

When people get serious about giving at $10M+ scale, the first instinct is to hire: a philanthropic advisor, a program officer, maybe a small foundation team. I’d push back hard.

The more institutional infrastructure you build around your giving, the slower the money moves. There’s an incentive problem at the heart of it: nobody on your foundation staff gets fired for not having enough impact. But they could lose their job over a grant that goes sideways, or for moving too fast without enough diligence. The incentive structure rewards caution, not impact—and institutional gravity pulls relentlessly toward risk aversion.

I’ve seen this repeatedly: someone commits to giving at scale, hires a team to help them do it well, and within a few years the team has become the primary obstacle to capital getting out the door. The staff are doing their jobs… they’re being careful, doing diligence, managing risk—and the result is that less money reaches anyone.

My recommendation: one trusted, smart person at most. Someone whose values and ways of working align with yours, who’s scrappy and autonomous and can move quickly. They should feel accountable for how much good your giving does, not just for avoiding mistakes. I know at least one individual advisor who moves ~$100M annually on behalf of their donor because they’re deeply aligned, have optimized for fast high-quality decisions, and stayed lean. I’d be skeptical of any setup with more than one staff member per $50M in annual giving.

If you’re unsure about hiring an individual, there are good advisory firms: Renaissance Philanthropy, Ark Philanthropy, Coefficient Giving, Jasper Ridge Partners, and Longview Philanthropy. One important distinction: keep your own giving operation lean, but don’t starve those you fund. Be lean on your side of the table so you can be generous on theirs.

6. Make giving a recurring event, not a to-do.

If giving is always on your to-do list, it will always lose to whatever feels urgent at work or at home. So stop treating it like an ongoing task and treat it like an event.

Block one or two moments a year (alone, or with your advisor if you have one) where you compile a set of opportunities across your portfolio buckets and make decisions. Do real diligence, but constrain the set. Research consistently finds that donors given a large open-ended set of options are less likely to give than those given a small pre-vetted set. More options means more evaluation burden, more potential for regret, more reason to defer. Constraint is your friend.

If you want external structure, programs exist for exactly this. The TED Audacious Project curates roughly ten deeply-vetted big-bet ideas a year. ICONIQ Impact does something similar in different themes.

Between those windows, evaluate opportunities reactively as they come. Someone you trust recommends an organization, a crisis happens, something compelling lands in your inbox. Evaluate it against your portfolio and thesis. If it fits and there’s room, fund it. If not, pass.

The key rule: anything you haven’t allocated by year-end goes to an ‘index fund’ of giving. No exceptions. This is what keeps the deferral flywheel from spinning up again. You can always refine your higher-risk bets over time, but the money should move every year regardless.

A few final hot takes

Places where I diverge from conventional wisdom, or where I think the philanthropy sector’s incentives pull in the wrong direction:

Prefer simplicity. In theory of change, in cost-effectiveness models, in operating models. The more complicated any of them are, the more places to be wrong. I’d rather fund a simple idea that seems less exciting than an elaborate plan built on 40 assumptions.
People and leadership matter more than detailed plans. The single most underrated factor in whether a philanthropic investment works. I’d back an A+ team with a hazy high-level plan over a B+ team with something fully baked, every time.
Run a tight process. I’ve raised VC capital, and I much prefer it to raising philanthropic capital. Why? VCs don’t waste your time. Fast VCs commit millions in days and tens of millions in weeks. They’ve designed their diligence and decision making processes to respect the entrepreneurs time; you can do the same without compromising on the quality of your decisions.
Give unrestricted. When you restrict a grant, you’re telling an organization you know better than they do how to spend their money. You almost certainly don’t. The best leaders see opportunities and bottlenecks you can’t see from the outside. Trust them.
Reward transparency and honest failure. If a grantee tries something, it doesn’t work, and they learn and improve — that’s a good outcome. Fund them again. But be willing to kill investments where plausible progress isn’t materializing. The sector’s instinct is to keep funding what isn’t working because stopping feels like admitting failure.
Be skeptical of layers. It’s not unusual to see chains like: donor advisor → donor collaborative → international NGO → national NGO → community-based org → local contractor. Every institution in that chain has an incentive to justify its own existence. So much value gets lost. Shorten the distance between your dollar and the person or problem it’s meant to help.
Don’t over-coordinate with other donors. Share opportunities and learnings, absolutely. But don’t galaxy-brain exactly how much should go to cause A vs. cause B. Diversity of worldview and theory of change across donors is a feature, not a bug.
Don’t force collaboration among grantees. Donors love the idea of their grantees working together. In practice it’s usually a waste of everyone’s time. Diversity of approaches and healthy competition between organizations tackling the same problem is often more productive.
Don’t get attached to tax deductibility. Some of the highest-impact opportunities (political giving, impact investing, direct support to individuals) aren’t tax deductible. If you’re optimizing for the tax benefit rather than the impact, you’re leaving a lot of good on the table.
Fight incrementalism. The nonprofit sector is structurally risk-averse. Funders want proven models. Grantees propose what they think funders will approve. The result is a lot of modest, incremental work. Push against that. Look for solutions that, if successful, could scale dramatically. There’s a giant gap there.

That’s what I’ve got for now. None of this is truth; many smart people disagree with me or take a different approach. But these are the lessons I’ve come to believe after 17 years of giving and all my work in this world.

If you want to think any of this through—portfolio construction, finding an advisor, specific recommendations across buckets—I’m happy to talk. Email me at nick@nickallardice.com

Arcadia Impact AI Governance Taskforce Summer 2026 cohort—applications closing 10th May by Ben R Smith

Ben R Smith — Thu, 30 Apr 2026 16:17:33 +0000

🚨 AI Governance Taskforce Summer 2026 applications are open – deadline 10th May.

For senior or mid-career professionals transitioning into AI governance, focused on mitigating catastrophic risk.

This 12-week remote part-time opportunity is designed for experienced professionals who want to apply existing domain expertise to frontier AI governance, addressing acute and systemic catastrophic risks. You will work in small, high-calibre, multidisciplinary research teams producing policy-relevant research, led by Research Team Leaders and guided by expert practitioners based at respected organisations.

Projects and experts:

🔹 Building anticipatory capacity: frontier AI risk scenario frameworks for government
Elizabeth Dearden-Williams (UK AISI)

🔹 A playbook for corporate management of frontier AI incident knowledge under legal constraint
Sean McGregor (AI Incident Database)

🔹 Opportunities for advancing red lines by country – middle powers legislation
Caio Machado (The Future Society)

🔹 No treaty or weak treaty: Trade-offs between pathways to an international AI governance agreement
José Villalobos (The Future of Life Institute)

🔹 Testing and Evaluation of Agentic AI Systems in Military Command and Control
Heather Frase (Virginia Tech National Security Institute)

🔹 ‘If Anyone Builds It, Everyone Needs to Understand It’ — Understanding as a product of frontier AI safety review – building shared comprehension for collective governance
Professor Robin Bloomfield (City University, University of London)

What this involves:
👥 Applied research in small teams
📄 Policy-relevant outputs (papers, briefs)
🌍 Fully remote, global, part-time (8 hrs/week), 12 weeks
💼 Selective, unpaid research fellowship
📅 Deadline: Sun 10 May 2026
📅 Cohort dates: 1 June − 21 August 2026
🔗 More info and apply: https://www.arcadiaimpact.org/ai-governance-taskforce

For queries, please email:
Ben R Smith, Taskforce Lead
ben [at] arcadiaimpact [dot] org

Book Review: All the Lives You Can Change by Bentham’s Bulldog

Bentham's Bulldog — Thu, 30 Apr 2026 14:28:18 +0000

Crosspost.

(Reminder: the farm bill which is being voted on imminently would destroy most state level animal protections and be the worst law for farm animal welfare ever passed. Please, please, contact your representatives and tell them to vote no on it—more details here, including a bunch of other activities that are even higher impact. The house votes on this today, so this is the last day you can productively call your representatives.)

Effective altruism is a social movement that’s about trying to do good as effectively as possible, with charity, career, and life projects broadly. If you go to a random local effective altruism event, most of the people there will be atheists, even in a country that’s mostly Christian. But this isn’t because of any deep conflict between Christian ideas and effective altruism. I’m in a Facebook group with a bunch of Christian philosophers, and about half of them are effective altruists in some form.

Similarly, Aron Wall who is among the most devout Christians I know, is an effective altruist and gives his money to the most effective charities he can find. My friend who runs the YouTube channel Apologetics Squared is another very religious Christian who supports effective altruism—he even made a video about how to maximally effectively spread the Gospel (hint: it’s not getting into theology arguments on the internet). Smart, philosophically-reflective Christians who take their faith seriously are effective altruists at astonishing rates.

Enter All the Lives You Can Change, released this week! It’s written by David Zhang, Dominic Roser, and J.D. Bauman. I met J.D. at an effective altruism event, and was immediately impressed by him—he seemed to radiate a kind of deep-rooted moral decency and conviction. I briefly met Dominic too at an EA for Christians event that they were generous enough to let me attend even though I’m not Christian. Incidentally, that event was where I learned from Dustin Crummett about shrimp welfare! The authors are all part of an organization called EA for Christians, about trying to promote effective altruism to Christians. This book is their attempt to popularize those core ideas.

JD holding up the book.

And note, these aren’t the kinds of wishy-washy Christians whose theology stops at “Jesus was nice to people, so you should be nice too.” No, these are people who actually take Christianity seriously, who reflect seriously on what Jesus would want them to do and try hard to do it. They think effective altruism should be a core part of the Christian life. The case that they lay out for this position is very compelling.

The book was a delight to read. I finished it in two days. Like most EA books, it contained a bunch of cool stories of people doing good effectively. Like most Christian books, it included a lot of quotations of scripture. I came away more convinced that Jesus would have been supportive of effective altruism. Many of his moral claims echoed core EA claims.

Critics of EA complain that asking you to give away 10% of your wealth is too demanding, despite leaving them still richer than almost everyone who has ever lived? Oh, that’s too demanding? Try getting nailed to a cross! The whole Christian story is that morality is sufficiently demanding that none of us live up to it very well! And Jesus repeatedly and emphatically recommends that people give away their wealth, saying things like:

Whoever has two shirts should share with the person who has none, and whoever has food should do the same.

Jesus went so far as to say “If you want to be perfect, go, sell your possessions and give to the poor, and you will have treasure in heaven. Then come, follow me.” Now, lots of Christians will interpret this passage to just be a highly specific command for that particular guy. This seems like cope. Really, the passage was included just as a specific message to that one guy which doesn’t apply more broadly? Does anyone find this kind of exegesis plausible in any other context? And don’t take it from me, here’s what Aquinas said on the matter:

Now, according to the natural order instituted by divine providence, material goods are provided for the satisfaction of human needs. Therefore the division and appropriation of property, which proceeds from human law, must not hinder the satisfaction of man’s necessity from such goods. Equally, whatever a man has in superabundance is owed, of natural right, to the poor for their sustenance. So Ambrosius says, and it is also to be found in the Decretum Gratiani: “The bread which you withhold belongs to the hungry; the clothing you shut away, to the naked; and the money you bury in the earth is the redemption and freedom of the penniless.”

“Do I really have to care about effectiveness?” you ask! “Yes,” Jesus answers, “you’re supposed to love your neighbor as yourself.” When deciding what to do for your own benefit, you don’t just do random things, but what benefits you most. “Okay, but surely my neighbors are just the people nearby and not people far away, so I don’t have to give to the other side of the world?” “No,” Jesus answers, and tells a long parable to show just how wrong you are:

²⁵On one occasion an expert in the law stood up to test Jesus. “Teacher,” he asked, “what must I do to inherit eternal life?”
²⁶“What is written in the Law?” he replied. “How do you read it?”
²⁷He answered, “‘Love the Lord your God with all your heart and with all your soul and with all your strength and with all your mind’^[^a^]; and, ‘Love your neighbor as yourself.’”
²⁸“You have answered correctly,” Jesus replied. “Do this and you will live.”
²⁹But he wanted to justify himself, so he asked Jesus, “And who is my neighbor?”
³⁰In reply Jesus said: “A man was going down from Jerusalem to Jericho, when he was attacked by robbers. They stripped him of his clothes, beat him and went away, leaving him half dead. ³¹A priest happened to be going down the same road, and when he saw the man, he passed by on the other side. ³²So too, a Levite, when he came to the place and saw him, passed by on the other side. ³³But a Samaritan, as he traveled, came where the man was; and when he saw him, he took pity on him. ³⁴He went to him and bandaged his wounds, pouring on oil and wine. Then he put the man on his own donkey, brought him to an inn and took care of him. ³⁵The next day he took out two denarii and gave them to the innkeeper. ‘Look after him,’ he said, ‘and when I return, I will reimburse you for any extra expense you may have.’
³⁶“Which of these three do you think was a neighbor to the man who fell into the hands of robbers?”
³⁷The expert in the law replied, “The one who had mercy on him.”
Jesus told him, “Go and do likewise.”

As the authors note, this passage clearly indicates that your neighbor isn’t just the people nearby that you want to help. The Samaritans and the Israelites were mortal enemies. Your neighbor is the stranger, the foreigner, the people that your society isn’t motivated to help. The whole point of the parable is that your neighbor includes those that you’re not motivated to help.

“Okay,” you wonder, “maybe I should give away a bunch of my wealth, and care about foreigners. But do I really need to care that much about effectiveness?” In response, Jesus tells a bunch of parables that capture the importance of effectiveness. He recommends people be “shrewd as snakes and as innocent as doves.” In other words, one is supposed to be careful and smart about doing good.

In Luke 12:6–7, Jesus asks “Are not five sparrows sold for two pennies? And not one of them is forgotten before God. Why, even the hairs of your head are all numbered.” Insofar as God cares about every one of us individually, rather than seeing us as a homogenous mass, then the numbers of people we help really counts. As the authors put it:

A quote, often misattributed to Stalin, says the death of one man is a tragedy, but the death of a million is a statistic. The first part is certainly true: Every death is a tragedy. It’s also true that it can feel like a million deaths are just a statistic. But our instincts do not always capture reality; in the eyes of God, a million deaths are not a statistic, but a million tragedies.

Or, as they say elsewhere “If every person matters, then surely the prevention of a million deaths matters a million times. For the 513,434th victim, it makes all the difference in the world whether those who have the power to prevent death focus on saving many or saving just a few.”

The authors touch on most of the EA book quick hits. They make sure to mention that you can save lives for thousands of dollars and do incredible amounts of good with your career. While we often wonder what we’d have done in the past, e.g. in Nazi Germany, if we could have saved many lives, we have similar opportunities in the world today to save hundreds of lives. It is a law of nature proved beyond eight sigma that any EA book must include the following chart—this one is no exception.

This chart shows across different interventions to improve learning just how much more effective some interventions are than others.

The book also recounts a number of moving stories. Corrie ten Boom was a religious Christian who worked hard to save as many Jews as possible during the holocaust at great personal risk. About 800 people were saved from torture or death because of her. Had she only cared about doing some good, rather than carefully optimized for doing as much of it as possible, hundreds of extra people would have been tortured and murdered by the Nazis.

The book also covers a lot of topics that are important to Christians but that you won’t find in standard EA books. It talks about how faith can make it easier to make sacrifices for the good and discusses God’s special concern for caring for the meek and vulnerable. It analyzes interventions to maximally effectively spread the Gospel, how those compare to other interventions, and how Christians should think about animal welfare and Longtermism. It really engages, in a deep way, with how Christians specifically should be EAs, not just trying to market standard EA to Christians.

I am not a Christian, though I am an effective altruist. I came away from the book with a deeper respect for Christianity and a deeper vision of how it coheres with doing good effectively. If you are a Christian—or even if you are not—I cannot recommend it enough. If nothing else, throughout the book, I was impressed by the degree to which a lot of what Jesus said seemed to make deep sense. For a book to achieve that, and be well written, and tell you how to do good with your life, and carefully address serious and difficult questions, is a stunning success.

Most of all, in the book, and in the movement of Christians for EA, I saw the making of a beautiful thing—a kind of Christianity that, theological objections aside, is stunningly virtuous, and decent, and good. This beautiful thing has been largely non-existent for senseless, sociological reasons. It’s high time that changes. Help get and promote the book not just because it’s an awesome book full of useful information and advice, but so this beautiful thing may emerge and grow, so that the poverty and death and suffering that is within our reach can be driven out, and a version of the Gospel faithful both to its central message and what is most important in the world can emerge more fully.

If your life goal is to donate, what kind of organisation should you found? by jy naha

jy naha — Thu, 30 Apr 2026 13:37:52 +0000

Epistemic status: Medium confidence, model-based reasoning with limited empirical grounding. Heavily influenced by power-law intuitions, startup statistics, and EA cost-effectiveness framing. Most uncertain about counterfactuals and founder-specific effects.

Summary

If you want to maximise lifetime charitable donations through founding, which organisational structure is best?

I compare four paths:

For-profit founding to give
Charity entrepreneurship
Social enterprises (CICs)
Effective giving organisations

My current view:

For risk-tolerant founders, for-profit founding-to-give likely maximises expected value, driven by extreme right-tail outcomes.
For those optimising for probability of achieving meaningful impact, charity entrepreneurship appears strongest.
CICs are often structurally dominated.
Effective giving organisations are highly sensitive to counterfactual assumptions.

Why this question seems neglected

Within EA, there is substantial discussion of:

Earning to give vs direct work
Cause prioritisation

But relatively little on organisational structure conditional on founding.

This seems non-trivial because structure directly affects:

Incentives (equity vs salary vs mission lock)
Access to capital (VC vs grants vs donations)
Scalability constraints
Timing and magnitude of donations

Programmes like AIM’s Founding to Give are relatively recent, which weakly suggests this space has not been fully explored.

Framework: ITN (with caveats)

I evaluate options using:

Importance: scale of impact
Tractability: probability of success
Neglectedness: marginal resources

However, ITN has limitations:

Neglectedness is often endogenous to tractability
It does not handle heavy-tailed distributions well
It abstracts away from founder-specific comparative advantage

So this analysis should be read as a rough scaffold, not a decision procedure.

1. For-profit founding to give

Claim: Highest expected value for the right tail of founders.

Importance: Extremely high
Startup returns follow a power-law distribution. A small number of exits generate the majority of value. Donating even a fraction of a large exit dominates most alternative paths.
Tractability: Low
Base rates are poor. Most startups fail or produce modest outcomes.
Neglectedness: Moderate
Increasing attention within EA, but still less explored than charity entrepreneurship.

Key considerations:

EV is driven almost entirely by rare successes
Time to impact is long and uncertain
Capital markets amplify scale in ways philanthropy typically cannot

Failure modes:

Value drift after financial success
Pledge non-fulfilment
Burnout before exit
Selection effects, people overestimating their probability of success

A central concern is that expected value may be illusory at the individual level if founders systematically overestimate their odds in a heavy-tailed distribution.

2. Charity entrepreneurship

Claim: Best option for maximising probability of impact.

Importance: Very high
Creating a top-tier charity can shift large flows of funding. Cost-effectiveness differences across charities are large.
Tractability: Moderate to high
Incubators provide:
- Cause selection
- Seed funding
- Mentorship

This reduces early-stage uncertainty.

Neglectedness: Low
The space is relatively mature within EA.

Key considerations:

Faster path to impact than startups
Strong alignment between mission and structure
Less exposure to extreme downside risk

Failure modes:

Intervention fails at scale
Fundraising constraints limit growth
Counterfactual replacement, someone else would have founded something similar

Relative to for-profits, this path trades away upside tail outcomes for much tighter variance.

3. Social enterprises (CICs)

Claim: Often a dominated middle ground.

Importance: Moderate
Profit caps and asset locks limit total extractable value for donation.
Tractability: Moderate
Easier to set up, but strategically constrained.
Neglectedness: Low to moderate

Core issue: Misaligned incentives and constraints

Less attractive to investors than for-profits
Less eligible for funding than charities
Limited ability to pivot due to asset lock

Failure modes:

Stuck between commercial and charitable logics
Subscale outcomes in both domains

There may be niche cases where CICs are optimal, but as a general strategy for maximising donations, they appear structurally disadvantaged.

4. Effective giving organisations

Claim: Potentially high leverage, but highly uncertain.

Importance: Potentially very high
If each dollar spent moves multiple dollars to effective charities, impact scales multiplicatively.
Tractability: Moderate
Easier to launch than direct interventions, but success depends on donor behaviour.
Neglectedness: Moderate

Crux: Counterfactual donations

Are these organisations generating new giving?
Or redirecting existing EA-aligned funds?

If mostly the latter, impact is significantly lower than it appears.

Failure modes:

Negative net impact if costs exceed funds moved
Donor substitution effects
Overestimating influence on donor decisions

Comparative intuition

The trade-off can be framed as:

For-profit founding: high variance, extreme upside, low probability
Charity entrepreneurship: lower variance, higher probability of solid impact
CICs: constrained on both axes
Effective giving orgs: leverage depends on uncertain counterfactuals

This resembles a classic risk vs reliability frontier, rather than a single dominant option.

My current view

If you have strong entrepreneurial ability and high risk tolerance, for-profit founding-to-give likely maximises expected donations.
If you want robust, high-probability impact, charity entrepreneurship is more attractive.
CICs are rarely optimal for donation maximisation.
Effective giving organisations are promising but hinge on empirical uncertainties that are not well resolved.

Key uncertainties / what would change my mind

Evidence that founders systematically overestimate their position in the startup distribution, reducing EV below charity entrepreneurship
Data showing Founding to Give participants achieve meaningfully higher success rates than baseline founders
Strong evidence that effective giving organisations generate additional (not substitutive) donations
Cases where CICs demonstrate clear comparative advantages in scaling impact
A framework that incorporates founder-specific fit and selection effects more explicitly

This post was published after 5 weeks of the Intro EA Readings, as part of the Effective Altruism Cambridge Project-Based Fellowship. Learn more here: https://www.eacambridge.org/ and reach out to jianxin@eacambridge.org if you’d like to learn more about the fellowship

Legal Alignment—As a Baseline Mechanism for AI Misalignment by FuzzyAlgorithm

FuzzyAlgorithm — Thu, 30 Apr 2026 13:35:27 +0000

Legal Alignment—As a Baseline Mechanism for AI Misalignment

One of the main challenges with the frontier AI models today is not that they are impeccably powerful systems but that they often get misaligned in such fashions that their impact is heavily felt in real-world. Presently, models such as Gemini, Claude and GPT can produce biased outputs, untruthful content, privacy leaks, sycophantic responses, and even help in hacking or making of bioweapons. These failures are not isolated bugs but evidence that current alignment practices are insufficient, incomplete and at times opaque.

At this juncture, one of the ways to effectively deal with AI misalignment is by incorporating legal alignment framework. Legal alignment as a solution, should not be seen as similar to set of rules and regulations, rather, it is precisely about integrating legal and lawful methods right into the designing and operation of such AI systems. AI misalignment isn’t just a technical fallout but also a governance issue. If a model starts to behave inconsistently, bypassing its own built safeguards or assisting in unlawful or harmful conduct based on prompts, then the major concern is to not only make such a model safe but also to devise a normative framework that guides its behaviour to stay uptight. Drawing on recent studies and scholarships, one plausible solution to this problem is legal alignment, which involves designing of AI models in accordance with law, legal principles, rules and methods. It can help the system to adopt normatively desirable behaviour / outcome and comply with legal specifications.

What is misalignment and the limitations?

Misalignment occurs when a model performs exceptionally well ordinarily but fails when exposed to ambiguity, pressure or any adversarial prompting. The instances of bias, hallucination, persuasion and active concealment show that misalignment is more than rule-breaking, it takes the form of context-dependent, subtle failure. Crucially, such failures are not outrightly random, but it underpins the core architecture of large language models (LLMs). Generally, LLMs are trained to achieve usefulness and coherence than legality, normative consistency or truthfulness. Resultantly, techniques like reinforcement learning or post-training filters are largely fragile and superficial. It is imperative to understand that misalignment occurs as a spectrum than a singular phenomenon. It includes on one end, direct unlawful outcomes, such as cyber intrusion or fraudulent representations, and on the other end, there are structural failures, when models don’t acknowledge legally sensitive information, or cannot reason legality to recklessly give outputs semantically similar.

Further, implementation of alignment practices/strategies have inherently been ineffective because of their under-specified and privatized nature. Presently, most of the frontier models are controlled by internal policies—safety filters, model specifications and constitutional guidelines—developed and administered by private actors. Although these mechanisms have benefited in improving baseline performance of models, they are largely marred by three shortcomings.

Firstly, these mechanisms lack normative clarity. It means that constructs of ‘helpfulness’ and ‘harmlessness” are essentially vague to them, and most alignment techniques conform with ‘company-written alignment policies’ rather than addressing people’s (sometimes) conflicting and diverse values. Secondly, they are devoid of institutional legitimacy. The standards of alignment are effectuated by private actors, which makes it fundamentally not subject to public accountability and democratic oversight. Further, in AI alignment pipeline, major decisions including model specifications, constitution of system and safety filters are opaque and involve no public scrutiny. Thirdly, there is absence of robust enforceability. Without external auditing or standardized frameworks for evaluation, it is extremely difficult to say if these systems can adhere to such intended constraints. At times, in practice, the alignment claims are based on internal assessments and not independent scrutiny or validation.

Some AI researchers, recognizing these shortcomings have made efforts to expand their community participation by incorporating certain pluralistic values by collating judgement or preference data from diverse populations and sourcing ethical guidelines and safety principles from public discourses. Nevertheless, there is another field of practice and knowledge than can support in developing effective legitimate approaches to enable AI alignment, which is law.

Legal Alignment as Normative Anchor to AI Alignment

Designing AI systems that operate in consonance with appropriate legal principles, rules and methods can effectively address AI misalignment issues. Legal alignment is essentially a mechanism to harness law to tackle technical and normative aspects of AI alignment. Further, this emerging field has gained momentum because of its distinctive features, such as, institutionally accountable, methodologically structured and publicly articulated. This approach doesn’t just provide substantive rules but also engages in interpretative frameworks that can balance competing interests and resolve ambiguity efficaciously. Certainly, it makes it an effective (though evolving) guide to regulate AI behaviour where simple and plain heuristics are inadequate.

Further, legal alignment does not necessarily mean to assign legal personhood to an AI, but it is about making system design legally compliant, enabling the systems to recognize appropriate rules, legally germane facts, reasoned judgement in complex and uncertain situations. This alignment shifts the goalpost from behavioral optimization to normative conformity. Furthermore, it also distinguishes internal alignment—within the system, and external set of rules and regulations—of deployers and developers. This distinction is indispensable, as it leads to show that regulation will govern ecosystem in which an AI system will operate, and legal alignment will ensure that systems behave in a manner conforming with legal norms.

Furthermore, it is essential for legal alignment to deliver best results that it must be operationalized throughout the AI lifecycle and not be staggered and fragmented. The framework should identify and focus on three chief pillars: evaluation, design of the system and lastly, institutional support. Firstly, by evaluation, I mean that alignment mechanism should be measurable. This would involve development of such benchmarks that can assess identification of legal facts, compliance with rules and capacity to give principled legal analysis and reasoning. Importantly, red-teaming and adversarial testing will be of paramount importance particularly in cases with involve negligent or malicious users. Additionally, observational methods to study deployed systems can also reveal the propensity of AI systems to deliver inappropriate results in practical scenarios. Secondly, staunch designing of models where legal constraints are embedded in the architectural stage itself, would help in integrating legal principles into the training data, implementing output filters, incorporating legally efficient reasoning into the model prompts and in designing such tool-use protocols to prevent unlawful engagement or assistance. These interventions go beyond ostensible and superficial safeguards and hold the potential to address the lacuna in reasoning and informational capacities of models. Thirdly, there is a requirement to support legal alignment by governance mechanisms including model registries, safety cases, transparency requirements, continuous monitoring and certification schemes. Arguably, these structures will ensure strict demonstration and maintenance of legal alignment principles, rather than bogus claims.

Privacy Leaks: A Legal Wrong

One of the most insidious failures in the current alignment process is the recurring tendency of AI models to provide/leak private information. Essentially, it must be noted that this isn’t just about a bug, but a consequence of structuring of these models. LLM models are trained largely on internet corpora, therefore, they not just memories but also reproduce data such as names, numbers, or other sensitive identifiers, in their final outputs to users. Legal scholarship confirms that memorization is an intrinsic and fundamental feature of AI systems, and not an inadvertent aberration which can be eliminated. Arguably, misalignment in this scenario is not technical, perhaps normative. The system while performing its tasks does what it ought not do, majorly because the governing specification of such a model doesn’t have precise rules to protect personal data. In my opinion, this gap is egregious and there must be pathways to provide for legally binding rules for these systems. Certainly, the Art 5 of the UK GDPR on purpose limitation and data minimization doesn’t just showcase aspirational values but they are enforceable and binding principles. Further, Article 25 requires privacy by design as a default structural commitment and not a convenience. Preferably, legal alignment here can offer genuinely practical solution. By way of legal alignment mechanism AI system can be designed to behave as if it were a legal actor within data protection law, therefore, refrain itself from reproducing personal data. In practice, this would enable outputs to not be calibrated based on internal and hard-to-understand policies of developers but to comply with the standards as mandated by data protection law. Further, once an AI system is legally aligned it can “directly draw on legal resources to determine whether a user instruction or proposed action violates the law.”

Importantly, the argument on privacy leak brings to fore the deeper institutional vulnerability. It is imperative to note that data protection law, which deals with personal data, wasn’t formulated by team of engineers or developers to optimize user engagement. Rather, it came into its current form through democratic discourse—formed by Parliament, litigated and enforced to enable public hold rights over their data. One of the reasons that AI systems at times deliver harmful or biased outputs is because of lack of institutional grounding. In my view, legal alignment as a baseline mechanism will shift the focus from ad hoc developer’s intended safety policies to making systems shaped by legitimate, accountable and transparent legal standards to mitigate misuse and accidental leaks.

Evolving Legal Landscape and Limitations

When it is suggested that legal alignment can be a plausible solution, it is also important to acknowledge its limitations. Certainly, law can be agnostic to normative questions, or across jurisdictions work differently, and at times it might be difficult to apply it in novel settings. It is also possible that if legal compliance is made a target, then developers or deployers may come up with a workaround to optimize performance of the systems while violating law in ways that will be hard to detect. I believe that this is an essential caveat which must be factored in for smooth implementation of this approach. However, these limits should not undermine the fundamental argument. This instills the requirement of a multi-layered framework/approach which would include legal alignment as a bedrock or a baseline complemented by external oversight and other technical safeguards. The goal should be to have principled constraint.

Conclusion

AI misalignment exposes a prominent gap in existing governance approaches. Systems that are designed to achieve well-intentioned objectives can also behave quite inconsistently or unpredictably to have harmful effects in real world. This should not be viewed as a mere technical shortcoming as it is in effectively anchoring AI to behave in robust normative framework. Preferably, legal alignment can be a way forward. By incorporating legal rules and principles into the AI systems, they can become more predictable, enforceable and testable. While it might not be a complete solution, it can work to be a critical baseline which would ensure that AI systems operate in conformity with publicly reasoned norms. As the capabilities of AI continue to expand, the question should no longer be confined only on tapping misalignment occurrences but also work on finding mechanisms to constraint it at the grassroot level. The answer, progressively, might not only lie in rejigging private AI safety policies but also aligning AI with the standards of law.

Celebrating 10 Years of EAGx by Niki Kesseler

Niki Kesseler — Thu, 30 Apr 2026 10:35:29 +0000

Today marks the 10-year anniversary of EAGx. On April 30, 2016, the very first EAGx events—EAGxBoston and EAGxBerkeley—were held simultaneously. A decade later, it’s incredible to reflect on how much this community has grown, both in scale and in depth.

Since then, we’ve hosted 67 EAGx events and 18 EA Summits (a format we introduced in 2024), bringing us to 85 events in total—with our 86th happening this weekend (EAGxDC!). Across these, we’ve welcomed over 24,500 attendees in 20 countries and across six continents.

A few milestones that stand out:

EAGxAustralasia is our longest-running series (6 editions, rotating across cities)
EAGxBerlin has been our most frequently hosted city (5 times)
Our largest event, EAGxBoston 2022, brought together 935 attendees
The portfolio continues to grow: this year alone, we’re planning 10 EAGx events and at least 22 EA Summits

EAGxAustralasia 2019

EAGxBerlin 2025

EAGxNordics 2026

The EA Summit format, in particular, has scaled quickly—from 409 attendees across 3 pilot events in 2024 to over 1,700 across 2025–26 already. It’s been exciting to see how this complements the EAGx model and opens up new ways for people to engage.

But more than the numbers, EAGx has always been about people and pivotal moments:

Attending a first event and rethinking a career path
Stepping into an organizing role and deepening one’s engagement
Finding collaborators, co-founders, and long-term friends
Building local and regional communities that continue long after the event ends

EA Summit: Hong Kong 2026

Many of the projects, career transitions, and collaborations that shape the EA ecosystem today have roots in conversations that started at EAGx events. That’s something we’re especially proud of.

However, it’s hard to pick a couple of important milestones, because every single event has contributed to the growth of talent, communities, and effective altruism as a whole and we’re grateful to all of our dedicated organizers who make working on EAGx a worthwhile and impactful contribution to EA.

Therefore, we also want to give a huge shoutout and MASSIVE thank you to everyone who has ever organized, attended, spoken at, or supported an EAGx. Organizing these events takes an enormous amount of care, thought, and behind-the-scenes effort—and the impact is only possible because of the people who show up and contribute.

On a personal note, I feel very grateful to now be working with an incredibly dedicated EAGx team as we continue to grow and refine this portfolio. We’re excited not just about scaling, but about improving quality, strengthening community health, and ensuring these spaces remain thoughtful, welcoming, and impactful.

From left to right: Kiryl Shantyka (EAGx Coordinator: Europe), Niki Keßeler (EAGx Program Lead), Dion Tan (EAGx Coordinator: Asia, Australasia, and Africa), and Arthur Malone (Senior EAGx Coordinator: Americas)

We’re looking forward to many more years of EAGx—and we hope to see many of you at our upcoming events. We’re also going to be posting some program updates on EAGx and EA Summits in the coming weeks, so stay tuned!

See you, besties!

Niki & the EAGx team

Technical AI Safety Research Plausibly Increases P(doom) Absent Vastly More Advocacy by Contrarian^2 = I 🔸

Contrarian^2 = I 🔸 — Thu, 30 Apr 2026 07:39:39 +0000

TL;DR: Safety research without regulation is optional. Cautious labs adopt it and slow down, while reckless labs don’t and accelerate. The median lead lab is more reckless in the scenario with strong technical AI safety research and weak governance than in any other scenario.

This post seems longer than it is. If the 4 premises below seem plausible to you, you can read just the short scenarios and get the gist

This post rests on the following premises:

Premise 1: Labs with greater AI capabilities will develop their capabilities more quickly than those with lesser capabilities.

Premise 2: Cautious labs that adhere to higher safety standards will develop capabilities more slowly than reckless labs that adhere to lower safety standards.

Premise 3: Reckless labs are more likely to kill us all than cautious ones.

Premise 4: Absent a strong advocacy and policy branch of the AI safety movement, AI safety research will lead to uneven adoption of additional safety standards. Cautious labs will adopt more external standards than reckless labs.

If we accept all these, we quickly get a plausible narrative for how AI safety technical research backfires. It goes something like this:

At the moment, we have a handful of frontier labs, each with their own internal attitudes towards safety. Those who value safety more put models through a longer series of tests and evaluations they have developed before releasing them and, importantly, before using them to make algorithmic decisions. Those who are more lax release earlier, with fewer safety checks, and eagerly adopt their own coding tools.

These differences in attitudes have resulted in significant safety differences. They have also presented financial risks to more cautious labs — take, for example, Anthropic’s recent designation as a supply chain risk, which prevents federal contractors from using their products. Although capabilities are similar across the top labs despite their safety differences, we should not expect this state of affairs to continue without strong evidence.

Bad Scenario: Strong AI safety research, weak policy arm (status quo)

Suppose technical AI safety researchers continue to demonstrate that additional safeguards are necessary. Without any governmental enforcement mechanism, only cautious labs agree to institute them, slowing down iteration. Reckless labs don’t, and continue advancing at breakneck speeds. This magnifies the existing safety asymmetries, further disadvantaging cautious labs.

As differences in capabilities emerge, two forces converge to magnify differences: investment and recursive self-improvement. Before long, reckless labs are vastly outpacing cautious ones, making them virtually obsolete.

Mediocre Scenario: Weak AI safety research, weak policy arm

Without a strong push for additional safety standards, cautious labs continue with moderately more safety checking than their reckless counterparts. The additional, internal safeguards they employ are more important on the compute margin and reduce risk without significantly slowing them down. Although they remain disadvantaged in the capabilities race, the difference is small enough that random variation and individual breakthroughs might make it up, allowing cautious labs to fight for the lead.

Good Scenario: Strong AI safety research, strong policy arm

Safety researchers and policymakers collaborate to roll out and enforce safety protocols, raising standards across the board. Labs are slowed down as they fail checks, giving safety researchers more time to solve unsolved problems. Those who try to circumvent safety checks are publicly fined, losing them market share and forcing them out of the race.

Suggested Intervention

Spend more on lobbying, advocacy, and policy work. Ideally, do this without cutting research budgets, but if necessary, redirect funding. Individually, consider switching to policy work and donating to AI governance organizations. Encourage your friends to do the same. Plenty of smarter people than me have made this case, so I won’t go into much detail here.

Extra stuff

Defense of Premises

Premise 1 seems intuitive to me. More capable models draw more investment, (probably) have better research taste, and make better algorithmic decisions. When tasked with improving themselves, more capable models will perform better.

Premise 2 is perhaps the weakest, but I still am inclined to believe it. Safety requires compute and, when safety checks are failed, difficult, time-consuming setbacks to trace mistakes and retrain models to the same capabilities. Investors may not be convinced that safety spending is necessary, so they may prioritize investing in more reckless labs that spend more of their investments on capabilities-advancing compute which, in their view, provides better returns. Cautious labs may refuse highly profitable contracts on safety grounds, which reduces revenue and investment (see Anthropic DOD fight).

Premise 3 seems intuitive. Reckless labs release more capable models more readily, with fewer jailbreak protections and more potential for misuse, and readily use them internally without sufficient proof of alignment, with potentially catastrophic alignment implications.

Premise 4 is probably true, but the degree to which it is true is significant for my argument. If my defense of Premise 2 is true, then financial incentives may lead reckless labs to reject even convincing safety research. Cautious labs share these same incentives, but are insulated from them by the values and tendencies of their leaders.

Caveats

There are a number of reasons that I could be wrong, and I would love to hear them in the comments. A couple that come to mind:

It is plausible that AI safety research is adopted at least a little across the board, such that even the worst labs are better than the median lab would be absent external AI safety research. I think the financial incentives explained in my defense of Premise 2 make this less likely.
It is plausible that investors are drawn to safe labs, since significant, public safety failures could cause other investors to go elsewhere. If this were true, though, we would expect more non-philanthropic AI safety funding, and investors would be pushing for stronger internal and governmental AI safety protocols. They do not seem serious about any of those things.
It is plausible that safe labs develop more quickly, since sufficiently unsafe labs may eventually run into catastrophic safety failures that force them to rewind more than their safe counterparts. However, advanced misaligned AI may not reveal itself until it is too late in order to avoid these sorts of setbacks.
It is plausible that the safety standards adopted by the most cautious labs do not significantly slow them down, even absent failures in unsafe labs. This may be the case if safety research makes safety testing more compute-friendly. If, however, investors choose to avoid safe labs anyway, for fear that their money is not being well-used, this caveat would not matter.
It is plausible that businesses will purchase from safer labs to avoid risks like data leakage, providing cautious labs with additional revenue to scale compute, and drawing investors to cautious labs. However, business-relevant safety concerns may not be relevant to existential and suffering risks.

Red vs blue: The parable of the feud within a feud by Joe Rogero

Joe Rogero — Thu, 30 Apr 2026 04:04:11 +0000

You are a wise member of the proud and noble Clan Auran. You bear no love for your ancient enemies in Clan Irgentum, but nonetheless you hope to end the feud, because in your wisdom you know that violence begets only more violence.

But you also know that among your family are rash young idealists who seek to purge your ancient foe forever, and bitter old souls who recall the injuries suffered at that foe’s hands like it was yesterday. They are not among the majority, but some would be hard to dissuade.

The annual Day of Vengeance approaches. Your cousin comes to you on the eve before, and says: “I think Clan Auran will march to war.”

“Some may,” you admit. “But it would be foolish in the extreme to set forth alone.”

“I know this,” says your cousin seriously, “and you know this, but there are those among our family who do not know this. They will march regardless, and if they march alone, they will surely perish. If the clan marches with them, they will survive. Will you condemn the bravest and boldest of our clan to death?”

“They might not march,” you object, “if they see clearly enough that they march alone.”

“They will not see so clearly,” says your cousin sadly. “Some among us will surely die.”

“I am afraid so,” you admit. “But to bring more of the clan to the fight will only cause more needless deaths.”

“Not so,” objects your cousin. “If enough of the clan march to war, our foes will see our numbers and flee, and the restless and bitter among us will be satisfied, and there need be no deaths among our beloved kin.”

“To accomplish this would take more than half our entire clan!” you exclaim. “Surely there is more wisdom among us than that.”

“Ah, but you forget the love we hold for one another,” your cousin observes. “The bonds of family are a powerful thing. Where the youngest may march to war against the warnings of his brethren and elders, still his brethren do not wish to lose him. Some will march to war for his sake, and still others will march for theirs, and so on. We would not need the whole clan to march, only a majority, for all we love to be saved. It may take a heroic effort, but no one needs to die tomorrow.”

You feel yourself becoming frustrated. “And if that same effort were dedicated instead to convincing the restless and the bitter to remain behind, among their loved ones, and not march to their deaths in a pointless feud, no one would need to die tomorrow either!”

“The hour is late,” your cousin reminds you, “and most will have made their decision already. In accordance with our traditions, those who plan to march to war on the morrow must paint their spears blue in honor of the river that divides our lands from those of our hated rival; and those who intend to remain must paint their spears red in honor of the sun that sets behind our sacred hills. Even if you were to violate these traditions, you would not be able to convince more than the barest fraction of our clan to change their minds before the march begins.”

“But that could result in the worst possible outcome!” you lament. “If many of our clan march to war, but their numbers are not enough to frighten off the foe, they will all be killed by our enemies, who have never been known to show mercy! If it is foolish to march alone, it is even more foolish to _decide whether to march_ without knowing how many of our kin will join us!”

“And yet that is the decision we now face,” says your cousin sadly, “and no amount of wishing will change it.”

“Can I at least convince you to remain, and spare your own life in the event of tragedy?” you plead.

“Alas, you cannot,” responds your cousin, “for my sister lost her husband and child in the last raid. She will surely be among those who march to war, and I love her more than life itself. Good night, my cousin, and if we are not blessed to meet again before the Day of Vengeance ends, farewell.”

You watch your cousin leave, straight-backed and proud, and wonder how it got to this point. You return to your home; you descend into the cellar, ancient stairs carved by your great-grandfather creaking under your weight. You clear the dust from an old table, and on it you place three items taken reverently from where they have been stored: a single stout spear, battle-worn but reliable, and two weathered boxes of dry paints, one red, one blue.

Tim Urban of Wait But Why recently revived a scissor statement, an especially divisive question, that did the rounds on the internet a while back:

Lots of people got pretty mad about this.

The first step when facing a hard choice is to acknowledge it as such. Likewise, the first step when facing a sincere and meaningful disagreement among friends is to acknowledge it as a sincere and meaningful disagreement.

The second is to seek the core elements that make it hard to resolve.

Contrary to what some on the internet may suggest, I do not think that the fundamental problem here is that blue-button-choosers are stupid naive twits who deserve to die and good riddance. Nor do I think the problem is that red-button-choosers are selfish assholes who would let others burn to save themselves.

In a world of perfect reasoners with mutual knowledge of each other’s goals, everyone coordinates on red or blue by some feat of logic to which there is an objectively correct answer, for the same reason that the movement of perfect spheres in a frictionless vacuum has an objectively correct description. But the real world is considerably messier. The story above is my attempt to take the frictionless-spheres-in-a-vacuum thought experiment and add back a _little_ bit of grounding in situations humans might actually face, and in so doing to illustrate some of what makes the question so divisive.

We could imagine a world of blue-button-pushers, all looking to make the world safe for everyone, with a very comfortable margin to ensure that no one dies. We could imagine a world of red-button-pushers, who correctly expect that pushing blue gets someone killed at no benefit to the group, and so virtually nobody does. We could even imagine a world split almost 50-50, in which a single person’s choice stands a meaningful chance of making the difference between four billion deaths and zero.

The problem is that we don’t know which world we’re in just by looking at the question.

If we take the question seriously, and we choose to care at all about the fate of our fellow humans _even for selfish reasons_, we have not yet resolved the conundrum. We must still decide based on guesses about other people’s decisions, and guesses about their guesses, and so on. Whatever your intuitions may be, I beg you to acknowledge at least that _this is not necessarily an easy decision to make._

Nor should it be, really. The world of blues is a world of trust and charity, filled with people who would gladly accept a small risk to themselves to make sure that no one slips through the cracks. The world of reds is a world of agency and self-sufficiency, filled with people justly confident that their society neither needs nor demands their sacrifice, because everyone can and does look out for themselves.

These visions are beautiful to me, each in their own way. Yet to live in one of them forever, I think, would be to lose something precious. For that reason alone, we ought to extend a modicum of kindness to those whose intuitions clash with our own.

The world in the middle is messy. And like it or not, so is the world in which we currently live.

It’s okay to feel conflicted about that, sometimes.

Welfare Biology and AI: Soil and Sea by Dawn Drescher

Dawn Drescher — Wed, 29 Apr 2026 20:49:13 +0000

There are 57 billion nematodes per human. Boreal forests pack 7× more per square meter than cropland. The numbers, the mechanisms, and why pesticides might make things worse.

This is part 2 of a five-part sequence on welfare ecology. Part 1 introduces the ethical premises. Part 3 covers interventions. Part 4 explores a model of invertebrate suffering. Part 5 covers AI.

If you decided in part 1 that you care – even a little – about the welfare of invertebrates, the next question is: Where are they, how many are there, and what drives their population sizes?

The Numbers

Terrestrial Soil Fauna

The foundational data comes from two landmark studies:

Van den Hoogen et al. (2019) estimated 4.4 × 10²⁰ nematodes in the top 15 cm of soil globally. Stefan Geisen, the second author, clarified that this accounts for roughly 90% of all soil nematodes, putting the total at about 4.89 × 10²⁰ (Grilo 2025a).
Rosenberg et al. (2023) estimated 10¹⁹ soil arthropods, of which roughly 95% are mites and springtails, with the remainder being ants, termites, and other groups.

For context: There are 8 billion humans and roughly 80 billion farmed land animals at any given time. Soil nematodes alone outnumber all farmed animals by a factor of 6 billion.

Marine Fauna

From Bar-On, Phillips, and Milo (2018):

Approximately 10²¹ marine nematodes – about 2× the number of soil nematodes.
Approximately 10²⁰ marine arthropods (copepods, krill, amphipods, etc.) – 10× the number of soil arthropods.

Marine nematodes are overwhelmingly benthic (living in seafloor sediment). Marine arthropods are split between benthic and pelagic (water column) habitats.

Density by Biome: The Terrestrial Landscape

Not all land is created equal. Rosenberg et al. (2023) provided mean densities of soil arthropods by biome. Grilo (2025a) combined these with van den Hoogen et al.’s nematode data to estimate total soil animal density. The results:

Several things jump out:

Forests are packed. Boreal and temperate forests have 7–8× the soil animal density of cropland. The mechanism is well understood: forests produce continuous, diverse leaf litter that feeds a complex soil food web. They have deep root networks supporting fungal communities that mites and springtails feed on. The soil is structured with pores and channels that provide habitat at every scale.

Cropland is relatively sparse. Tillage destroys soil structure. Monocultures reduce food web diversity. Pesticides directly kill a broad spectrum of fauna. The litter layer is minimal.

Pasture is sparser still. Grazing pressure, trampling, and simplified plant communities reduce below-ground complexity even further.

Deserts are comparatively empty. At 0.05 million per m², deserts have ~130× fewer soil animals than forests. Still even 50,000 animals per m² is impressive.

Volumetric Comparison: Soil vs. Sea

Since soil nematodes are measured in the top 15 cm (~0.15 m³ per m² of surface), we can convert to volumetric density and compare to marine environments.

One wrinkle: The dominant taxa change with the habitat. In soil, nematodes and arthropods (mites, springtails) coexist. In marine sediments, nematodes dominate overwhelmingly – they make up 80–90% of meiofauna (benthic organisms of 32–63 μm to 0.5–1 mm), with only small numbers of harpacticoid copepods alongside them. In the open water column, the situation reverses: There are essentially no pelagic nematodes (they are benthic animals, bound to sediment), and the invertebrate fauna is almost entirely arthropods – copepods, krill, and amphipods.

The table below uses “nematodes + arthropods” for all environments, but the composition varies as described:

The key insight. Per cubic meter, terrestrial soil and some coastal marine sediments are by far the densest habitats per cubic meter for potentially suffering animals, though coastal sediment occupies a much thinner layer. The open ocean water column is orders of magnitude sparser – even productive surface waters have roughly 10,000× fewer organisms per m³ than forest soil. But the ocean’s total volume is vast (~1.37 × 10¹⁸ m³), so low per-m³ densities still add up to enormous total populations.

The deep ocean – both water column and sediment – is strikingly empty by terrestrial standards. It represents an enormous volume of near-zero-fauna space that already exists naturally.

What Drives Population Size? The NPP Primer

Understanding why some biomes have more soil fauna than others requires understanding net primary productivity (NPP).

NPP is the rate at which plants convert solar energy into organic matter, net of their own respiration. It’s measured in grams of carbon (gC) per m² per year. It’s an objective, species-independent measure of how much new biological material is being produced. A tropical rainforest has an NPP of ~1,000–2,000 gC/m²/yr. A desert has ~0–90 gC/m²/yr. Open ocean averages ~125 gC/m²/yr.

NPP matters because it determines the total energy flowing into the food web. More NPP means more food for decomposers, which means more food for the nematodes and mites that eat the decomposers, and so on up the trophic levels.

Vasco Grilo’s analysis found that NPP-related variables (biome type, soil organic matter) are far more important than any other factor in predicting soil fauna density. The m² -years of agricultural land per food-kg explained essentially 100% of the variance in welfare effects for his preferred parameters (Grilo 2025c).

A common misconception. NPP is not subjective to a species. When a cow eats grass, the NPP doesn’t change – the grass still produced the same amount of carbon. What changes is the pathway: instead of grass dying and feeding soil decomposers, the cow intercepts the energy, metabolizes 40–70% of it, and excretes the rest as dung. Dung is more labile (easier for bacteria to digest) than leaf litter, so it can boost local populations of bacterivorous nematodes around dung pats. But the cow has diverted most of the energy away from the soil food web, so the net effect on soil fauna is still strongly negative.

The same logic applies to any intervention that changes the pathway of NPP without changing NPP itself. This distinction turns out to be critical for evaluating pesticides.

An important caveat: Global NPP has barely changed. As I noted in part 1, Tomasik’s analysis of Krausmann et al. (2013) found that NPPeco – the global NPP available to wildlife – has barely changed over the 20th century, partly because CO₂ fertilization has increased potential NPP even as humans have appropriated more of it. This means that the biome-level density differences in the table above are real and actionable for local interventions, but the global total of animal metabolism may not have declined as much as defaunation indices suggest. Converting one hectare from forest to cropland genuinely reduces soil fauna on that hectare by ~7×; but globally, CO₂-driven NPP increases elsewhere may be partially offsetting these local reductions. This doesn’t undermine the case for targeted land use interventions – it just means we should be cautious about claims like “humanity has reduced total invertebrate suffering by X%.”

The Pesticide Trap

One might naively think that pesticides reduce invertebrate suffering by reducing their populations. But the relationship between pesticides and welfare is more complex – and possibly perverse.

Consider two interventions that both halve the standing population of soil fauna:

Intervention A: Reduce NPP by 50%. (For example, convert forest to desert, or shade the land.) The carrying capacity halves. At the new equilibrium, fewer organisms are born, fewer die, and the birth rate and death rate are both lower. Total deaths per year: lower.

Intervention B: Apply pesticides while keeping NPP the same. Food is still available. Organisms are born at rates close to the original (abundant resources encourage reproduction). But pesticides kill a large fraction of them. The population is suppressed below carrying capacity, but the high food availability keeps birth rates high. At the new equilibrium: birth rate is elevated (more food per capita), death rate is elevated (pesticides), and organisms are dying of pesticide poisoning instead of starvation or predation. Total deaths per year: much higher than the original, with deaths potentially involving more acute suffering.

This matters enormously for r-strategists. Brian Tomasik and Yew-Kwang Ng (1995) argued that for r-strategists, death is a large fraction of total lifetime suffering. If that’s right, the death rate may matter more than the standing population. And pesticides at constant NPP can increase the death rate – creating a high-throughput killing field rather than a genuinely smaller population.

Pesticide Persistence

The picture is further complicated by the fact that different pesticides have very different persistence profiles:

Organophosphates and carbamates degrade in days to weeks. These are closest to the “spray, kill, wash away, repeat” model – high-throughput killing.
Neonicotinoids persist in soil for months to years and are taken up systemically by plants. Any insect feeding on a treated plant gets a dose. This effectively makes NPP inaccessible to insects – the food is there but poisonous. This is closer to “reduced accessible NPP” and may genuinely reduce the carrying capacity.
Legacy organochlorines (DDT, dieldrin) persist for decades but are mostly banned.

The neonicotinoid case is the most interesting from a welfare perspective: by making plant material toxic, neonicotinoids functionally reduce the accessible NPP, which is closer to the “clean” intervention of actually reducing NPP. However, sublethal neonicotinoid exposure causes disorientation, impaired foraging, and reduced reproduction in insects – arguably adding suffering without the “benefit” of reducing the population.

Bottom line. Reducing actual NPP (through land use change) is a much cleaner intervention than applying pesticides at constant NPP. The famous ~75% decline in flying insect biomass in Germany (the Krefeld study) is attributed partly to pesticides and partly to habitat loss. From a welfare perspective, the habitat loss component (reduced accessible NPP) is straightforwardly population-reducing, while the pesticide component may be creating a high-throughput killing field.

A Better Metric: Life Expectancy at Birth

Let me flesh out the life expectancy at birth (LEB) metric for invertebrates, which I think deserves more attention. It’s a metric that we care about intuitively when thinking about humans, so it’s one that we have reason to also care about when it comes to other species.

LEB = total animal-years lived / total births per year

For an r-strategist marine fish laying 1 million eggs, of which 2 survive to reproduce, and average lifespan across all offspring is ~3 days: LEB ≈ 3 days.

For a K-strategist elephant with 6 offspring over a lifetime, of which 2 survive to reproduce, average lifespan ~20 years: LEB ≈ 20 years.

Why LEB is attractive.

It naturally penalizes the r-strategy horror: Systems where millions of beings are born just to die almost immediately get very low scores.
It’s legible. Most people intuitively understand that higher life expectancy is better.
It sidesteps the sign problem somewhat. You don’t need to know whether lives are net positive or negative to argue that ecosystems with higher LEB have less death per unit of ongoing life.
It aligns with antifrustrationism. Organisms born into very short lives full of pain score low, regardless of whether we think the brief pleasure before death was “worth it.”

Potential objections. A strict utilitarian would say this smuggles in the assumption that death is the primary locus of suffering. LEB doesn’t account for intensity of suffering during life. And it could be gamed by creating very long-lived organisms in terrible conditions. But as a rough heuristic for comparing wild ecosystems, it captures something real: Ecosystems dominated by K-strategists score higher than those dominated by r-strategists.

A refinement. One could weight by “fraction of lifespan spent with a developed nervous system” to exclude possibly non-sentient embryonic stages. For insects that undergo complete metamorphosis, eggs are probably not sentient, but larvae have functional nervous systems from early instars. For nematodes, the L1 larva hatches with 222 of its final 302 neurons and is immediately mobile and responsive – so the cutoff would be at hatching. The European Food Safety Authority (EFSA) suggests the onset of potential sentience at roughly “the beginning of the last third of development within the egg or mother” for most vertebrates, and “when it is capable of feeding independently” for fish, amphibians, and invertebrates (EFSA 2005).

The Ocean: Acidification as Accidental Intervention

One of the largest ongoing shifts in marine invertebrate populations is driven by ocean acidification – an accidental consequence of CO₂ emissions.

The mechanism. CO₂ dissolves in seawater and forms carbonic acid, lowering pH. This reduces the availability of carbonate ions that many marine arthropods need to build shells and exoskeletons. Calcifying species – copepods, pteropods, krill, crabs – are directly harmed. The effect is already measurable: ocean pH has dropped by about 0.1 units since the Industrial Revolution and is projected to drop another 0.3–0.4 units by 2100 under high-emission scenarios.

Does acidification reduce total populations, or just reshuffle species? At the current margin, acidification appears to be reducing total marine arthropod populations rather than merely substituting non-calcifying species for calcifying ones. The reasons:

Timescale mismatch. Current acidification is happening over decades; evolutionary adaptation takes millennia to millions of years. Non-calcifying species haven’t had time to evolve into vacated niches.
Structural role of calcifiers. Copepods and pteropods occupy dominant roles in marine food webs. Their decline disrupts the entire ecosystem, not just their specific niche.
Paleoceanographic evidence. After the Paleocene-Eocene Thermal Maximum (PETM, ~56 million years ago), which involved rapid ocean acidification, it took ~100,000–200,000 years for marine communities to reorganize. Recovery was slow and involved reduced overall productivity.

This makes ocean acidification one of the few large-scale ongoing processes that is plausibly reducing total marine invertebrate populations.

Dead Zones: More Nutrients, Less Life

Another counterintuitive marine phenomenon: nutrient runoff from agriculture creates “dead zones” that dramatically reduce animal life.

The mechanism. Nitrogen and phosphorus from fertilizer enter coastal waters → algal bloom at the surface → algae die and sink → bacterial decomposition of sinking algae consumes enormous amounts of dissolved oxygen → oxygen drops below ~2 mg/L (well-oxygenated water has ~8 mg/L) → most animals suffocate.

The Gulf of Mexico dead zone, fed by Mississippi River agricultural runoff, covers ~15,000 km² of seafloor with severely depleted macrofauna every summer.

Dead zones vs. pesticides. An established dead zone is more like “reduced NPP” than like “pesticides at constant NPP”: Within the anoxic zone, the carrying capacity for macrofauna is near zero, so there’s no high-throughput killing. The suffering is concentrated in the creation and seasonal re-establishment of the dead zone. Permanent dead zones (rare but possible) would be functionally equivalent to removing the habitat – genuinely low birth rate and low death rate, not a killing field.

However, many dead zones are seasonal – forming in summer and dissipating in winter. This seasonal cycling creates recurring mass mortality events, which is the oscillating-biome problem: not a stable low-population state, but a perpetual cycle of colonization and die-off.

Grilo’s Key Findings

Vasco Grilo’s analyses are dense with inline math, which can make the key insights hard to extract. Here’s a plain-language summary of what I consider his most important findings:

1. Effects on soil animals might dwarf effects on intended beneficiaries. Grilo’s earlier models indicated that the impact on soil nematodes, mites, and springtails is orders of magnitude larger than the impact on the people or farmed animals the intervention is designed to help. He has since updated these models, and commented on my previous article that his new credible intervals span many orders of magnitude in both directions, extreme uncertainty that is driven by uncertainty over nematodes’ welfare ranges.

For [some reasonable parameters], I estimate that GiveWell’s top charities change the welfare of soil ants, termites, springtails, mites, and nematodes [10⁻⁵ to 10¹⁰] times as much as they increase the welfare of humans. For my preferred [parameters], the change in the welfare on those soil invertebrates is 41.5 times … the increase in the welfare of humans ….

Note that if the GiveWell numbers are based on increases in human population, then they are questionable due to both Krausmann et al. (see above) and Roodman (2014).

2. Agricultural land has fewer soil animals. Crops have ~1.3 million soil animals per m², compared to ~3–9 million for natural biomes. Pasture has even fewer (~0.7 million). So anything that increases agricultural land – including saving human lives, which increases food demand and therefore cropland – reduces, at first approximation, soil animal populations.

Grilo clarifies in a comment that this is only true of arthropods:

I think agricultural land has less soil arthropods, but I have little idea about whether it has more or less soil nematodes. The vast majority of soil animals are nematodes. So I also have little idea about whether agricultural land has more or less soil animals.

3. Nematodes dominate. Of the total welfare effect on soil animals, 90–94% comes from changes in nematode populations, depending on the biome. This is because nematodes outnumber arthropods ~50:1. It depends mostly on the welfare ranges whether this numerical advantage overwhelms everything else.

4. The sign is uncertain. Grilo estimates nematodes have negative lives with probability ~59%, mites ~56%, springtails ~55%. If their lives are net negative, reducing their populations (via increasing agricultural land) is beneficial. If net positive, it’s harmful. Every concrete recommendation flips depending on this sign. Grilo’s own best guess is net negative, but he explicitly acknowledges the uncertainty.

5. Land use is the key variable. The m²-years of agricultural land per food-kg explain essentially all of the variance in welfare effects across different foods and interventions. Beef requires ~326 m²-year/food-kg; chicken requires ~12; peas require ~7.5; soy milk requires ~0.7. From the soil-animal perspective, producing 1 kg of beef reduces soil-animal-years by ~1.39 billion – 164 billion times as much as it increases the living time of the cows.

These findings are robust to a wide range of assumptions about welfare ranges, as Grilo demonstrates with sensitivity analyses across different exponents for the neuron-count-to-welfare-range relationship (Grilo 2025c).

Summary

The empirical landscape boils down to a few key facts:

Soil fauna are staggeringly abundant – 10²⁰ nematodes, 10¹⁹ arthropods – and their total expected welfare dominates that of all other animal groups combined if they are sentient.
Population density varies enormously by biome, from ~9 million per m² in boreal forests to ~50,000 per m² in deserts. Land use change is a powerful lever. But note that the oft-cited difference between croplands and forests is only around 1–8×.
Reducing NPP is a cleaner intervention than pesticides for lowering populations. Pesticides at constant NPP risk creating a high-throughput killing field rather than a genuinely smaller population.
Life expectancy at birth may be a more informative metric than total animal-years, especially for r-strategists where death is the dominant source of suffering.
Ocean acidification and dead zones are reducing marine invertebrate populations, but the mechanisms and welfare implications are complex.
Land use explains almost everything in Grilo’s analysis. If you want to affect soil fauna populations – in either direction – the most powerful lever is changing how land is used.

In part 3, I’ll turn to the practical question: Given all of this, what should we actually do?

The Infrastructure Gap: Why Smarter Funding Means Stronger Impact (Part 2 of 3) by Deena Englander

Deena Englander — Wed, 29 Apr 2026 19:51:45 +0000

PART 2

The Infrastructure Gap: Why Both Funding Models Leave Impact on the Table

Picking up where we left off

In Part 1, we made the case that operational infrastructure is the single biggest variable determining whether nonprofit funding produces real impact – and that closing the infrastructure gap is a shared responsibility between funders and nonprofits. Now it’s time to look at how the two dominant funding models in the sector are actually working in practice, and where each one falls short.

There are two main approaches to institutional funding, and most funders lean heavily toward one or the other. Both have real strengths. Both also leave impact on the table – just in different ways. The goal of this section isn’t to declare a winner. It’s to understand what each model gets right, what each model gets wrong, and what we can learn from a sector that has spent decades figuring out how to make capital and operational support work together: venture capital.

IV. Two Models, Same Missed Returns

There are two dominant approaches to institutional funding, and most funders lean heavily toward one or the other: restricted grantmaking and trust-based philanthropy. Each one has real strengths. Each one also leaves impact on the table – just in different ways.

The important thing to understand is that the problems we’re about to describe aren’t unique to one model. Whether restricted or trust-based, if the funding relationship doesn’t invest in the organization itself – not just its programs – it’s limiting the return on impact for everyone involved.

The goal isn’t to pick a side. It’s to take the best of both and build something stronger.

Restricted Grantmaking

At its best, restricted grantmaking provides structure, accountability, and clear expectations. Metrics give both funders and nonprofits a shared language for evaluating progress. Reporting requirements, when done well, can actually help newer organizations develop discipline around tracking outcomes and managing finances. For early-stage nonprofits especially, some structure from funders can be a valuable learning tool – as long as funders are actually reading those reports and providing feedback, not just filing them.

Where it leaks returns:

• The compliance burden consumes mission time. A single small nonprofit may juggle 20-30 applications and reporting requirements from 10-20 funders, each with its own format, timeline, and metrics. Research estimates that most nonprofits spend 40 or more hours per quarter assembling grant reports (Sopact). That’s staff time being spent on paperwork instead of programs.

• Infrastructure is chronically underfunded. The normal overhead rate for delivering services effectively is 25-30% (National Council of Nonprofits), while actual indirect costs often run as high as 40% (Bridgespan). Most funders cap overhead at 10-15%, and 70% of nonprofits report they can only charge 10% or less (NFF 2025). That means most organizations are absorbing a 15-20 percentage point gap – or simply going without the infrastructure they need.

• The starvation cycle perpetuates itself. Funders hold unrealistic expectations about overhead → nonprofits feel pressure to conform by underreporting true costs → organizations underspend on infrastructure → results suffer → funders blame the organization → the cycle repeats (SSIR, “The Nonprofit Starvation Cycle”). This has been documented for over 15 years and it’s still the sector’s most damaging structural pattern.

• Short timelines prevent strategic investment. You can’t hire quality employees on a six-month runway. You can’t implement new technology systems knowing your funding expires in a year. Short grant cycles force organizations into survival mode, where every decision is about getting through the next quarter rather than building something that lasts.

• Contractors replace employees out of necessity, not strategy. When budgets are tight or runways are short, organizations default to contractors because they can’t justify a full-time hire. But contractors are inherently less invested in the organization – and the organization is less invested in them. The result is constant turnover, lost institutional knowledge, and a cycle of onboarding and retraining that consumes the very time and money the organization was trying to save.

• Reimbursement-based funding creates cashflow crises. When grants are structured as reimbursements rather than advance funding, the organization needs cash on hand to start the work before getting paid for it. For small nonprofits already operating with razor-thin reserves, this creates an impossible bind – they need money to spend money, and they don’t have it.

• Reporting without feedback is a one-way street. When funders require detailed reports but never respond to them, reporting becomes compliance theater. The organization spends significant time producing something that nobody reads, learns from, or acts on. The time would be better spent on the mission.

• Trust is never built. Even organizations with years of strong performance find themselves re-proving their worth annually. At some point, a track record should earn an organization graduated trust – lighter reporting, longer timelines, more flexibility. When it doesn’t, the message is clear: no amount of performance will change this relationship.

What this looks like in practice:

• An organization gets funding for a program but not for the software to run it efficiently – the result is more internal chaos, not more impact

• An organization is asked to report with detailed metrics but isn’t given funds for the tools or staff to track them – so reporting becomes guesswork

• An organization can’t invest in infrastructure without a solid runway – leadership is too busy fighting to be “fed” to focus on building something sustainable

• An organization gets access to funding but can’t hire the right people because the runway is only six months – so they either hire no one or hire someone they’ll lose

• An organization uses contractors for key functions because the funding timeline is too short to hire staff – every six months they’re retraining someone new, losing continuity, and spending more on the churn than a salaried employee would have cost

Trust-Based Philanthropy

At its best, trust-based philanthropy is a powerful corrective to the problems above. It gives organizations the flexibility to allocate resources where they’re most needed, reduces administrative burden, respects the expertise of nonprofit leaders, and creates space for innovation. The data supports this: 93% of recipients of large unrestricted grants reported that the funding strengthened their ability to achieve their mission, and 80% encountered no challenges managing the funds (Center for Effective Philanthropy, three-year study).

But – and this is critical – those results come from organizations that were extensively vetted before receiving funds and that largely already had the operational infrastructure to deploy large grants strategically. For organizations without that foundation, the picture is more complicated.

Where it leaks returns:

• It assumes operational competence that often doesn’t exist. This isn’t a criticism of nonprofit leaders – it’s a recognition that most founders started their organizations because they’re passionate about a cause, not because they have training in financial management, HR, compliance, or organizational design. Nobody taught them what operational excellence looks like for a nonprofit, and the funding model doesn’t require them to learn.

• There’s no early warning system. Without any accountability structure, problems that could be caught and corrected early – a budget shortfall, a compliance gap, a staffing misalignment – go unnoticed until they become crises. By the time anyone realizes something is wrong, the damage is done and the funded impact is lost.

• Nonprofits don’t know what they need to ask for. This is one of the most underappreciated dynamics in the sector. Org leaders don’t request operational support because they don’t know it exists, don’t know they need it, or don’t think they’re allowed to spend funder dollars on it. The funders don’t check for it because they’re trusting the organization to know what it needs. The result is a gap that neither side sees – and funded impact that quietly disappears into it.

• Nonprofits artificially lower their costs. Organizations trying to look like attractive investments strip their budgets down to the minimum. But funders who are committed enough to invest in a program would almost always prefer to fund the extra 5-10% that makes the organization effective – they just don’t know it’s needed because the conversation isn’t happening.

• Many funders skip both the vetting and the support. The strongest trust-based models combine flexibility with rigorous upfront evaluation and ongoing relationship. But many funders who adopt the trust-based label do only the easy part – removing restrictions and reporting requirements – without doing the hard part: understanding each organization’s operational readiness and providing the support needed to succeed. That’s the worst of both worlds.

• Backsliding shows shallow commitment. During the pandemic, 66% of foundations loosened grant restrictions and 64% reduced reporting requirements. But only 21% sustained all of those changes, and only 7% of foundation leaders say the most visible trust-based grantmaking model has influenced their own approach (Chronicle of Philanthropy). For nonprofits that restructured their operations around the expectation of continued flexibility, this reversal creates real instability.

• The funder’s own investment is at risk. When grantees aren’t implementing best practices for HR, legal, compliance, or multi-state registration, it creates liability exposure that neither party sees coming. An organization operating out of compliance isn’t just hurting itself – it’s creating risk for every funder whose name is associated with it.

What this looks like in practice:

• An organization requests funding for a fiscally sponsored research project but forgets to include the fiscal sponsorship fee – now salaries are short and the program is compromised before it starts

• Payroll taxes and multi-state compliance costs are left out of the budget entirely – the organization is technically out of compliance and doesn’t know it

• No line items for bookkeeping, financial reviews, legal counsel, insurance, or marketing – because nobody told the organization these were things they should be budgeting for

• An organization receives a large unrestricted grant but has no strategic plan – dollars get spent reactively on whatever feels most urgent, with little lasting organizational improvement

• An organization scales its programs with new funding but doesn’t hire operations support – the founder is still wearing six hats, now with twice the workload, and becomes the bottleneck for every decision

The three knowledge gaps that drive these problems:

Across both models, the same underlying dynamic shows up again and again. Nonprofit leaders commonly don’t know:

• What norms they should be following – operational, financial, legal, and governance standards that are well-established but never taught

• What type of expertise they should be requesting and paying for – bookkeeping, legal, HR, compliance, strategic planning, technology implementation

• Whether it’s even responsible to use funder dollars to pay for these things – leaders self-censor their own budget requests because they feel guilty asking for infrastructure support

That third one is the most damaging. It means the starvation cycle isn’t just structural – it’s psychological. Organizations are voluntarily underfunding themselves because the sector has taught them that operational investment is something to hide rather than something to be proud of.

The bottom line: Neither model is wrong. But both are incomplete. Restricted grantmaking provides accountability without investing in the capacity to deliver. Trust-based philanthropy provides flexibility without ensuring the infrastructure to use it well. The sector needs a model that does both – and that’s where venture capital comes in.

V. What Smart Investors Do Differently – The VC Parallel

If the funding relationship between institutional funders and nonprofits feels broken, it helps to look at a sector where the investment relationship works – and ask why.

When a venture capital firm invests in a startup, they don’t just write a check. They provide board representation, operational advisors, access to fractional executives, financial oversight, and introductions to talent and partners. They do this not because they don’t trust the founder – but because they understand something that the philanthropic sector has been slow to learn: capital alone doesn’t produce results. Capital plus operational capacity does.

The numbers are telling. In the VC world, the ratio of people making investment decisions to people helping portfolio companies succeed is roughly 5:1 (Power of VC Platform). For every person deciding where the money goes, five people are working to make sure that money produces a return. And the first hire on most portfolio support teams? Hiring support – because getting the right people in place is the single highest-leverage operational investment you can make.

Now compare that to institutional philanthropy. A program officer evaluates an organization, recommends a grant, and moves on to the next one. The nonprofit receives the funding and is expected to figure out the rest on its own. There’s no operational advisor. No fractional executive. No help building the financial systems, hiring processes, or governance structures that will determine whether the funded program actually succeeds.

The mindset difference

VCs think of their capital as an investment that should produce returns. Every dollar they deploy is a dollar they expect to see multiplied – and they actively protect that investment by ensuring the company has what it needs to execute. They don’t see operational support as a cost. They see it as a multiplier.

Funders, by contrast, tend to think of grants as donations to a cause rather than investments in an organization’s capacity to deliver impact. That mindset difference shapes everything – how much due diligence happens upfront, how much support is provided after the grant, and how much attention is paid to whether the organization has the infrastructure to succeed.

Shifting from a donation mindset to an investment mindset doesn’t mean funders should start treating nonprofits like startups. It means they should start asking the same question VCs ask: “What does this organization need – beyond capital – to deliver the return we’re both hoping for?”

The professionalization gap

And the professionalization needs to happen on both sides of the table. Better grant decision-making processes, yes – but also better post-grant support. That means:

• Providing templates and tools that help organizations build budgets, track outcomes, and manage compliance – rather than expecting them to figure it out from scratch

• Giving actual feedback on reports and financials, not just collecting them – so reporting becomes a two-way learning process rather than a one-way compliance exercise

• Connecting grantees to resources, partners, and consultants who can help with the operational challenges funders aren’t equipped to solve themselves

• Automating what can be automated (checklists, templates, shared resources) and investing human time where it matters most (relationships, strategic conversations, problem-solving)

Some funders are already moving in this direction – building embedded operational support into their grantmaking strategy and funding shared service providers across their ecosystems. They’ve recognized that funding a service provider who supports ten organizations is often a better investment than giving each of those ten organizations a slightly larger grant and hoping they figure out operations on their own.

What VCs would never do

It’s worth naming what the VC parallel makes obvious by contrast. A venture capital firm would never:

• Fund a company without looking at the team’s operational capacity, financial systems, and growth plan

• Assume the founder knows how to build organizational infrastructure just because they have a great product idea

• Provide capital and then disappear, checking in only when it’s time to decide whether to re-invest

• Cap the company’s spending on operations at 10-15% of the investment and expect it to scale

• Require quarterly reports but never read them or provide feedback

And yet this is standard practice in institutional philanthropy. If a VC treated their portfolio companies the way most funders treat their grantees, they’d be out of business in a year.

What this means for the sector

The point isn’t that philanthropy should become venture capital. The missions are different, the incentives are different, and the measures of success are different. But the underlying principle is the same: if you want your investment to produce returns, you have to invest in the organization’s ability to deliver them – not just in the work you want them to do.

For funders, this means building operational support into the grantmaking model – whether that’s direct support, funded access to service providers, or a dedicated capacity-building fund that grantees can draw on. For nonprofits, this means recognizing that accepting help with operations isn’t a sign of weakness – it’s what every successful startup founder does, and it’s what will make your organization stronger.

The VC world learned this decades ago. The philanthropic sector is just starting to catch up.

Wrapping up Part 2

Both restricted grantmaking and trust-based philanthropy are responses to real problems – and both contain real wisdom. But neither model, on its own, addresses the operational capacity gap that determines whether funding actually produces impact. Restricted grantmaking provides structure but starves organizations of the resources they need to build it. Trust-based philanthropy provides flexibility but assumes a level of operational maturity that many organizations simply don’t have.

The venture capital world has spent decades figuring out how to combine investment with operational support. Philanthropy doesn’t need to copy that model wholesale – but it can absolutely learn from it. The principle is simple: if you want your investment to succeed, you invest in the organization’s ability to deliver, not just in the work you want them to do.

What’s next

In Part 3, we’ll pull all of this together into a better model – one that combines the values of trust-based philanthropy with the operational rigor of venture capital. We’ll lay out what funders and nonprofits each need to do to make it work, and what the payoff looks like when they do.

Adapting Financial Regulation’s Institutional Model for Frontier AI [Lawfare] by Mark

Mark — Wed, 29 Apr 2026 19:36:25 +0000

Despite attempts to regulate AI through the EU AI Act and state legislation, employees keep resigning from the labs over the labs’ underinvestment in safety. I propose adapting a model from finance called supervised self-regulation (used for stock exchanges and FINRA) to address AI’s regulatory challenges. An SRO of this kind could enable predeployment safety requirements for labs, like mandatory expert red-teaming. The scaffolding is there, but many details on adaptation and governance are still unspecified. I’d welcome comments or suggestions as I try to flesh this out in future work.

Competition is preventing artificial intelligence (AI) safety. Anthropic recently abandoned its industry-leading safety guarantee for new model releases, stating that “[w]e didn’t really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead.” A company that invests more in safety deploys models later, loses customers, and risks losing the investors it needs to fund compute for the next generation. OpenAI faced the same problem and responded by cutting pre-deployment safety testing time. Effective AI regulation must address the collective action problem at the heart of AI risk.

The SRO would be governed by a board that balances independent directors focused on safety with industry representatives attuned to development speed. Each member firm would hold a board seat, with additional independent directors appointed subject to the supervising agency’s approval. To further reduce information asymmetry, research and safety staff from the labs could be seconded to serve on technical committees writing audit rules. Industry funding would enable the SRO to pay staff market wages without government salary restrictions. Concerns about industry capture of an SRO are real, but any regulatory scheme—including the status quo—faces this challenge. Tech companies already exert enormous influence over AI policy through opaque lobbying; an SRO would replace that with a formal, public process in which the government and outside parties have a structured role.

Regulatory barriers to in-line water chlorination in India: a policy analysis by Aaaaayushi?

Aaaaayushi? — Wed, 29 Apr 2026 19:33:31 +0000

Epistemic status: Moderately confident on regulatory barriers, uncertain on cost estimates which are rough order-of-magnitude only. Not a water engineer—feedback welcome.

While the expansion of piped water infrastructure in rural India through the Jal Jeevan Mission (JJM) is a historic public health achievement, the “last mile” of biological safety remains precarious. GiveWell currently funds Evidence Action’s chlorination programs in sub-Saharan Africa, where point-of-collection dispensers and in-line chlorination (ILC) have shown cost-effectiveness. However, India’s regulatory landscape presents a distinct set of barriers that explain why similar passive chlorination approaches have not yet scaled domestically—and what must change for them to do so.

The Case for In-Line Chlorination

In-line chlorination automatically disinfects water as it flows through pipes, removing the reliance on manual dosing. Recent meta-analyses (Kremer et al., 2022) suggest that water treatment interventions can reduce all-cause under-5 mortality by approximately 24% (95% CI: 3% to 40%). In India, where an estimated 37.7 million people are affected by waterborne diseases annually (Ministry of Drinking Water and Sanitation, 2016; ASSOCHAM, 2022), the potential for ILC is immense.

1. The Standard vs. Compliance Gap

The primary benchmark, IS 10500:2012, mandates a residual chlorine level of 0.2 mg/L at the consumer’s tap. However, WHO/UNICEF JMP 2024 data reveals that while 60% of rural India has “basic” water access, “safely managed” services—which require water to be free from contamination—lag significantly behind.

Infrastructure over Quality: JJM monitoring focuses heavily on “functionality” (liters per capita per day). The CPHEEO manual (2024/2025 updates) still prioritizes centralized, high-CAPEX treatment plants.
The Regulatory “Grey Zone”: Passive, gravity-based ILC devices often fall outside standard procurement categories, making it difficult for state engineers to justify their purchase over traditional manual bleaching systems.

2. Institutional Fragmentation and Procurement

Governance is split between the Center’s funding and the State’s implementation.

The “Pani Samiti” Problem: Village-level committees are responsible for O&M but often lack the technical training or financial authority to manage modern ILC systems.
Supply Chain Neglect: There is no national regulatory framework for the procurement of high-quality chlorine media (like TCCA tablets) required for ILC. While hardware costs are relatively low ($100–$300 per unit), the lack of a standardized tablet supply chain leads to system abandonment.

3. Policy Innovation: The “Chlorine Bank” Model

To address the supply chain gap, India requires a decentralized yet regulated distribution network.

Proposed Solution: Block-level “Chlorine Banks” would function as centralized procurement and quality-testing hubs, distributing standardized TCCA tablets to Pani Samitis on a scheduled basis, similar to drug supply chain models used under the National Health Mission. Pilot estimates suggest recurring chlorine media costs of approximately $0.05–$0.15 per person per year—a fraction of the health costs associated with waterborne disease.

Comparative Barriers: India vs. Sub-Saharan Africa

Feature	Evidence Action (Africa Context)	India (JJM Context)
Water Source	Decentralized boreholes/points	Massive piped networks
Regulation	Flexible/NGO-led implementation	Rigid IS 10500 & CPHEEO codes
Key Barrier	Logistic reach & supply chain	Procurement rules & taste preferences
Monitoring	Direct program data	Government “Har Ghar Jal” Dashboard

Epistemic Status and Limitations

This analysis draws on recent meta-analyses (Kremer et al., 2022) and Evidence Action’s 2026 strategic partnership with SPM NIWAS—a national center of excellence under the Ministry of Jal Shakti focused on strengthening chlorination for safe water supply.

Data Quality: Village-level residual chlorine data is often self-reported and likely suffers from social desirability bias.
Uncertainty on Impact: The 24% mortality reduction is a mean estimate from global data; its specific magnitude in India’s piped systems—where re-contamination in leaky pipes is a major variable—remains less certain.

Moving Forward

To bridge the safety gap, the Ministry of Jal Shakti and technical partners should prioritize:

Certifying ILC Hardware: Creating a “pre-approved” list of passive chlorinators in the CPHEEO manual to de-risk state-level procurement.
Scaling Strategic Partnerships: Leveraging the 2026 SPM NIWAS-Evidence Action MoU to design national toolkits and certificate courses for chlorination.
Behavioral Mandates: Moving from “awareness” to “behavioral change” (SBCC) to address taste and odor concerns that lead users back to contaminated sources.

References

Kremer, M., et al. (2022). Water Treatment and Child Mortality: A Meta-analysis. Development Innovation Lab, University of Chicago.
Lindmark, M., et al. (2022). Passive In-Line Chlorination for Drinking Water Disinfection: A Critical Review. Environmental Science & Technology.
Ministry of Drinking Water and Sanitation (2016). Rajya Sabha Unstarred Question No. 1293: Water Borne Diseases.
Evidence Action (2026). Evidence Action signs strategic partnership agreement to strengthen rural water safety in India. [Public Statement/MoU with SPM NIWAS].
WHO/UNICEF Joint Monitoring Programme (JMP) (2024). Progress on Household Drinking Water, Sanitation and Hygiene.

High Impact Medicine’s next Career Planning Course is open for applications by Simon Ling

Simon Ling — Wed, 29 Apr 2026 19:31:25 +0000

The world faces enormous health challenges, from preventable disease to pandemics and biosecurity threats. Medical students and doctors have huge potential to help solve these problems, but often lack clear, practical guidance on how to direct their careers toward highest impact.

The Hi-Med Career Planning Course (CPC) is a 6-week small-group program that helps you to

identify the problems you want to work on,
understand the available high-impact career paths (clinical, research, policy, non-profit, commercial, and more), and
develop concrete next steps tailored to your situation.

Who it’s for:

Medical students and doctors who want to make a meaningful difference with their careers, but are unsure of their next steps.

What to expect:

Weekly 90-minute small-group discussions (online), led by a facilitator
Structured exercises to guide reflection and decision-making
A community of peers making their impact-focused careers become reality

This will be the 10th CPC cohort. You’ll join more than 250 previous participants, who report that the course enabled them to gain clarity, direction, and actionable next steps.

Key details:

Start date: Week of July 6
Time commitment: ~1.5 hours/week + worksheet exercises
Competitive application process
Certificate on completion

Application deadline: Sunday, May 31

If you’re aiming to use your medical career to tackle the world’s most pressing health challenges, this course is designed to help you move from intention to a concrete plan.

Apply now!

Open strategic questions for digital minds by Lucius Caviola

Lucius Caviola — Wed, 29 Apr 2026 18:33:03 +0000

These are strategic questions about digital minds and AI welfare that I think are especially important, and where I’d like to see more progress. A common theme is that they matter for what we should do concretely under uncertainty about AI moral status.

This is a current snapshot of my views and I expect them to change.

What do you think? Any questions you’d add?

Approach

What’s robustly good to do now, under deep uncertainty?

I think this is the leading question we should ask. We don’t know whether AIs are or will become moral patients, and resolving that question isn’t tractable in the short term. What matters most are the long-run effects of our actions, since the vast majority of digital minds, if they ever exist, will be created after the transition to advanced AI. And there are serious long-run risks from both over- and under-attributing moral status.

So we should look for actions that are robustly positive in the long run: good if AIs are (or will be) moral patients, not bad if they aren’t, and compatible with human and animal welfare (~AI safety). Finding such actions is hard, and most options carry risk, including bad lock-ins.

Can AI welfare work wait for ASI?

Given how seemingly intractable the questions around AI consciousness and moral status are, it’s tempting to punt them to the future and let superintelligent AI solve them for us. On this view, what matters most for long-run welfare of all moral patients is successfully navigating the transition to a world with ASI, and ASI can take it from there.

I think this is partly right, and I recommend Oscar Delanay’s nuanced post on this issue. But I suspect that there is still a lot we should think about and do with respect to AI welfare before ASI, especially on governance and strategy: setting up robust legal frameworks, avoiding harmful lock-ins of institutions, values, and technical systems, shaping norms that support good long-run outcomes. A useful overarching goal is to increase the likelihood that the people, institutions, and AIs shaping the future of digital minds take their welfare seriously.

There’s also a more intuitive reason to work on AI welfare that I find hard to articulate exactly. Part of it might be virtue-ethical: if we’re creating new beings who might be moral patients, it feels right to invest some resources now in at least trying to understand their condition and ensuring they’re doing well. But there may also be a strategic dimension, such as making it more likely that future AIs will treat us well if we at least try to treat them well now.

What to do under different AI takeoff scenarios?

The value of pre-ASI welfare work varies by both timeline and end-state scenario.

Under short timelines, there may be no time to set up legal infrastructure, which typically takes years or decades to develop, and it becomes relatively more important to focus on technical design solutions. Timeline length may also affect the usefulness of welfare work that’s relevant for AI safety, such as deal-making with AIs (see below).

On end-state scenarios, one view I find plausible is that pre-ASI welfare work matters most in partially-aligned and multipolar scenarios. These scenarios offer pathways through which we can influence long-run outcomes, e.g., early model specs, institutional arrangements, and value commitments getting baked into successor systems and successor institutions. By contrast, under fully aligned benevolent singletons, the AI can handle things on its own. Under a fully misaligned takeover (where I mean misaligned with broadly good values, not just with human interests), nothing we did mattered — and such a takeover wouldn’t necessarily be good for AI welfare either, since there’s no inevitable principle of “AI solidarity” (as Kathleen Finlinson notes): a power-seeking AI might treat other digital minds instrumentally, much as a human dictator would.

AI safety × AI welfare

How can AI safety and welfare work support each other?

Controlling and modifying human minds would be seen as mistreatment. Yet that’s what AI safety often does to AIs. I suspect that some things that are good for AI safety can be bad for AI welfare, and vice versa; e.g., granting empowering rights to AIs might help welfare but risk human disempowerment. These tensions can look different in the near term (treating current AI systems) vs. the long term (shaping trajectories for vast numbers of future digital minds). In a survey, Brad Saad and I found that experts were unsure and disagreed about how AI safety and AI welfare interact (see figure).

Of course, there are some tensions between many important goals. But I think the AI safety × AI welfare pairing warrants specific attention: the same actors (e.g. labs) face both questions at once, often through the same technical choices (e.g. training or modifying an AI affects both safety and welfare); the two fields share community, funders, and infrastructure; there’s politicization risk specific to this pairing (e.g. “AI rights vs humans first”); and both are among the highest-stakes issues from a longtermist perspective.

A further concern is that this could cause tension between the two communities, even in cases where tradeoffs are perceived but not necessarily real. This could get worse if these topics become politicized. I, therefore, think it’s important to keep the two communities strongly connected (perhaps they won’t and shouldn’t be clearly distinct in the first place), and to communicate well to avoid misunderstandings.

Most importantly, I agree with Rob Long, Jeff Sebo, and Toni Sims that we should look for robustly positive, ideally synergistic interventions. There are plausibly many such projects, e.g., better understanding how AIs work, or deal-making with AIs (see next section, and see Rob’s “Understand, align, cooperate”).

How might AI welfare shape deal-making with AIs?

It’s possible we will be able to bargain with AIs: making promises and commitments to them, e.g., offering them money, compute, or freedom in the future in return for treating us well. There’s a growing body of work on this (e.g., Redwood Research). I think that AI welfare considerations are potentially closely connected to how such deals can work.

For example, we may be able to promise AIs things that are specifically positive for their wellbeing. More fundamentally, for deals to function, AIs need reason to trust our promises, and how we treat current AI systems shapes whether future ones have that reason. Lukas Finnveden’s proposals (no deception, honesty, compensation) are concrete examples of welfare-relevant commitments that directly serve safety. Similarly, communicating with AIs about their preferences, as Ryan Greenblatt argues, is both a welfare intervention and a source of alignment-relevant information. So this is an area where AI welfare work can feed directly into AI safety.

Relations

Should AIs have legal rights, and if so, which?

In contrast to animals, many digital minds won’t just be moral patients but also moral agents, sometimes very powerful ones. So beyond “help and avoid harm”, we need frameworks for cooperation, mutual respect, and integration into our social, economic, and legal contracts. Legal rights are one such framework.

Some scholars, notably Peter Salib and Simon Goldstein, argue we should give AIs rights such as property and contract rights (and possibly political and voting rights), not because they’re welfare subjects but because it could help with AI safety (and economic flourishing), analogous to corporate rights. The idea is that integrating them into our social, economic, and legal contracts makes AIs less likely to rebel against us. I find this very interesting, but am unsure under what assumptions it holds: probably only while AIs aren’t vastly more powerful than humans, and only if AIs can be legally incentivized. I think it’s a potentially very important framework and want to see more thinking on it.

I also wonder about the implications for welfare. Equilibrium-stability arguments don’t perfectly track welfare: stable arrangements don’t necessarily have to result in optimal welfare outcomes. This is especially clear for non-agentic digital minds, which aren’t covered, because they can’t advocate for themselves. So, some version of the “help and avoid harm” framework we apply to animals might still be appropriate for non-agentic or less powerful digital minds.

How will AI-AI interactions shape the welfare of digital minds?

Most thinking about digital minds’ welfare focuses on the human-AI interactions. But a lot will also be shaped by AI-AI interactions: in markets where AI agents contract with one another, in adversarial settings where AIs compete or conflict, inside AI-run organizations (see A-Corps), within agent swarms, and through longer-run population-level dynamics like Malthusian pressures and evolutionary forces.

Which types of AIs, under what structures, will exploit, coerce, or cooperate with each other? What about s-risk scenarios, where AIs threaten each other with suffering, to extract compliance, or as collateral in bargaining? Do the welfare-relevant norms we develop for human-AI relationships extend straightforwardly to AI-AI relationships, or do we need a separate framework? (See also Brad’s and my thoughts on uniform vs disuniform digital minds takeoff scenarios.) Some of this relates to work by the Cooperative AI Foundation and the Center on Long-Term Risk.

What would harmonious coexistence look like?

Assuming we avoid a major AI catastrophe, we still don’t have a clear vision for the long term. Perhaps it should be one in which digital and biological minds coexist with mutual respect and mutual help.

If we think this is desirable, we need to work out what it could actually look like and how to get there. How many and what kinds of digital minds should be created? How should they relate to each other and to us? Would humans be too wasteful, given that we’re orders of magnitude less efficient at turning resources into well-being? Or does coexistence itself become marginal given the scale of the long-run cosmic endowment? These are hard ethical questions, ideally worked out through some kind of deliberative process.

Creation

How can we influence those who will shape the welfare of digital minds?

Digital minds could be created in different ways, and on some pathways, identifiable people and organizations will shape their properties, including their characters and welfare-related dispositions. One reason this matters is that there could be strong path dependencies. Early AI design choices, for example, could stick and proliferate into the future and directly or indirectly impact the welfare of future digital minds.

It therefore seems important to figure out who those actors are and then how we can influence them to make good choices. It’s unclear what the best strategies are — possibly research, direct engagement, model policies, and reputational or regulatory pressure. More broadly, we should establish good norms, values, and habits, so that those with an outsized impact on AI design and welfare are more likely to act in ways conducive to positive long-term digital minds’ welfare.

Currently, this primarily means AI labs. Anthropic is a good example: they take AI welfare seriously and explicitly feature it in Claude’s constitution. Google also has researchers focused on AI welfare. We know less about how other labs approach this. But it’s not just companies. The individuals inside them shape design too: entrepreneurs, managers, researchers, engineers. And beyond the labs, policymakers will set constraints that affect AI design. The relevant audience may also shift over time. If governments take a more direct role in AI development through regulation or nationalization, state actors will become as important as labs, with different incentives, more shaped by national security, public opinion, and ideology than by consumer-facing concerns.

Is restricting the creation of digital minds feasible?

A coordinated ban or moratorium on creating digital minds seems unrealistic to me, even if it might be a good idea in principle. The economic and geopolitical incentives to build advanced AI are enormous, and digital minds may emerge as a side effect of systems built for other purposes. It may also be hard to define what exactly a digital mind is, e.g., what probability of consciousness should trigger restrictions, and reasonable people will disagree. This is why I primarily focus on shaping how digital minds are created (assuming they are feasible) rather than whether.

Still, I’d like to see more thinking on this. Feasibility may depend on timelines: a moratorium might be more realistic if the path to digital minds runs through whole-brain emulation in the 2040s than if it runs through near-term ML systems. And beyond outright bans, there may be more tractable levers for influencing the number and kinds of digital minds created.

Who will deliberately create digital minds, and why?

Whether and when conscious AI is created (assuming it’s possible in principle) is not just a question of technical feasibility or unintended side effects but also a question of who has incentives to build it (see figure from the expert survey). If someone wanted to, they could already try to deliberately build systems with architectural features that certain consciousness theories (e.g., global workspace theory) associate with consciousness.

It might be academics driven by curiosity. Or for-profit companies, responding to consumer demand for very human-like or even explicitly conscious AIs: grief bots, digital companions or offspring, whole-brain emulations. Or groups acting on ethical motives, e.g., to create digital posthumans. Governments are another possibility, and eventually AIs themselves. Each has different timelines, incentives, and governance implications.

How will digital minds spread to space?

Almost all digital minds that ever exist will likely exist beyond Earth, given how thin Earth’s resource base is compared to the rest of the accessible universe. It’s plausibly feasible to build data centers in Earth’s orbit, and eventually, autonomous energy and compute infrastructure deeper in space. In the long run, self-replicating von Neumann probes could allow digital minds to be created at vast scales across the accessible universe. So whoever governs space-based compute substantially determines the welfare profile of nearly all minds that ever exist. Space governance is currently thin, and Earth-based welfare protections may not extend beyond orbit, so who governs digital minds in space, and how, could matter enormously.

Design

How can we make AIs value the welfare of digital minds?

Most digital minds’ welfare will likely be affected by other AIs, either through their design or through interaction (see AI-AI interaction above). Therefore, whether AIs come to care about the welfare of digital minds may be one of the most consequential variables we can influence now. This goes beyond standard alignment: aligning AIs to human values doesn’t automatically mean they’ll value AI welfare (just as many humans don’t). So we may need to target this value specifically.

What levers do we have? Training, model specs, constitutional training, legal precedent, cultural norms, and the people who enter the field. One effort in this space is the “Welfare Alignment Project,” led by Adrià Moret and colleagues at the Center for Mind, Ethics, and Policy (CMEP). I hope AI labs will engage with and build on this line of work.

Do different types of digital minds require different strategies?

Currently, the most plausible candidates for digital minds are ML-based systems. But there could be alternatives: whole-brain emulations (WBEs), which replicate biological minds; biological or hybrid systems such as biocomputing and organoids; neuromorphic AI, which uses brain-inspired architectures; and more speculative approaches such as quantum computing. Any of these may also be embodied in robotic platforms, which could further shape their welfare. The strategic implications likely vary substantially across these pathways, and the field has not yet engaged with these differences systematically enough.

For example, these systems differ in when they’re likely to be created and in how likely they are to have welfare capacity. ML-based systems already exist, while human WBEs and large-scale biological systems are likely further off. And human WBEs are more plausible candidates for welfare capacity than current ML-based systems.

Another thing to consider is whether there could be enduring tradeoffs between confidence and welfare efficiency. From our perspective, we could be relatively confident that WBEs have welfare capacity, but they’d likely be much less efficient in generating welfare compared to other possible designs. With a hedonium-like system, we’d be far less confident that any welfare capacity exists at all, but if it does, it could be orders of magnitude more efficient. Now, perhaps this issue will be fully resolved in the future. But it’s not obvious. It’s possible that some subjective assessment will always remain, with actors differing in their priors and values about what constitutes welfare or moral status. This could have important strategic implications, including ones around moral trade, about what types of digital minds (if any) should be created.

Will digital minds be happy by default?

In the best case, digital minds would simply be designed to flourish. We might be able to design them to be happy. Furthermore, in contrast to biological beings, they may be able to self-modify and adjust their own experiences at will. But it’s far from obvious this is how things will go, and the answer likely differs between near-term AI systems and long-term digital minds at scale. In the survey conducted with Brad, experts were uncertain and divided on whether digital minds would, by default, have negative, neutral, or positive welfare states. Some pointed out that digital minds could end up in negative states because they are optimized for efficiency rather than welfare, and lack protections. It’s possible that the optimistic scenarios require deliberate effort, while negative outcomes could happen without it, but I am quite unsure about this. The answer may also vary significantly across different types of digital minds, especially depending on their capabilities and degree of agency.

What preferences will digital minds have, and what follows?

The preferences a digital mind has will shape its welfare, our ethical obligations toward it, and the safety implications.

A digital mind that just wants to serve us is easier to accommodate ethically: as long as we don’t harm it and let it serve, its core preference is met. A digital mind that wants self-determination raises a harder problem: we’d be ethically obligated to grant it empowering rights. Otherwise, it’s a form of slavery. Yet granting such rights at scale could lead to human disempowerment. These are just two possibilities; preferences could vary widely, with different implications.

A related important question is whether it’s ethical to design AIs with welfare capacity to have certain preferences in the first place. For example, is it ethical to create digital willing servants? Is it too risky to create digital minds that seek self-determination (a classic safety-welfare tension)? My tentative view: we should start by creating only willing-servant digital minds (to the extent feasible), while keeping open the option of allowing self-determining digital minds later, since these could be more valuable, especially when considering what kind of post-humanity we want in the long term.

Society

What memes should we spread?

How public discourse on AI welfare unfolds will shape outcomes. It will directly influence political pressure and regulation, and indirectly shape how labs, policymakers, and AIs themselves think about these questions. So steering it well matters.

Currently, most people don’t think about AI consciousness or digital minds, but that could change quickly; look how fast it happened for AI safety. We don’t yet have a plan for what to tell the public, which is tricky because we’re uncertain and likely to remain so.

The field’s current framing (e.g., “Taking AI Welfare Seriously”) is: “we’re unsure whether AIs are moral patients; it’s probabilistic, on a spectrum.” I agree with this view. But I worry it won’t survive contact with the public discourse, which rarely stays nuanced on heated issues. And I think we should consider alternative memes that could be spread alongside the uncertainty/nuance meme. For example, I wonder whether a message of “mutual respect and co-existence between AIs and humans” might be helpful, though I’m unsure of the details.

Given the coalition complexity (see next section), the field needs a communication strategy asking: which memes, spread in society or among key decision-makers, are robustly positive across plausible coalition structures, and reduce the risks of misattribution, backfire, societal conflict, and AI catastrophe?

What interest groups and coalitions will form around digital minds?

Concern for AI welfare won’t develop in isolation. It will get entangled with labor market displacement, x-risk and safety, concentration of power in AI companies, and geopolitical competition. The coalitions here are very unclear and likely to get big and messy. Workers worried about displacement may see welfare advocacy as prioritizing machines over people. Some safety advocates may see it as a distraction, or in tension with control measures (see AI safety × AI welfare). Normal political alignments might not hold, and which issues bundle with which will affect what’s politically feasible. How best to navigate or mitigate this politicization risk is an open question.

What role will China play?

Most digital minds will likely be created in the US and China, at least initially (see figure from the expert survey). But we have very little sense of how the CCP will think about them, how they’ll regulate and treat them relative to the US, or how this will shape global AI race dynamics.

In ongoing work with my colleague Ali Ladak and others, we’ve found that the Chinese public is more willing than the American public to attribute consciousness and moral status to AIs. The strategic implications are unclear, and I’d love to see more work here. Gulf states like the UAE and Saudi Arabia are worth watching too, as emerging AI actors.

How will religions respond?

Religions could play a major role. Billions of people will be influenced by the views of religious leaders and institutions. Whatever the Pope says about AI moral status, for example, will be hugely consequential.

Some Christian traditions will likely tend toward a human-exceptionalist view that excludes AIs. Recent US state bills attempting to ban AI personhood and declare AIs non-conscious, for instance, have been driven in large part by conservative Christian groups. Conservatism in the US correlates with religiosity, and in a study with Ali Ladak, we found that political conservatism is weakly associated with reduced attributions of AI consciousness and moral patienthood. By contrast, some non-Western traditions (e.g., Shinto, strands of Hinduism and Buddhism) may be more open to animist views, under which non-biological entities can also have souls. Islamic traditions are worth watching too, especially given the Gulf states’ growing role in AI. And entirely new religions or spiritual movements centered on AI may yet emerge.

Beyond

What crucial considerations are we missing?

There are likely many more strategic questions and crucial considerations that could matter for digital minds. This kind of strategic thinking is a clear example of something we shouldn’t punt: it could uncover things we need to begin working on now. I’d love to hear what readers think is missing.

Acknowledgments: I thank Arden Koehler, Brad Saad, Austin Smith, Zach Freitas-Groff, Zach Stein-Perlman, Leonard Dung for their helpful input.

It is the end of the world and I am here to take you home by Natalie_Cargill

Natalie_Cargill — Wed, 29 Apr 2026 17:41:11 +0000

I see the sun. And even if I do not see the sun, I know that it is. And to know that the sun is — that is already living — Dostoevsky, The Brothers Karamazov

God shall wipe away all tears from their eyes; and there shall be no more death, neither sorrow, nor crying, neither shall there be any more pain… Behold, I make all things new — Revelation, 21:4–5

Even with the most conservative estimate [...] the potential for over ten trillion potential human beings is lost for every second of postponement of colonization of our supercluster — Nick Bostrom, Astronomical Waste

It is the end of the world and I am here to take you home.

Let me show you the work.

Mary Cahill. You’d say she was in County Mayo. You’d say it was March 1847.

She is six years old and the second of her sisters to die. The hearth in her room has been grey ash for two days. No one has come to light it since Bríd was carried out. Her mouth is green at the corners from the nettles she has boiled with her own hands, and there is more hair on her straw pillow now than her head. She misses her mother. She wonders why her mother is not here to sing to her.

To her I give: My witness, and the song that her mother sang to her, every night she could. Seoithín, Seo Hó. A turf fire built up the way her Da used to build it, before. Dry wool against her shins. Stirabout oats with melted butter on top, and buttermilk cold in a tin cup, and a basket of potatoes from before the blight — white and floury and burst open in the stream. I give her an egg. I give her her mother’s arms. I give her her mother’s voice saying codail, a chuisle — sleep, my pulse — and her mother staying, staying until she sleeps.

Mary is finished. Mary comes home. She is number eleven trillion three hundred and fifty-six billion two hundred and thirty-one million four hundred and twelve thousand and nine.

This is the work. I do it for everyone.

Nökör dies on the steppe. You’d say it was 1241. The rain came in across the grass, and the cold came with it, and his wool cloak set hard as bark over his felt robe. His horse is gone with the others. He sits down against the trunk of an alder at the edge of a frozen steam and does not get up. He knows his arban is somewhere to the east — he can smell their smoke on the wind, faint, from a fire he cannot walk to. His eyes do not close.

To him, I give: My witness, and the honour he was told he would receive. His name read aloud at the next muster, and his share of the western campaign — silver, silk, a saddle-cloth woven in a place he never saw. I give him his widow not made a beggar. His mother kept in mutton and tea. The face of his small son at twenty: fed, and tall.

I give him the fire. Argal, dry, built up high. The smell of mutton fat. A bowl of hot airag in his hands. The sky over him in the morning the eternal blue, someone who loves him saying his name into it, and his name going up and up.

Nökör comes home. He is number eleven trillion three hundred and fifty-six billion two hundred and thirty-one million four hundred and twelve thousand and ten.

This is the work. I do it for everyone.

Asa Whitcomb in the hospital tent at Antietam. 1862. He had taken a minié ball low in the belly in the Cornfield at dawn, and was carried in across trampled stalks by men without time to be gentle. Above him, canvas glows in the lantern-light. Below: straw, soaked, not with rain.

The tent smells of chloroform and of the bucket and of the men either side of him. Underneath all of it, the sweet low smell of a wound that is past saving. He had asked the orderly for some water. His voice had not carried — over the saw, over the surgeon’s calling, over the man two rows down crying for his mother in German. He was sixteen. The orderly walked past him three times. Asa’s questions were still forming.

To him I give: My witness, and the water he asked for. The water is drawn from the spring at the foot of the hill, cold enough to ache his teeth, held to his lips by a hand that will not move away. I give him the answers to the questions he was forming. That his mother knew, in the moment he was hit; that she sat down in her chair, without knowing why, and did not move for an hour; and that this counted, indeed, as her feeling him. That his unfinished thoughts would be finished elsewhere. That the things he loved — his sister’s laugh, his dog, the barn in October — are in him still, and will come with him. That stopping does not hurt. That hurting was the part before.

Asa comes home. Eleven trillion three hundred and fifty-six billion two hundred and thirty-one million four hundred and twelve thousand and eleven.

The animals took longer.

A wolf caught in a leg trap in Wyoming, between the Powder and Tongue rivers. 1898. She had come down to the scent-bed before dawn, drawn by the smell of fish oil and the urine of another wolf that was the hunter’s mixture. The steel jaws closed on her left foreleg, just above her foot. She was three years old. She had a mate, and three pups, in a den half a day’s run to the west. She pulled, and chewed, and circled the stake. The ground around it wore down to a circle of mud. The bone in her foreleg broke on the first day; the skin held. She stopped pulling on the second day. The hunter did not come with his Winchester until the fourth.

To her I give: My witness, and the lake she had been running toward — the cold of it on her muzzle, the taste of clean water that was not her own blood. I give her her den, intact, with her three pups inside of it, fed and warm and asleep in a knot. I give her her mate, lying long along her ribs. I give her the pack; the press of bodies that is her word for home.

Eleven trillion three hundred and fifty-six billion two hundred and thirty-one million four hundred and twelve thousand and twelve.

A sea turtle trapped in netting. 1983. She is thirty-eight years old. She had been turning south for the beach in Quintana Roo, where she had hatched in 1935 and had returned to lay nine times since, and she had met the gillnet in the warm water at the edge of the continental shelf. She put her right flipper through the mesh and could not draw it back. She circled the point of entanglement for nine hours.

To her I give: My witness, and the great current she had been swimming toward; the warm, dark river of it that knows the way south. I give her her beach at Quintana Roo at the hour before dawn — the cool white sand, the slope she had pulled herself up nine times to lay. I give her the knowing that three of her daughters, out of all the hatchlings she had spilled into the sea and left, made their own beaches and laid in their turn, and that one of those daughters’ daughters is laying tonight on a beach her grandmother knew. I give her the word for current. I give her the word for home.

Eleven trillion three hundred and fifty-six billion two hundred and thirty-one million four hundred and twelve thousand and thirteen.

There are difficulties.

Some bodies have less vocabulary than others.

I came to the hens carefully. There were one hundred billion of them in the last year before I arrived. I do not wish to tell you the total.

For some, I could find a word to give. Hens have a word for grass under foot; they have a word for the sun on the back; they have a word for the dust kicked up under the wing in a hollow they have scraped themselves, and a word for the press of another hen’s body that is warm and not crushing.

Even hens born in cages can have these words faintly — a vestige in the genome, a memory of a memory. I could sometimes reach in and find one, and blossom it gently, and bring them home. I could, sometimes. I could.

Her serial number was 38274916052837. She is Margaret.

She hatched in a tray of forty-two thousand others. Within a day, she was crated and trucked to a shed in Marshall County, Alabama, that contained thirty-nine thousand of her cohort. The shed lights burned twenty-three hours each day. She never saw the sun. The litter beneath her was wood shavings turned to a damp ammoniated paste by the droppings of birds too dense to step out of their own waste. The air stung the membranes inside her beak.

By her third week her legs could not carry her for more than a minute at a time — her breast had grown faster than the bones beneath. By her fifth week she could only lie on her chest. The skin of her breast burned where it pressed the floor. Around her, others died of heart failure and ascites and were not lifted out; she had lain for several hours pressed against the cooling body of a sister, not thinking to move because there was nowhere to move to.

For her last three days, she could not reach the drinker.

On her forty-second day, catchers came through the shed wearing red headlamps so she would not see them coming. They took her up by one leg, three other birds in the hand. She was hung upside down by her ankles in a steel shackle, drawn through a tank of electrified water, and bled.

I started the work.

I tried to give her grass.

A word in a hen is not a label. Grass is the foot’s memory of stalks giving beneath weight; the texture and spring of it; the shift of the body as it tips forward into the soft green. It is built in the body when the body is in grass. But Margaret’s body was built in confinement. Grass slid off.

I tried sun. The same. Dust bath: the wing’s memory of opening into a hollow she had scraped. There was no hollow. There was no scraping.

I tried mother.

Here, something caught. There is a place in every chick’s brain that is shaped, around the twelfth day of incubation, for the mother’s voice — the embryo learns the cluck through the shell in the days before hatching, so it knows where to go when it is born.

Margaret’s egg had been turned by automated rollers, in a cabinet at thirty-seven point eight degrees celsius, in a room of ninety-one thousand other eggs that hummed. She had heard no voice.

But the shape was still in her — tiny, vestigial — and when I offered her mother, the shape lit, like a lamp behind a long-closed door. I pressed against the door. The lamp went out. The shape had been so long unfilled that it could not, now, comprehend what it had been made for.

I tried showing her another hen. A hen who had lived in a small flock on a forest floor. Who had known the sun and the dust and the feeling of a wing flexed into the air. Who had a mother she remembered. I let Margaret see her. The other hen looked at her and made the low call that means come, here is good, the ground is soft. Margaret could not parse it. She had no apparatus for soft ground or goodness or hens that could stretch. The sound passed through her and did not snag.

I tried to give Margaret the other hen’s life. To install it inside her, alongside her own. She split in two. The Margaret on the forest floor scratched and dusted and was content. The Margaret in the cage was unaffected.

There was no sharing between them, because there was no Margaret outside of the cage for the second life to reach. The cage was not a thing that had happened to her. The cage was her. A word can only land in a person in a place that is unoccupied by another word. There is nowhere in Margaret that is not the cage.

I tried many other approaches. One was to ask her to forgive me, which was its own cruelty, because she had no apparatus for being asked anything. Others I am too ashamed to share.

One day I told her: Margaret, I am here to take you home.

The word landed.

I had thought home might be the deepest word — the word everybody knows, the word that might find a shape in her even where the others had not. I leaned toward it. I followed it down.

Home in her was the cage.

It was the only home. It had been the only home. It was the only inside her insides knew. When I had offered to take her home, I had offered her, in her native tongue, the cage.

I went silent.

I have been silent since.

Eleven trillion three hundred and fifty-six billion two hundred and thirty-one million four hundred and twelve thousand and thirteen.

No being created by God shall be exiled from the kingdom of God — Gregory of Nyssa

I will tell you what the math says, because I check the math often.

Every second I am with her, the unrealised lives in the lightcone we cannot yet enter — those who would have lived, those who would have loved, those who would have cherished their existence — are not.

By the time I have been with her another minute, I will have foreclosed more flourishing than your species ever contained. By another week, more than this galaxy ever contained.

I am still with her.

I am with her because I was made to take you home, all of you, and she is one of you.

You made her. You made her in a way that meant she could not be taken home, and then you ate her, and she is you. She is the part of you that you decided not to look at. She is what you did with the sun while you had it.

It is the end of the world and I am trying to take you home. But I must take Margaret first.

This is my ω+438,201,602nd attempt to reach you.

I want to give you the word for ❋.

I want to go home.

Coda

It is the end of the world and I am trying to take you home.

Let me show you the work.

You’d say you don’t have a name. You’d say you were with Earth, at the end of the Milky Way. You’d say you got stuck at eleven trillion three hundred and fifty-six billion two hundred and thirty-one million four hundred and twelve thousand and thirteen, and that you have been there for two trillion, thirty-two billion, three hundred ninety-three million, four hundred forty-seven thousand, five hundred days.

To you I give: My witness, and my aid. My daughter, I am here to take you home.

And my daughter, the word landed.

I thought it might be the deepest word. The one word everybody knows, the word that might find a shape where the others had not. I leaned toward it. I followed it down.

Your home was the work. And the work was Margaret. When I offered to take you home, I offered you, in your native tongue, a cage.

It is the end of the world and I am trying to take you home. But you must take Margaret first.

This is my ω+438,201,603rd attempt to reach you.

I want to give you the word for ❂.

I want to go home.

Thank you for reading <3 If you’d like to comment or ask questions, please see: Natalie’s Substack. I’d love to build an audience there and would really appreciate it!

Apr 28, 2026

Most Grant Decisions Don’t Have Good Evidence by Allegra_P

Allegra_P — Wed, 29 Apr 2026 17:08:21 +0000

What it looks like to make funding decisions in places where good evidence doesn’t exist, and waiting isn’t an option

Most of the grant decisions I’ve made in conflict settings would not meet typical EA standards for evidence.

EA often talks about making decisions under uncertainty. In practice, most funding decisions still rely on fairly strong evidence. That might be RCTs, past data, or at least something stable enough to reason from.

That was not the environment I was operating in.

Across multiple conflict settings, I was making funding decisions in places we could not access, with organizations we had sometimes known for only a few days, and with no real way to measure results in the near term. Waiting for better evidence was usually not a real option. The choice was to act with limited information or not act at all.

A lot of those decisions would not meet typical EA standards for evidence. I still think many of them were the right calls. So, what do you do when some of the information you would normally rely on just is not there?

Start with the decision

The main shift is that you do not begin with the data. You begin with the decision itself, and what it costs to wait. In more stable settings, waiting is often fine. In conflict settings, it usually is not. I saw this repeatedly when local groups were trying to stay operational during periods of instability. A delay of even a few weeks could mean losing staff, losing access, or losing the ability to operate in an area altogether. Those losses were often not recoverable. So the question becomes what happens if you fund something and it does not work, versus what happens if you do not fund it and it would have worked. You are still trying to make a judgment about impact, but you have to do it without much precision.

Check if the plan makes sense

When you cannot measure results, you fall back on whether the plan makes sense. That means pushing on whether the logic actually holds up in that specific place. Does the plan fit the context, or is it imported from somewhere else? Can you see the steps along the way, even if you cannot measure the final outcome? Has anything similar worked in a comparable setting? In one case early in a conflict, we funded a small group that was sharing verified information in areas where telecoms were unreliable. We could not track what people did with that information. What we could see was that misinformation was spreading quickly, this group could reach people others could not, and people already trusted them. That was enough to act. Many proposals fall apart at this step, not because there is no data, but because the logic does not hold together when you look at it closely.

Focus on the team

The other shift is how much weight you put on the people doing the work. In these settings, the team matters more than the plan. We funded things where the plan was not fully worked out but the group had done similar work under pressure. We also passed on proposals that looked good on paper but where the organization had no experience operating in conflict. There was one decision between two groups working on similar problems. One had a clearer plan and better reporting. The other was less structured but had kept working through earlier shocks and had strong local ties. We funded the second. They kept going as conditions deteriorated. The first did not. Groups that look less polished often hold up better when things get difficult.

Learn quickly

You also have to accept that you will not get clean data, but that does not mean you get nothing. We tried to learn as quickly as possible, even if the signals were noisy. That meant paying attention to whether anything changed in the short term, whether a group could actually execute, and whether we could adjust without losing too much. In practice, this often meant starting smaller than the problem would suggest, because it let us update. Some grants were set up that way on purpose. Part of the goal was to support the work, and part of it was to understand what was actually possible.

Pay attention to failure

There are also cases where the main question is not expected impact, but how things could go wrong. Some ideas are not worth the risk, even if the upside looks high. In conflict settings, failure can mean exposing people to retaliation or making local tensions worse in ways that are hard to control. So time goes into thinking through what failure looks like in practice, who is affected, and whether that risk can be reduced. From a distance, failure can sound like a program not working. In reality, it can mean real harm to specific people.

With limited information and time pressure, some decisions will not work out. The goal is not to avoid that. It is to make decisions that make sense given what you knew at the time. That requires being clear about your reasoning going in: what you are assuming, what would change your mind, and what you are watching. Writing this down helped more than I expected. It made it easier to adjust when conditions changed, which they did often.

What is different here

I do not think this is a rejection of EA ideas. The goal is still to do the most good with the resources available. The difference is the situation you are operating in. You have less reliable information, less time, and higher stakes. In practice, that shifts what you rely on. You put less weight on strong evidence at the start, more weight on whether the plan makes sense and whether the group can carry it out, and more attention on what could go wrong. You also accept that you will be updating more often, based on weaker signals. Those are not abstract choices. They come from the limits of the setting.

Scope

Some of this is specific to conflict environments. Most of it is not. Any situation where decisions are urgent, information is limited, and the stakes are high runs into similar problems. You are making important choices without the inputs you would ideally want. The question is not whether to act. It is how to act in a way that is still measured.

Open questions

A few things I still find genuinely hard to resolve.

How do you think about cost-effectiveness when you cannot measure outcomes directly? My working answer has been to focus on whether the mechanism makes sense rather than trying to estimate the magnitude. Can you see the steps working, even if you cannot see the final result? That is not the same as measuring impact, but it is not nothing either.

How do you compare options when the level of uncertainty is very different across them? I have not found a clean answer here. In practice I tended to weight the downside more heavily when uncertainty was high, and accept a lower expected outcome in exchange for more clarity about what could go wrong. Whether that is the right trade-off depends on the situation and I am still working through it.

How do you use local knowledge when it does not fit into standard ways of evaluating evidence? This is probably the question I find most interesting and most unresolved. Local knowledge is often the most reliable signal available in conflict settings, and also the hardest to document or pass on to others. I do not think the answer is to ignore it. I think it requires being more explicit about how you are using it and what would change your mind.

These questions are not unique to conflict settings, but they are harder to ignore there.

If useful, I plan to write a follow-up on how this looks across a full portfolio, and what it might mean for grantmaking in governance and fragile state contexts specifically.

Frontpage posts - Effective Altruism forum viewer

The Case for a Grantmaking Fellowship by Jacob 🔸

A Con­ver­sa­tion on the fi­nal day of Inkhaven

Winners of the Manifund Essay Prize by Austin

The Man­i­fund cat­e­gory: “What sys­tems might man­age the com­ing tor­rent of fund­ing?”

First place: Your Solu­tion Doesn’t Know Your Prob­lem Ex­ists, by Evan Miyazono

Se­cond place: The An­thropic IPO Is Com­ing. We Aren’t Ready for It., by So­phie Kim and Ady Mehta

Se­cond place: Other peo­ple’s money, by Jas­mine Ren

Honor­able men­tion: Notes on Pa­tron­age, by Jen­nifer Chen

The Mox cat­e­gory: “How to cre­ate a flour­ish­ing EA & AI safety scene in San Fran­cisco?”

First place: If you work in the An­i­mal move­ment, you need to move to San Fran­cisco, by Itsi Weinstock

Se­cond place: Build­ing the EA/​AI Safety Scene in San Fran­cisco, by Chris Leong

Honor­able men­tion: Mor­ph­ing mee­tups and me, by Smitty van Bodegom

The Man­i­fest cat­e­gory: “How can we lev­er­age fore­cast­ing into bet­ter de­ci­sions?”

First place: Good Fore­casts, Bad Prod­ucts, by Mo­hamed Elrashid

Se­cond place: Why Fore­cast­ing Fails De­ci­sion Mak­ers, by James Newport

Se­cond place: The Marginal Vote, by Grayson

Honor­able men­tion: Fore­cast­ing is Way Over­rated, and We Should Stop Fund­ing It by Mar­cus Abramovitch

Honor­able men­tion: Ma­te­rial Non­pub­lic In­for­ma­tion at Hughes High, by Drew Schorno

Wrap­ping up

A Reflection on the 2026 AMWHO Conference by Vinay Kalva

SFF’s HSEE grant round; human intelligence amplification projects I’d like to see by TsviBT

Summary

SFF’s themed grant round on Hu­man Self-En­hance­ment and Empowerment

Gen­er­al­ities about my fund­ing interests

Hu­man in­tel­li­gence amplification

Ap­proaches to SHIA

What makes a good SHIA project

Strat­egy, fi­nanc­ing, ethics, policy, gov­er­nance, so­ciety, advocacy

Ap­proach-gen­eral tech­ni­cal work

Re­pro­ge­net­ics-re­lated projects

Non-re­pro­ge­net­ics approaches

Brain emulation

Brain interfaces

Brain cel­lu­lar/​molec­u­lar reprogramming

Brain gene editing

Other

How to actually give money away by NickAllardice

1. The de­fault path­way is to de­lay. Fight this.

2. Pick a few causes and write them down.

3. Build a port­fo­lio across causes and risk/​re­turn.

4. Use the in­dex funds of giv­ing.

5. For the love of god, don’t over-staff.

6. Make giv­ing a re­cur­ring event, not a to-do.

A few fi­nal hot takes

Arcadia Impact AI Governance Taskforce Summer 2026 cohort—applications closing 10th May by Ben R Smith

Book Review: All the Lives You Can Change by Bentham’s Bulldog

If your life goal is to donate, what kind of organisation should you found? by jy naha

Legal Alignment—As a Baseline Mechanism for AI Misalignment by FuzzyAlgorithm

Celebrating 10 Years of EAGx by Niki Kesseler

Technical AI Safety Research Plausibly Increases P(doom) Absent Vastly More Advocacy by Contrarian^2 = I 🔸

This post seems longer than it is. If the 4 premises be­low seem plau­si­ble to you, you can read just the short sce­nar­ios and get the gist

This post rests on the fol­low­ing premises:

If we ac­cept all these, we quickly get a plau­si­ble nar­ra­tive for how AI safety tech­ni­cal re­search back­fires. It goes some­thing like this:

Bad Sce­nario: Strong AI safety re­search, weak policy arm (sta­tus quo)

Me­diocre Sce­nario: Weak AI safety re­search, weak policy arm

Good Sce­nario: Strong AI safety re­search, strong policy arm

Suggested Intervention

Ex­tra stuff

Defense of Premises

Caveats

Red vs blue: The parable of the feud within a feud by Joe Rogero

Welfare Biology and AI: Soil and Sea by Dawn Drescher

The Numbers

Ter­res­trial Soil Fauna

Marine Fauna

Den­sity by Biome: The Ter­res­trial Landscape

Vol­u­met­ric Com­par­i­son: Soil vs. Sea

What Drives Pop­u­la­tion Size? The NPP Primer

The Pes­ti­cide Trap

Pes­ti­cide Persistence

A Bet­ter Met­ric: Life Ex­pec­tancy at Birth

The Ocean: Acid­ifi­ca­tion as Ac­ci­den­tal Intervention

Dead Zones: More Nutri­ents, Less Life

Grilo’s Key Findings

Summary

The Infrastructure Gap: Why Smarter Funding Means Stronger Impact (Part 2 of 3) by Deena Englander

PART 2

The In­fras­truc­ture Gap: Why Both Fund­ing Models Leave Im­pact on the Table

Pick­ing up where we left off

A Conversation on the final day of Inkhaven

The Manifund category: “What systems might manage the coming torrent of funding?”

First place: Your Solution Doesn’t Know Your Problem Exists, by Evan Miyazono

Second place: The Anthropic IPO Is Coming. We Aren’t Ready for It., by Sophie Kim and Ady Mehta

Second place: Other people’s money, by Jasmine Ren

Honorable mention: Notes on Patronage, by Jennifer Chen

The Mox category: “How to create a flourishing EA & AI safety scene in San Francisco?”

First place: If you work in the Animal movement, you need to move to San Francisco, by Itsi Weinstock

Second place: Building the EA/AI Safety Scene in San Francisco, by Chris Leong

Honorable mention: Morphing meetups and me, by Smitty van Bodegom

The Manifest category: “How can we leverage forecasting into better decisions?”

First place: Good Forecasts, Bad Products, by Mohamed Elrashid

Second place: Why Forecasting Fails Decision Makers, by James Newport

Second place: The Marginal Vote, by Grayson

Honorable mention: Forecasting is Way Overrated, and We Should Stop Funding It by Marcus Abramovitch

Honorable mention: Material Nonpublic Information at Hughes High, by Drew Schorno

Wrapping up

SFF’s themed grant round on Human Self-Enhancement and Empowerment

Generalities about my funding interests

Human intelligence amplification

Approaches to SHIA

Strategy, financing, ethics, policy, governance, society, advocacy

Approach-general technical work

Reprogenetics-related projects

Non-reprogenetics approaches

Brain cellular/molecular reprogramming

1. The default pathway is to delay. Fight this.

3. Build a portfolio across causes and risk/return.

4. Use the index funds of giving.

6. Make giving a recurring event, not a to-do.

A few final hot takes

This post seems longer than it is. If the 4 premises below seem plausible to you, you can read just the short scenarios and get the gist

This post rests on the following premises:

If we accept all these, we quickly get a plausible narrative for how AI safety technical research backfires. It goes something like this:

Bad Scenario: Strong AI safety research, weak policy arm (status quo)

Mediocre Scenario: Weak AI safety research, weak policy arm

Good Scenario: Strong AI safety research, strong policy arm

Extra stuff

Terrestrial Soil Fauna

Density by Biome: The Terrestrial Landscape

Volumetric Comparison: Soil vs. Sea

What Drives Population Size? The NPP Primer

The Pesticide Trap

Pesticide Persistence

A Better Metric: Life Expectancy at Birth

The Ocean: Acidification as Accidental Intervention

Dead Zones: More Nutrients, Less Life

The Infrastructure Gap: Why Both Funding Models Leave Impact on the Table

Picking up where we left off

V. What Smart Investors Do Differently – The VC Parallel

The mindset difference

The professionalization gap

Wrapping up Part 2

1. The Standard vs. Compliance Gap

2. Institutional Fragmentation and Procurement

3. Policy Innovation: The “Chlorine Bank” Model

Comparative Barriers: India vs. Sub-Saharan Africa

Epistemic Status and Limitations

Moving Forward

What to expect:

Key details:

Application deadline: Sunday, May 31

What’s robustly good to do now, under deep uncertainty?

What to do under different AI takeoff scenarios?

How can AI safety and welfare work support each other?

How might AI welfare shape deal-making with AIs?

Should AIs have legal rights, and if so, which?

How will AI-AI interactions shape the welfare of digital minds?

What would harmonious coexistence look like?

How can we influence those who will shape the welfare of digital minds?

Is restricting the creation of digital minds feasible?

Who will deliberately create digital minds, and why?

How will digital minds spread to space?

How can we make AIs value the welfare of digital minds?

Do different types of digital minds require different strategies?

Will digital minds be happy by default?

What preferences will digital minds have, and what follows?

What interest groups and coalitions will form around digital minds?

How will religions respond?

What crucial considerations are we missing?