Author name: Kris Guyer

Congress warned that NASA’s current plan for Artemis “cannot work”

artemis, congress, griffin, NASA, Space, spacex, starship / Kris Guyer / December 4, 2025

As for what to do about it, Griffin said legislators should end the present plan.

“The Artemis III mission and those beyond should be canceled and we should start over, proceeding with all deliberate speed,” Griffin said. He included a link to his plan, which is not dissimilar from the “Apollo on Steroids” architecture he championed two decades ago, but was later found to be unaffordable within NASA’s existing budget.

“There need to be consequences”

Other panel members offered more general advice.

Clayton Swope, deputy director of the Aerospace Security Project for the Center for Strategic and International Studies, said NASA should continue to serve as an engine for US success in space and science. He cited the Commercial Lunar Payload Services program, which has stimulated a growing lunar industry. He also said NASA spending on basic research and development is a critical feedstock for US innovation, and a key advantage over the People’s Republic of China.

“When you’re looking at the NASA authorization legislation, look at it in a way where you are the genesis of that innovation ecosystem, that flywheel that really powers US national security and economic security, in a way that the PRC just can’t match,” Swope said. “Without science, we would never have had something like the Manhattan Project.”

Another witness, Dean Cheng of the Potomac Institute for Policy Studies, said NASA—and by extension Congress—must do a better job of holding itself and its contractors accountable.

Many of NASA’s major exploration programs, including the Orion spacecraft, Space Launch System rocket, and their ground systems, have run years behind schedule and billions of dollars over budget in the last 15 years. NASA has funded these programs with cost-plus contracts, so it has had limited ability to enforce deadlines with contractors. Moreover, Congress has more or less meekly gone along with the delays and continued funding the programs.

Cheng said that whatever priorities policymakers decide for NASA, failing to achieve objectives should come with consequences.

“One, it needs to be bipartisan, to make very clear throughout our system that this is something that everyone is pushing for,” Cheng said of establishing priorities for NASA. “And two, that there are consequences, budgetary, legal, and otherwise, to the agency, to supplying companies. If they fail to deliver on time and on budget, that it will not be a ‘Well, okay, let’s try again next year.’ There need to be consequences.”

Congress warned that NASA’s current plan for Artemis “cannot work” Read More »

Why won’t Steam Machine support HDMI 2.1? Digging in on the display standard drama.

AMD, drivers, gaming, steam, steam machine, steamos, valve, Windows / Kris Guyer / December 4, 2025

When Valve announced its upcoming Steam Machine hardware last month, some eagle-eyed gamers may have been surprised to see that the official spec sheet lists support for HDMI 2.0 output, rather than the updated, higher-bandwidth HDMI 2.1 standard introduced in 2017. Now, Valve tells Ars that, while the hardware itself actually supports HDMI 2.1, the company is struggling to offer full support for that standard due to Linux drivers that are “still a work-in-progress on the software side.”

As we noted last year, the HDMI Forum (which manages the official specifications for HDMI standards) has officially blocked any open source implementation of HDMI 2.1. That means the open source AMD drivers used by SteamOS can’t fully implement certain features that are specific to the updated output standard.

“At this time an open source HDMI 2.1 implementation is not possible without running afoul of the HDMI Forum requirements,” AMD engineer Alex Deucher said at the time.

Doing what they can

This situation has caused significant headaches for Valve, which tells Ars it has had to validate the Steam Machine’s HDMI 2.1 hardware via Windows during testing. And when it comes to HDMI performance via SteamOS, a Valve representative tells Ars that “we’ve been working on trying to unblock things there.”

That includes unblocking HDMI 2.0’s resolution and frame-rate limits, which max out at 60 Hz for a 4K output, according to the official standard. Valve tells Ars it has been able to increase that limit to the “4K @ 120Hz” listed on the Steam Machine spec sheet, though, thanks to a technique called chroma sub-sampling.

Why won’t Steam Machine support HDMI 2.1? Digging in on the display standard drama. Read More »

Microsoft drops AI sales targets in half after salespeople miss their quotas

agentic AI, AI, AI adoption, AI agents, AI infrastructure, AI spending, azure, Biz & IT, Cloud Computing, copilot, enterprise AI, foundry, generative ai, machine learning, microsoft, Microsoft Copilot, simulated reasoning, SR models / Kris Guyer / December 3, 2025

Microsoft has lowered sales growth targets for its AI agent products after many salespeople missed their quotas in the fiscal year ending in June, according to a report Wednesday from The Information. The adjustment is reportedly unusual for Microsoft, and it comes after the company missed a number of ambitious sales goals for its AI offerings.

AI agents are specialized implementations of AI language models designed to perform multistep tasks autonomously rather than simply responding to single prompts. So-called “agentic” features have been central to Microsoft’s 2025 sales pitch: At its Build conference in May, the company declared that it has entered “the era of AI agents.”

The company has promised customers that agents could automate complex tasks, such as generating dashboards from sales data or writing customer reports. At its Ignite conference in November, Microsoft announced new features like Word, Excel, and PowerPoint agents in Microsoft 365 Copilot, along with tools for building and deploying agents through Azure AI Foundry and Copilot Studio. But as the year draws to a close, that promise has proven harder to deliver than the company expected.

According to The Information, one US Azure sales unit set quotas for salespeople to increase customer spending on a product called Foundry, which helps customers develop AI applications, by 50 percent. Less than a fifth of salespeople in that unit met their Foundry sales growth targets. In July, Microsoft lowered those targets to roughly 25 percent growth for the current fiscal year. In another US Azure unit, most salespeople failed to meet an earlier quota to double Foundry sales, and Microsoft cut their quotas to 50 percent for the current fiscal year.

Microsoft drops AI sales targets in half after salespeople miss their quotas Read More »

On Dwarkesh Patel’s Second Interview With Ilya Sutskever

Dwarkesh / Kris Guyer / December 3, 2025

Some podcasts are self-recommending on the ‘yep, I’m going to be breaking this one down’ level. This was very clearly one of those. So here we go.

As usual for podcast posts, the baseline bullet points describe key points made, and then the nested statements are my commentary.

If I am quoting directly I use quote marks, otherwise assume paraphrases.

What are the main takeaways?

Ilya thinks training in its current form will peter out, that we are returning to an age of research where progress requires more substantially new ideas.
SSI is a research organization. It tries various things. Not having a product lets it punch well above its fundraising weight in compute and effective resources.
Ilya has 5-20 year timelines to a potentially superintelligent learning model.
SSI might release a product first after all, but probably not?
Ilya’s thinking about alignment still seems relatively shallow to me in key ways, but he grasps many important insights and understands he has a problem.
Ilya essentially despairs of having a substantive plan beyond ‘show everyone the thing as early and often as possible’ and hope for the best. He doesn’t know where to go or how to get there, but does realize he doesn’t know these things, so he’s well ahead of most others.

Afterwards, this post also covers Dwarkesh Patel’s post on the state of AI progress.

Ilya opens by remarking how crazy it is all this (as in AI) is real, it’s all so sci-fi, and yet it’s not felt in other ways so far. Dwarkesh expects this to continue for average people into the singularity, Ilya says no, AI will diffuse and be felt in the economy. Dwarkesh says impact seems smaller than model intelligence implies.
1. Ilya is right here. Dwarkesh is right that direct impact so far has been smaller than model intelligence implies, but give it time.
Ilya says, the models are really good at evals but economic impact lags. The models are buggy, and choices for RL take inspiration from the evals, so the evals are misleading and the humans are essentially reward hacking the evals. And that given they got their scores by studying for tons of hours rather than via intuition, one should expect AIs to underperform their benchmarks.
1. AIs definitely underperform their benchmarks in terms of general usefulness, even for those companies that do minimal targeting of benchmarks. Overall capabilities lag behind, for various reasons. We still have an impact gap.
The super talented student? The one that hardly even needs to practice a specific task to be good? They’ve got ‘it.’ Models don’t have ‘it.’
1. If anything, models have ‘anti-it.’ They make it up on volume. Sure.

Humans train on much less data, but what they know they know ‘more deeply’ somehow, there are mistakes we wouldn’t make. Also evolution can be highly robust, for example the famous case where a guy lost all his emotions and in many ways things remained fine.
1. People put a lot of emphasis on the ‘I would never’ heuristic, as AIs will sometimes do things ‘a similarly smart person’ would never do, they lack a kind of common sense.
So what is the ‘ML analogy for emotions’? Ilya says some kind of value function thing, as in the thing that tells you if you’re doing well versus badly while doing something.
1. Emotions as value functions makes sense, but they are more information-dense than merely a scalar, and can often point you to things you missed. They do also serve as training reward signals.
2. I don’t think you ‘need’ emotions for anything other than signaling emotions, if you are otherwise sufficiently aware in context, and don’t need them to do gradient descent.
3. However in a human, if you knock out the emotions in places where you were otherwise relying on them for information or to resolve uncertainty, you’re going to have a big problem.
4. I notice an obvious thing to try but it isn’t obvious how to implement it?
Ilya has faith in deep learning. There’s nothing it can’t do!

Data? Parameters? Compute? What else? It’s easier and more reliable to scale up pretraining than to figure out what else to do. But we’ll run out of data soon even if Gemini 3 got more out of this, so now you need to do something else. If you had 100x more scale here would anything be that different? Ilya thinks no.
1. Sounds like a skill issue, on some level, but yes if you didn’t change anything else then I expect scaling up pretraining further won’t help enough to justify the increased costs in compute and time.
RL costs now exceed pretraining costs, because each RL run costs a lot. It’s time to get back to an age of research, trying interesting things and seeing what happens.
1. I notice I am skeptical of the level of skepticism, also I doubt the research mode ever stopped in the background. The progress will continue. It’s weird how every time someone says ‘we still need some new idea or breakthrough’ there is the implication that this likely never happens again.

Why do AIs require so much more data than humans to learn? Why don’t models easily pick up on all this stuff humans learn one-shot or in the background?
1. Humans have richer data than text so the ratio is not as bad as it looks, but primarily because our AI learning techniques are relatively primitive and data inefficient in various ways.
2. My full answer to how to fix it falls under ‘I don’t do $100m/year jobs for free.’
3. Also there are ways in which the LLMs learn way better than you realize, and a lot of the tasks humans easily learn are regularized in non-obvious ways.
Ilya believes humans being good at learning is mostly not part of some complicated prior, and people’s robustness is really staggering.
1. I would clarify, not part of a complicated specialized prior. There is also a complicated specialized prior in some key domains, but that is in addition to a very strong learning function.
2. People are not as robust as Ilya thinks, or most people think.
Ilya suggests perhaps human neurons use more compute than we think.

Scaling ‘sucked the air out of the room’ so no one did anything else. Now there are more companies than ideas. You need some compute to bring ideas to life, but not the largest amounts.
1. You can also think about some potential techniques as ‘this is not worth trying unless you have massive scale.’
SSI’s compute all goes into research, none into inference, and they don’t try to build a product, and if you’re doing something different you don’t have to use maximum scale, so their $3 billion that they’ve raised ‘goes a long way’ relative to the competition. Sure OpenAI spends ~$5 billion a year on experiments, but it’s what you do with it.
1. This is what Ilya has to say in this spot, but there’s merit in it. OpenAI’s experiments are largely about building products now. This transfers to the quest for superintelligence, but not super efficiently.
How will SSI make money? Focus on the research, the money will appear.
1. Matt Levine has answered this one, which is that you make money by being an AI company full of talented researchers, so people give you money.
SSI is considering making a product anyway, both to have the product exist and also because timelines might be long.
1. I mean I guess at some point the ‘we are AI researchers give us money’ strategy starts to look a little suspicious, but let’s not rush into anything.
2. Remember, Ilya, once you have a product and try to have revenue they’ll evaluate the product and your revenue. If you don’t have one, you’re safe.

Ilya says even if there is a straight shot to superintelligence deployment would be gradual, you have to ship something first, and that he agrees with Dwarkesh on the importance of continual learning, it would ‘go and be’ various things and learn, superintelligence is not a finished mind.
1. Learning takes many forms, including continual learning, it can be updating within the mind or otherwise, and so on. See previous podcast discussions.
Ilya expects ‘rapid’ economic growth, perhaps ‘very rapid.’ It will vary based on what rules are set in different places.
1. Rapid means different things to different people, it sounds like Ilya doesn’t have a fixed rate in mind. I interpret it as ‘more than these 2% jokers.’
2. This vision still seems to think the humans stay in charge. Why?

Dwarkesh reprises the standard point that if AIs are merely ‘as good at’ humans at learning, but they can ‘merge brains’ then crazy things happen. How do we make such a situation go well? What is SSI’s plan?
1. I mean, that’s the least of it, but hopefully yes that suffices to make the point?
Ilya emphasizes deploying incrementally and in advance. It’s hard to predict what this will be like in advance. “The problem is the power. When the power is really big, what’s going to happen? If it’s hard to imagine, what do you do? You’ve got to be showing the thing.”
1. This feels like defeatism, in terms of saying we can only respond to things once we can see and appreciate them. We can’t plan for being old until we know what that’s like. We can’t plan for AGI/ASI, or AI having a lot of power, until we can see that in action.
2. But obviously by then it is likely to be too late, and most of your ability to steer what happens has already been lost, perhaps all of it.
3. This is the strategy of ‘muddle through’ the same as we always muddle through, basically the plan of not having a plan other than incrementalism. I do not care for this plan. I am not happy to be a part of it. I do not think that is a case of Safe Superintelligence.
Ilya expects governments and labs to play big roles, and for labs to increasingly coordinate on safety, as Anthropic and OpenAI did in a recent first step. And we have to figure out what we should be building. He suggests making the AI care about sentient life in general will be ‘easier’ than making it care about humans, since the AI will be sentient.
1. If the AIs do not care about humans in particular, there is no reason to expect humans to stay in control or to long endure.
Ilya would like the most powerful superintelligence to ‘somehow’ be ‘capped’ to address these concerns. But he doesn’t know how to do that.
1. I don’t know how to do that either. It’s not clear the idea is coherent.
Dwarkesh asks how much ‘room is there at the top’ for superintelligence to be more super? Maybe it just learns fast or has a bigger pool of strategies or skills or knowledge? Ilya says very powerful, for sure.
1. Sigh. There is very obviously quite a lot of ‘room at the top’ and humans are not anything close to maximally intelligent, nor to getting most of what intelligence has to offer. At this point, the number of people who still don’t realize or accept this reinforces how much better a smarter entity could be.
Ilya expects these superintelligences to be very large, as in physically large, and for several to come into being at roughly the same time, and ideally they could “be restrained in some ways or if there was some kind of agreement or something.”
1. That agreement between AIs would then be unlikely to include us. Yes, functional restraints would be nice, but this is the level of thought that has gone into finding ways to do it.
2. There’s been a lot of things staying remarkably close, but a lot of that is because rather than an edge compounding and accelerating for now catching up has been easier.
Ilya: “What is the concern of superintelligence? What is one way to explain the concern? If you imagine a system that is sufficiently powerful, really sufficiently powerful—and you could say you need to do something sensible like care for sentient life in a very single-minded way—we might not like the results. That’s really what it is.”
1. Well, yes, standard Yudkowsky, no fixed goal we can name turns out well.
Ilya says maybe we don’t build an RL agent. Humans are semi-RL agents, our emotions make us alter our rewards and pursue different rewards after a while. If we keep doing what we are doing now it will soon peter out and never be “it.”
1. There’s a baked in level of finding innovations and improvements that should be in anyone’s ‘keep doing what we are doing’ prior, and I think it gets us pretty far and includes many individually low-probability-of-working innovations making substantial differences. There is some level on which we would ‘peter out’ without a surprise, but it’s not clear that this requires being surprised overall.
2. Is it possible things do peter out and we never see ‘it’? Yeah. It’s possible. I think it’s a large underdog to stay that way for long, but it’s possible. Still a long practical way to go even then.
3. Emotions, especially boredom and the fading of positive emotions on repetition, are indeed one of the ways we push ourselves towards exploration and variety. That’s one of many things they do, and yes if we didn’t have them then we would need something else to take their place.
4. In many cases I have indeed used logic to take the place of that, when emotion seems to not be sufficiently preventing mode collapse.
“One of the things that you could say about what causes alignment to be difficult is that your ability to learn human values is fragile. Then your ability to optimize them is fragile. You actually learn to optimize them. And can’t you say, “Are these not all instances of unreliable generalization?” Why is it that human beings appear to generalize so much better? What if generalization was much better? What would happen in this case? What would be the effect? But those questions are right now still unanswerable.”
1. It is cool to hear Ilya restate these Yudkowsky 101 things.
2. Humans do not actually generalize all that well.
How does one think about what AI going well looks like? Ilya goes back to ‘AI that cares for sentient life’ as a first step, but then asks the better question, what is the long run equilibrium? He notices he does not like his answer. Maybe each person has an AI that will do their bidding and that’s good, but the downside is then the AI does things like earn money or advocate or whatever, and the person says ‘keep it up’ but they’re not a participant. Precarious. People become part AI, Neurolink++. He doesn’t like this solution, but it is at least a solution.
1. Big points for acknowledging that there are no known great solutions.
2. Big points for pointing out one big flaw, that the people stop actually doing the things, because the AIs do the things better.
3. The equilibrium here is that increasingly more things are turned over to AIs, including both actions and decisions. Those who don’t do this fall behind.
4. The equilibrium here is that increasingly AIs are given more autonomy, more control, put in better positions, have increasing power and wealth shares, and so on, even if everything involved is fully voluntary and ‘nothing goes wrong.’
5. Neurolink++ does not meaningfully solve any of the problems here.
6. Solve for the equilibrium.
Is the long history of emotions an alignment success? As in, it allows the brain to move from ‘mate with somebody who’s more successful’ into flexibly defining success and generally adjusting to new situations.
1. It’s a highly mixed bag, wouldn’t you say?
2. There are ways in which those emotions have been flexible and adaptable and a success, and have succeeded in the alignment target (inclusive genetic fitness) and also ways in which emotions are very obviously failing people.
3. If ASIs are about as aligned as we are in this sense, we’re doomed.
Ilya says it’s mysterious how evolution encodes high-level desires, but it gives us all these social desires, and they evolved pretty recently. Dwarkesh points out it is desire you learned in your lifetime. Ilya notes the brain as regions and some things are hardcoded, but if you remove half the brain then the regions move, the social stuff is highly reliable.
1. I don’t pretend to understand the details here, although I could speculate.

SSI investigates ideas to see if they are promising. They do research.
On his cofounder leaving: “For this, I will simply remind a few facts that may have been forgotten. I think these facts which provide the context explain the situation. The context was that we were fundraising at a $32 billion valuation, and then Meta came in and offered to acquire us, and I said no. But my former cofounder in some sense said yes. As a result, he also was able to enjoy a lot of near-term liquidity, and he was the only person from SSI to join Meta.”
1. I love the way he put that. Yes.
“The main thing that distinguishes SSI is its technical approach. We have a different technical approach that I think is worthy and we are pursuing it. I maintain that in the end there will be a convergence of strategies. I think there will be a convergence of strategies where at some point, as AI becomes more powerful, it’s going to become more or less clearer to everyone what the strategy should be. It should be something like, you need to find some way to talk to each other and you want your first actual real superintelligent AI to be aligned and somehow care for sentient life, care for people, democratic, one of those, some combination thereof. I think this is the condition that everyone should strive for. That’s what SSI is striving for. I think that this time, if not already, all the other companies will realize that they’re striving towards the same thing. We’ll see. I think that the world will truly change as AI becomes more powerful. I think things will be really different and people will be acting really differently.”
1. This is a remarkably shallow, to me, vision of what the alignment part of the strategy looks like, but it does get an admirably large percentage of the overall strategic vision, as in most of it?
2. The idea that ‘oh as we move farther along people will get more responsible and cooperate more’ seems to not match what we have observed so far, alas.
3. Ilya later clarifies he specifically meant convergence on alignment strategies, although he also expects convergence on technical strategies.
4. The above statement is convergence on an alignment goal, but that doesn’t imply convergence on alignment strategy. Indeed it does not imply that an alignment strategy that is workable even exists.
Ilya’s timeline to the system that can learn and become superhuman? 5-20 years.
Ilya predicts that when someone releases the thing that will be information but it won’t teach others how to do the thing, although they will eventually learn.
What is the ‘good world’? We have powerful human-like learners and perhaps narrow ASIs, and companies make money, and there is competition through specialization, different niches. Accumulated learning and investment creates specialization.
1. This is so frustrating, in that it doesn’t explain why you would expect that to be how this plays out, or why this world turns out well, or anything really? Which would be fine if the answers were clear or at least those seemed likely, but I very much don’t think that.
2. This feels like a claim that humans are indeed near the upper limit of what intelligence can do and what can be learned except that we are hobbled in various ways and AIs can be unhobbled, but that still leaves them functioning in ways that seem recognizably human and that don’t crowd us out? Except again I don’t think we should expect this.
Dwarkesh points out current LLMs are similar, Ilya says perhaps the datasets are not as non-overlapping as they seem.
1. On the contrary, I was assuming they were mostly the same baseline data, and then they do different filtering and progressions from there? Not that there’s zero unique data but that most companies have ‘most of the data.’
Dwarkesh suggests, therefore AIs will have less diversity than human teams. How can we get ‘meaningful diversity’? Ilya says this is because of pretraining, that post training is different.
1. To the extent that such ‘diversity’ is useful it seems easy to get with effort. I suspect this is mostly another way to create human copium.
What about using self-play? Ilya notes it allows using only compute, which is very interesting, but it is only good for ‘developing a certain set of skills.’ Negotiation, conflict, certain social strategies, strategizing, that kind of stuff. Then Ilya self-corrects, notes other forms, like debate, prover-verifier or forms of LLM-as-a-judge, it’s a special case of agent competition.
1. I think there’s a lot of promising unexplored space here, decline to say more.

What is research taste? How does Ilya come up with many big ideas?

This is hard to excerpt and seems important, so quoting in full to close out:

I can comment on this for myself. I think different people do it differently. One thing that guides me personally is an aesthetic of how AI should be, by thinking about how people are, but thinking correctly. It’s very easy to think about how people are incorrectly, but what does it mean to think about people correctly?

I’ll give you some examples. The idea of the artificial neuron is directly inspired by the brain, and it’s a great idea. Why? Because you say the brain has all these different organs, it has the folds, but the folds probably don’t matter. Why do we think that the neurons matter? Because there are many of them. It kind of feels right, so you want the neuron. You want some local learning rule that will change the connections between the neurons. It feels plausible that the brain does it.

The idea of the distributed representation. The idea that the brain responds to experience therefore our neural net should learn from experience. The brain learns from experience, the neural net should learn from experience. You kind of ask yourself, is something fundamental or not fundamental? How things should be.

I think that’s been guiding me a fair bit, thinking from multiple angles and looking for almost beauty, beauty and simplicity. Ugliness, there’s no room for ugliness. It’s beauty, simplicity, elegance, correct inspiration from the brain. All of those things need to be present at the same time. The more they are present, the more confident you can be in a top-down belief.

The top-down belief is the thing that sustains you when the experiments contradict you. Because if you trust the data all the time, well sometimes you can be doing the correct thing but there’s a bug. But you don’t know that there is a bug. How can you tell that there is a bug? How do you know if you should keep debugging or you conclude it’s the wrong direction? It’s the top-down. You can say things have to be this way. Something like this has to work, therefore we’ve got to keep going. That’s the top-down, and it’s based on this multifaceted beauty and inspiration by the brain.

I need to think more about what causes my version of ‘research taste.’ It’s definitely substantially different.

That ends our podcast coverage, and enter the bonus section, which seems better here than in the weekly, as it covers many of the same themes.

Dwarkesh Patel offers his thoughts on AI progress these days, noticing that when we get the thing he calls ‘actual AGI’ things are going to get fucking crazy, but thinking that this is 10-20 years away from happening in full. Until then, he’s a bit skeptical of how many gains we can realize, but skepticism is highly relative here.

Dwarkesh Patel: I’m confused why some people have short timelines and at the same time are bullish on RLVR. If we’re actually close to a human-like learner, this whole approach is doomed.

… Either these models will soon learn on the job in a self directed way – making all this pre-baking pointless – or they won’t – which means AGI is not imminent. Humans don’t have to go through a special training phase where they need to rehearse every single piece of software they might ever use.

Wow, look at those goalposts move (in all the different directions). Dwarkesh notes that the bears keep shifting on the bulls, but says this is justified because current models fit the old goals but don’t score the points, as in they don’t automate workflows as much as you would expect.

In general, I worry about the expectation pattern having taken the form of ‘median 50 years → 20 → 10 → 5 → 7, and once I heard someone said 3, so oh nothing to see there you can stop worrying.’

In this case, look at the shift: An ‘actual’ (his term) AGI must now not only be capable of human-like performance of tasks, the AGI must also be a human-efficient learner.

That would mean AGI and ASI are the same thing, or at least arrive in rapid succession. An AI that was human-efficient at learning from data, combined with AI’s other advantages that include imbibing orders of magnitude more data, would be a superintelligence and would absolutely set off recursive self-improvement from there.

And yes, if that’s what you mean then AGI isn’t the best concept for thinking about timelines, and superintelligence is the better target to talk about. Sriram Krishnan is however opposed to using either of them.

Like all conceptual handles or fake frameworks, it is imprecise and overloaded, but people’s intuitions about it miss that the thing is possible or exists even when you outright say ‘superintelligence’ and I shudder to think how badly they will miss the concept if you don’t even say it. Which I think is a lot of the motivation behind not wanting to say it, so people can pretend that there won’t be things smarter than us in any meaningful sense and thus we can stop worrying about it or planning for it.

Indeed, this is exactly Sriram’s agenda if you look at his post here, to claim ‘we are not on the timeline’ that involves such things, to dismiss concerns as ‘sci-fi’ or philosophical, and talk instead of ‘what we are trying to build.’ What matters is what actually gets built, not what we intended, and no none of these concepts have been invalidated. We have ‘no proof of takeoff’ in the sense that we are not currently in a fast takeoff yet, but what would constitute this ‘proof’ other than already being in a takeoff, and thus it being too late to do anything about it?

Sriram Krishnan: …most importantly, it invokes fear—connected to historical usage in sci-fi and philosophy (think 2001, Her, anything invoking the singularity) that has nothing to do with the tech tree we’re actually on. Makes every AI discussion incredibly easy to anthropomorphize and detour into hypotheticals.

Joshua Achiam (OpenAI Head of Mission Alignment): I mostly disagree but I think this is a good contribution to the discourse. Where I disagree: I do think AGI and ASI both capture something real about where things are going. Where I agree: the lack of agreed-upon definitions has 100% created many needless challenges.

The idea that ‘hypotheticals,’ as in future capabilities and their logical consequences, are ‘detours,’ or that any such things are ‘sci-fi or philosophy’ is to deny the very idea of planning for future capabilities or thinking about the future in real ways. Sriram himself only thinks they are 10 years away, and then the difference is he doesn’t add Dwarkesh’s ‘and that’s fucking crazy’ and instead seems to effectively say ‘and that’s a problem for future people, ignore it.’

Seán Ó hÉigeartaigh: I keep noting this, but I do think a lot of the most heated policy debates we’re having are underpinned by a disagreement on scientific view: whether we (i) are on track in coming decade for something in the AGI/ASI space that can achieve scientific feats equivalent to discovering general relativity (Hassabis’ example), or (ii) should expect AI as a normal technology (Narayanan & Kapoor’s definition).

I honestly don’t know. But it feels premature to me to rule out (i) on the basis of (slightly) lengthening timelines from the believers, when progress is clearly continuing and a historically unprecedented level of resources are going into the pursuit of it. And premature to make policy on the strong expectation of (ii). (I also think it would be premature to make policy on the strong expectation of (i) ).

But we are coming into the time where policy centred around worldview (ii) will come into tension in various places with the policies worldview (i) advocates would enact if given a free hand. Over the coming decade I hope we can find a way to navigate a path between, rather than swing dramatically based on which worldview is in the ascendancy at a given time.

Sriram Krishnan: There is truth to this.

This paints it as two views, and I would say you need at least three:

Something in the AGI/ASI space is likely in less than 10 years.
Something in the AGI/ASI space is unlikely in less than about 10 years, but highly plausible in 10-20 years, until then AI is a normal technology.
AI is a normal technology and we know it will remain so indefinitely. We can regulate and plan as if AGI/ASI style technologies will never happen.

I think #1 and #2 are both highly reasonable positions, only #3 is unreasonable, while noting that if you believe #2 you still need to put some non-trivial weight on #1. As in, if you think it probably takes ~10 years then you can perhaps all but rule out AGI 2027, and you think 2031 is unlikely, but you cannot claim 2031 is a Can’t Happen.

The conflation to watch out for is #2 and #3. These are very different positions. Yet many in the AI industry, and its political advocates, make exactly this conflation. They assert ‘#1 is incorrect therefore #3,’ when challenged for details articulate claim #2, then go back to trying to claim #3 and act on the basis of #3.

What’s craziest is that the list of things to rule out, chosen by Sriram, includes the movie Her. Her made many very good predictions. Her was a key inspiration for ChatGPT and its voice mode, so much so that there was a threatened lawsuit because they all but copied Scarlett Johansson’s voice. She’s happening. Best be believing in sci-fi stores, because you’re living in one, and all that.

Nothing about current technology is a reason to think 2001-style things or a singularity will not happen, or to think we should anthropomorphize AI relatively less (the correct amount for current AIs, and for future AIs, are both importantly not zero, and importantly not 100%, and both mistakes are frequently made). Indeed, Dwarkesh is de facto predicting a takeoff and a singularity in this post that Sriram praised, except Dwarkesh has it on a 10-20 year timescale to get started.

Now, back to Dwarkesh.

This process of ‘teach the AI the specific tasks people most want’ is the central instance of models being what Teortaxes calls usemaxxed. A lot of effort is going to specific improvements rather than to advancing general intelligence. And yes, this is evidence against extremely short timelines. It is also, as Dwarkesh notes, evidence in favor of large amounts of mundane utility soon, including ability to accelerate R&D. What else would justify such massive ‘side’ efforts?

There’s also, as he notes, the efficiency argument. Skills many people want should be baked into the core model. Dwarkesh fires back that there are a lot of skills that are instance-specific and require on-the-job or continual learning, which he’s been emphasizing a lot for a while. I continue to not see a contradiction, or why it would be that hard to store and make available that knowledge as needed even if it’s hard for the LLM to permanently learn it.

I strongly disagree with his claim that ‘economic diffusion lag is cope for missing capabilities.’ I agree that many highly valuable capabilities are missing. Some of them are missing due to lack of proper scaffolding or diffusion or context, and are fundamentally Skill Issues by the humans. Others are foundational shortcomings. But the idea that the AIs aren’t up to vastly more tasks than they’re currently asked to do seems obviously wrong?

He quotes Steven Byrnes:

Steven Byrnes: New technologies take a long time to integrate into the economy? Well ask yourself: how do highly-skilled, experienced, and entrepreneurial immigrant humans manage to integrate into the economy immediately? Once you’ve answered that question, note that AGI will be able to do those things too.

Again, this is saying that AGI will be as strong as humans in the exact place it is currently weakest, and will not require adjustments for us to take advantage. No, it is saying more than that, it is also saying we won’t put various regulatory and legal and cultural barriers in its way, either, not in any way that counts.

If the AGI Dwarkesh is thinking about were to exist, again, it would be an ASI, and it would be all over for the humans very quickly.

I also strongly disagree with human labor not being ‘shleppy to train’ (bonus points, however, for excellent use of ‘shleppy’). I have trained humans and been a human being trained, and it is totally shleppy. I agree, not as schleppy as current AIs can be when something is out of their wheelhouse, but rather obnoxiously schleppy everywhere except their own very narrow wheelhouse.

Here’s another example of ‘oh my lord check out those goalposts’:

Dwarkesh Patel: It revealed a key crux between me and the people who expect transformative economic impacts in the next few years.

Transformative economic impacts in the next few years would be a hell of a thing.

It’s not net-productive to build a custom training pipeline to identify what macrophages look like given the way this particular lab prepares slides, then another for the next lab-specific micro-task, and so on. What you actually need is an AI that can learn from semantic feedback on the job and immediately generalize, the way a human does.

Well, no, it probably isn’t now, but also Claude Code is getting rather excellent at creating training pipelines, and the whole thing is rather standard in that sense, so I’m not convinced we are that far away from doing exactly that. This is an example of how sufficient ‘AI R&D’ automation, even on a small non-recursive scale, can transform use cases.

Every day, you have to do a hundred things that require judgment, situational awareness, and skills & context learned on the job. These tasks differ not just across different people, but from one day to the next even for the same person. It is not possible to automate even a single job by just baking in some predefined set of skills, let alone all the jobs.

Well, I mean of course it is, for a sufficiently broad set of skills at a sufficiently high level, especially if this includes meta-skills and you can access additional context. Why wouldn’t it be? It certainly can quickly automate large portions of many jobs, and yes I have started to automate portions of my job indirectly (as in Claude writes me the mostly non-AI tools to do it, and adjusts them every time they do something wrong).

Give it a few more years, though, and Dwarkesh is on the same page as I am:

In fact, I think people are really underestimating how big a deal actual AGI will be because they’re just imagining more of this current regime. They’re not thinking about billions of human-like intelligences on a server which can copy and merge all their learnings. And to be clear, I expect this (aka actual AGI) in the next decade or two. That’s fucking crazy!

Exactly. This ‘actual AGI’ is fucking crazy, and his timeline for getting there of 10-20 years is also fucking crazy. More people need to add ‘and that’s fucking crazy’ at the end of such statements.

Dwarkesh then talks more about continual learning. His position here hasn’t changed, and neither has my reaction that this isn’t needed, we can get the benefits other ways. He says that the gradual progress on continual learning means it won’t be ‘game set match’ to the first mover, but if this is the final piece of the puzzle then why wouldn’t it be?

Discussion about this post

On Dwarkesh Patel’s Second Interview With Ilya Sutskever Read More »

OpenAI CEO declares “code red” as Gemini gains 200 million users in 3 months

AI, AI assistants, AI competition, Biz & IT, chatbots, chatgpt, Gemini, Gemini 3, Google, Google Gemini, GPT-5, large language models, machine learning, Marc Benioff, openai, sam altman, simulated reasoning, SR model, Sundar Pichai, vibemark, vibemarking / Kris Guyer / December 2, 2025

In addition to buzz about Gemini on social media, Google is quickly catching up to ChatGPT in user numbers. ChatGPT has more than 800 million weekly users, according to OpenAI, while Google’s Gemini app has grown from 450 million monthly active users in July to 650 million in October, according to Business Insider.

Financial stakes run high

Not everyone views OpenAI’s “code red” as a genuine alarm. Reuters columnist Robert Cyran wrote on Tuesday that OpenAI’s announcement added “to the impression that OpenAI is trying to do too much at once with technology that still requires a great deal of development and funding.” On the same day Altman’s memo circulated, OpenAI announced an ownership stake in a Thrive Capital venture and a collaboration with Accenture. “The only thing bigger than the company’s attention deficit is its appetite for capital,” Cyran wrote.

In fact, OpenAI faces an unusual competitive disadvantage: Unlike Google, which subsidizes its AI ventures through search advertising revenue, OpenAI does not turn a profit and relies on fundraising to survive. According to The Information, the company, now valued at around $500 billion, has committed more than $1 trillion in financial obligations to cloud computing providers and chipmakers that supply the computing power needed to train and run its AI models.

But the tech industry never stands still, and things can change quickly. Altman’s memo also reportedly stated that OpenAI plans to release a new simulated reasoning model next week that may beat Gemini 3 in internal evaluations. In AI, the back-and-forth cycle of one-upmanship is expected to continue as long as the dollars keep flowing.

OpenAI CEO declares “code red” as Gemini gains 200 million users in 3 months Read More »

India orders device makers to put government-run security app on all phones

Apple, IPhone, Policy / Kris Guyer / December 2, 2025

Consumers can also use the app or website to check the number of mobile connections in their name and report any that appear to be fraudulent.

Priyanka Gandhi of the Congress Party, a member of Parliament, said that Sanchar Saathi “is a snooping app… It’s a very fine line between ‘fraud is easy to report’ and ‘we can see everything that every citizen of India is doing on their phone.’” She called for an effective system to fight fraud, but said that cybersecurity shouldn’t be “an excuse to go into every citizen’s telephone.”

App may need “root level access”

Despite Scindia saying the app can be deleted by users, the government statement that phone makers must ensure its functionalities are not “disabled or restricted” raised concerns about the level of access it requires. While the app store version can be deleted, privacy advocates say the order’s text indicates the pre-installed version would require deeper integration into the device.

The Internet Freedom Foundation, an Indian digital rights advocacy group, said the government directive “converts every smartphone sold in India into a vessel for state mandated software that the user cannot meaningfully refuse, control, or remove. For this to work in practice, the app will almost certainly need system level or root level access, similar to carrier or OEM system apps, so that it cannot be disabled. That design choice erodes the protections that normally prevent one app from peering into the data of others, and turns Sanchar Saathi into a permanent, non-consensual point of access sitting inside the operating system of every Indian smartphone user.”

The group said that while the app is being “framed as a benign IMEI checker,” a server-side update could repurpose it to perform “client side scanning for ‘banned’ applications, flag VPN usage, correlate SIM activity, or trawl SMS logs in the name of fraud detection. Nothing in the order constrains these possibilities.”

India orders device makers to put government-run security app on all phones Read More »

Testing shows why the Steam Machine’s 8GB of graphics RAM could be a problem

AMD, gaming, intel, NVIDIA, steam deck, steam machine, steamos, Tech / Kris Guyer / December 2, 2025

By Valve’s admission, its upcoming Steam Machine desktop isn’t swinging for the fences with its graphical performance. The specs promise decent 1080p-to-1440p performance in most games, with 4K occasionally reachable with assistance from FSR upscaling—about what you’d expect from a box with a modern midrange graphics card in it.

But there’s one spec that has caused some concern among Ars staffers and others with their eyes on the Steam Machine: The GPU comes with just 8GB of dedicated graphics RAM, an amount that is steadily becoming more of a bottleneck for midrange GPUs like AMD’s Radeon RX 7060 and 9060, or Nvidia’s GeForce RTX 4060 or 5060.

In our reviews of these GPUs, we’ve already run into some games where the RAM ceiling limits performance in Windows, especially at 1440p. But we’ve been doing more extensive testing of various GPUs with SteamOS, and we can confirm that in current betas, 8GB GPUs struggle even more on SteamOS than they do running the same games at the same settings in Windows 11.

The good news is that Valve is working on solutions, and having a stable platform like the Steam Machine to aim for should help improve things for other hardware with similar configurations. The bad news is there’s plenty of work left to do.

The numbers

We’ve tested an array of dedicated and integrated Radeon GPUs under SteamOS and Windows, and we’ll share more extensive results in another article soon (along with broader SteamOS-vs-Windows observations). But for our purposes here, the two GPUs that highlight the issues most effectively are the 8GB Radeon RX 7600 and the 16GB Radeon RX 7600 XT.

These dedicated GPUs have the benefit of being nearly identical to what Valve plans to ship in the Steam Machine—32 compute units (CUs) instead of Valve’s 28, but the same RDNA3 architecture. They’re also, most importantly for our purposes, pretty similar to each other—the same physical GPU die, just with slightly higher clock speeds and more RAM for the 7600 XT than for the regular 7600.

Testing shows why the Steam Machine’s 8GB of graphics RAM could be a problem Read More »

Netflix quietly drops support for casting to most TVs

Apps, Google, google cast, netflix, smart tvs, streaming video, Tech / Kris Guyer / December 1, 2025

Have you been trying to cast Stranger Things from your phone, only to find that your TV isn’t cooperating? It’s not the TV—Netflix is to blame for this one, and it’s intentional. The streaming app has recently updated its support for Google Cast to disable the feature in most situations. You’ll need to pay for one of the company’s more expensive plans, and even then, Netflix will only cast to older TVs and streaming dongles.

The Google Cast system began appearing in apps shortly after the original Chromecast launched in 2013. Since then, Netflix users have been able to start video streams on TVs and streaming boxes from the mobile app. That was vital for streaming targets without their own remote or on-screen interface, but times change.

Today, Google has moved beyond the remote-free Chromecast experience, and most TVs have their own standalone Netflix apps. Netflix itself is also allergic to anything that would allow people to share passwords or watch in a new place. Over the last couple of weeks, Netflix updated its app to remove most casting options, mirroring a change in 2019 to kill Apple AirPlay.

The company’s support site (spotted by Android Authority) now clarifies that casting is only supported in a narrow set of circumstances. First, you need to be paying for one of the ad-free service tiers, which start at $18 per month. Those on the $8 ad-supported plan won’t have casting support.

Even then, Casting only appears for devices without a remote, like the earlier generations of Google Chromecasts, as well as some older TVs with Cast built in. For example, anyone still rocking Google’s 3rd Gen Chromecast from 2018 can cast video in Netflix, but those with the 2020 Chromecast dongle (which has a remote and a full Android OS) will have to use the TV app. Essentially, anything running Android/Google TV or a smart TV with a full Netflix app will force you to log in before you can watch anything.

Netflix quietly drops support for casting to most TVs Read More »

Achieving lasting remission for HIV

health, HIV, remission, Science, syndication / Kris Guyer / November 29, 2025

Promising trials using engineered antibodies suggest that “functional cures” may be in reach.

A digital illustration of an HIV-infected T cell. Once infected, the immune cell is hijacked by the virus to produce and release many new viral particles before dying. As more T-cells are destroyed, the immune system is progressively weakened. Credit: Kateryna Kon/Science Photo Library via Getty Images

Around the world, some 40 million people are living with HIV. And though progress in treatment means the infection isn’t the death sentence it once was, researchers have never been able to bring about a cure. Instead, HIV-positive people must take a cocktail of antiretroviral drugs for the rest of their lives.

But in 2025, researchers reported a breakthrough that suggests that a “functional” cure for HIV—a way to keep HIV under control long-term without constant treatment—may indeed be possible. In two independent trials using infusions of engineered antibodies, some participants remained healthy without taking antiretrovirals, long after the interventions ended.

In one of the trials—the FRESH trial, led by virologist Thumbi Ndung’u of the University of KwaZulu-Natal and the Africa Health Research Institute in South Africa—four of 20 participants maintained undetectable levels of HIV for a median of 1.5 years without taking antiretrovirals. In the other, the RIO trial set in the United Kingdom and Denmark and led by Sarah Fidler, a clinical doctor and HIV research expert at Imperial College London, six of 34 HIV-positive participants have maintained viral control for at least two years.

These landmark proof-of-concept trials show that the immune system can be harnessed to fight HIV. Researchers are now looking to conduct larger, more representative trials to see whether antibodies can be optimized to work for more people.

“I do think that this kind of treatment has the opportunity to really shift the dial,” Fidler says, “because they are long-acting drugs”—with effects that can persist even after they’re no longer in the body. “So far, we haven’t seen anything that works like that.”

People with HIV can live long, healthy lives if they take antiretrovirals. But their lifespans are still generally shorter than those of people without the virus. And for many, daily pills or even the newer, bimonthly injections present significant financial, practical, and social challenges, including stigma. “Probably for the last about 15 or 20 years, there’s been this real push to go, ‘How can we do better?’” says Fidler.

The dream, she says, is “what people call curing HIV, or a remission in HIV.” But that has presented a huge challenge because HIV is a master of disguise. The virus evolves so quickly after infection that the body can’t produce new antibodies quickly enough to recognize and neutralize it.

And some HIV hides out in cells in an inactive state, invisible to the immune system. These evasion tactics have outwitted a long succession of cure attempts. Aside from a handful of exceptional stem-cell transplants, interventions have consistently fallen short of a complete cure—one that fully clears HIV from the body.

A functional cure would be the next best thing. And that’s where a rare phenomenon offers hope: Some individuals with long-term HIV do eventually produce antibodies that can neutralize the virus, though too late to fully shake it. These potent antibodies target critical, rarely changing parts of HIV proteins in the outer viral membrane; these proteins are used by the virus to infect cells. The antibodies, able to recognize a broad range of virus strains, are termed broadly neutralizing.

Scientists are now racing to find the most potent broadly neutralizing antibodies and engineer them into a functional cure. FRESH and RIO are arguably the most promising attempts yet.

In the FRESH trial, scientists chose two antibodies that, combined, were likely to be effective against HIV strains known as HIV-1 clade C, which is dominant in sub-Saharan Africa. The trial enrolled young women from a high-prevalence community as part of a broader social empowerment program. The program had started the women on HIV treatment within three days of their infection several years earlier.

The RIO trial, meanwhile, chose two well-studied antibodies shown to be broadly effective. Its participants were predominantly white men around age 40 who also had gone on antiretroviral drugs soon after infection. Most had HIV-1 clade B, which is more prevalent in Europe.

By pairing antibodies, the researchers aimed to decrease the likelihood that HIV would develop resistance—a common challenge in antibody treatments—since the virus would need multiple mutations to evade both.

Participants in both trials were given an injection of the antibodies, which were modified to last around six months in the body. Then their treatment with antiviral medications was paused. The hope was that the antibodies would work with the immune system to kill active HIV particles, keeping the virus in check. If the effect didn’t last, HIV levels would rise after the antibodies had been broken down, and the participants would resume antiretroviral treatment.

Excitingly, however, findings in both trials suggested that, in some people, the interventions prompted an ongoing, independent immune response, which researchers likened to the effect of a vaccine.

In the RIO trial, 22 of the 34 people receiving broadly neutralizing antibodies had not experienced a viral rebound by 20 weeks. At this point, they were given another antibody shot. Beyond 96 weeks—long after the antibodies had disappeared — six still had viral levels low enough to remain off antiviral medications.

An additional 34 participants included in the study as controls received only a saline infusion and mostly had to resume treatment in four to six weeks; all but three were back on treatment within 20 weeks.

A similar pattern was observed in FRESH (although, because it was mostly a safety study, this trial did not include control participants). Six of the 20 participants retained viral suppression for 48 weeks after the antibody infusion, and of those, four remained off treatment for more than a year. Two and a half years after the intervention, one remains off antiretroviral medication. Two others also maintained viral control but eventually chose to go back on treatment for personal and logistical reasons.

It’s unknown when the virus might rebound, so the researchers are cautious about calling participants in remission functionally cured. However, the antibodies clearly seem to coax the immune system to fight the virus. Attached to infected cells, they signal to immune cells to come in and kill.

And importantly, researchers believe that this immune response to the antibodies may also stimulate immune cells called CD8+ T cells, which then hunt down HIV-infected cells. This could create an “immune memory” that helps the body control HIV even after the antibodies are gone.

The response resembles the immune control seen in a tiny group (fewer than 1 percent) of individuals with HIV, known as elite controllers. These individuals suppress HIV without the help of antiretrovirals, confining it mostly to small reservoirs. That the trials helped some participants do something similar is exciting, says Joel Blankson, an infectious diseases expert at Johns Hopkins Medicine, who coauthored an article about natural HIV controllers in the 2024 Annual Review of Immunology. “It might teach us how to be able to do this much more effectively, and we might be able to get a higher percentage of people in remission.”

One thing scientists do know is that the likelihood of achieving sustained control is higher if people start antiretroviral treatment soon after infection, when their immune systems are still intact and their viral reservoirs are small.

But post-treatment control can occur even in people who started taking antiretrovirals a long time after they were initially infected: a group known as chronically infected patients. “It just happens less often,” Blankson says. “So it’s possible the strategies that are involved in these studies will also apply to patients who are chronically infected.”

A particularly promising finding of the RIO trial was that the antibodies also affected dormant HIV hiding out in some cells. These reservoirs are how the virus rebounds when people stop treatment, and antibodies aren’t thought to touch them. Researchers speculate that the T cells boosted by the antibodies can recognize and kill latently infected cells that display even trace amounts of HIV on their surface.

The FRESH intervention, meanwhile, targeted the stubborn HIV reservoirs more directly through incorporating another drug, called vesatolimod. It’s designed to stimulate immune cells to respond to the HIV threat, and hopefully to “shock” dormant HIV particles out of hiding. Once that happens, the immune system, with the help of the antibodies, can recognize and kill them.

The results of FRESH are exciting, Ndung’u says, “because it might indicate that this regimen worked, to an extent. Because this was a small study, it’s difficult to, obviously, make very hard conclusions.” His team is still investigating the data.

Once he secures funding, Ndung’u aims to run a larger South Africa-based trial including chronically infected individuals. Fidler’s team, meanwhile, is recruiting for a third arm of RIO to try to determine whether pausing antiretroviral treatment for longer before administering the antibodies prompts a stronger immune response.

A related UK-based trial, called AbVax, will add a T-cell-stimulating drug to the mix to see whether it enhances the long-lasting, vaccine-like effect of the antibodies. “It could be that combining different approaches enhances different bits of the immune system, and that’s the way forward,” says Fidler, who is a co-principal investigator on that study.

For now, Fidler and Ndung’u will continue to track the virally suppressed participants — who, for the first time since they received their HIV diagnoses, are living free from the demands of daily treatment.

This story originally appeared at Knowable Magazine.

Knowable Magazine explores the real-world significance of scholarly work through a journalistic lens.

Achieving lasting remission for HIV Read More »

Reintroduced carnivores’ impacts on ecosystems are still coming into focus

Science, syndication, wolves, yellowstone national park / Kris Guyer / November 28, 2025

He said he was surprised by how few studies show evidence of wolves, bears, and cougars having an effect on elk, moose, and deer populations. Instead, the biggest driver of changing elk population numbers across the West is humanity.

“In most mainland systems, it’s only when you combine wolves with grizzly bears and you take away human hunting as a substantial component that you see them suppressing prey numbers,” Wilmers said. “Outside of that, they’re mostly background noise against how humans are managing their prey populations.”

In some studies, ungulate populations actually increased slightly in the presence of wolves and grizzlies, Wilmers said, likely because human wildlife managers overestimated the effects of predators as they reduced hunting quotas.

“This is a much-needed review, as it is well executed, and highlights areas where more research is needed,” said Rae Wynn-Grant, a wildlife ecologist and cohost of the television show Mutual of Omaha’s Wild Kingdom Protecting the Wild, in an email to Inside Climate News. Wynn-Grant was not involved in the paper, and her work was not part of its survey.

In her view, the paper showed that an increase in predators on the landscape doesn’t automatically balance plant communities. “Our world would be much simpler if it did,” she said, “but the evidence suggests that so many variables factor into if and how ecosystems respond to increases in carnivore population in North America.”

Yellowstone, with its expansive valleys, relatively easy access, and status as an iconic, protected landscape, has become a hotspot for scientists trying to answer an existential question: Is it possible for an ecosystem that’s lost keystone large carnivores to be restored to a pre-extinction state upon their reintroduction?

Wilmers doesn’t think scientists have answered that question yet, except to show that it can take decades to untangle the web of factors driving ecological shifts in a place like Yellowstone. Any changes that do occur when a predator is driven to extinction may be impossible to reverse quickly, he said.

Yellowstone’s alternative stable state was a point echoed by researchers in both camps of the trophic cascade debate, and it is one Wilmers believes is vital to understand when evaluating the tradeoffs of large-carnivore reintroduction.

“You’d be better off avoiding the loss of beavers and wolves in the first place than you would be accepting that loss and trying to restore them later,” he said.

This story originally appeared on Inside Climate News.

Reintroduced carnivores’ impacts on ecosystems are still coming into focus Read More »

The Big Nonprofits Post 2025

Nonprofits / Kris Guyer / November 27, 2025

There remain lots of great charitable giving opportunities out there.

I have now had three opportunities to be a recommender for the Survival and Flourishing Fund (SFF). I wrote in detail about my first experience back in 2021, where I struggled to find worthy applications.

The second time around in 2024, there was an abundance of worthy causes. In 2025 there were even more high quality applications, many of which were growing beyond our ability to support them.

Thus this is the second edition of The Big Nonprofits Post, primarily aimed at sharing my findings on various organizations I believe are doing good work, to help you find places to consider donating in the cause areas and intervention methods that you think are most effective, and to offer my general perspective on how I think about choosing where to give.

This post combines my findings from the 2024 and 2025 rounds of SFF, and also includes some organizations that did not apply to either round, so inclusion does not mean that they necessarily applied at all.

This post is already very long, so the bar is higher for inclusion this year than it was last year, especially for new additions.

If you think they are better places to give and better causes to back, act accordingly, especially if they’re illegible or obscure. You don’t need my approval.

The Big Nonprofits List 2025 is also available as a website, where you can sort by mission, funding needed or confidence, or do a search and have handy buttons.

Organizations where I have the highest confidence in straightforward modest donations now, if your goals and model of the world align with theirs, are in bold, for those who don’t want to do a deep dive.

A vibrant and heartwarming scene depicting various charity fundraising efforts happening in a community park. People of diverse backgrounds are engaged in activities like a bake sale with colorful desserts displayed on tables, a charity run with participants wearing numbers on their shirts, a live music performance on a small stage, and a silent auction with art pieces on display. Children are running a lemonade stand, while volunteers distribute flyers and collect donations in jars. The atmosphere is lively and festive, with balloons, banners, and smiling faces everywhere.

The SFF recommender process is highly time constrained, and in general I am highly time constrained.

Even though I used well beyond the number of required hours in both 2024 and 2025, there was no way to do a serious investigation of all the potentially exciting applications. Substantial reliance on heuristics was inevitable.

Also your priorities, opinions, and world model could be very different from mine.

If you are considering donating a substantial (to you) amount of money, please do the level of personal research and consideration commensurate with the amount of money you want to give away.

If you are considering donating a small (to you) amount of money, or if the requirement to do personal research might mean you don’t donate to anyone at all, I caution the opposite: Only do the amount of optimization and verification and such that is worth its opportunity cost. Do not let the perfect be the enemy of the good.

For more details of how the SFF recommender process works, see my post on the process.

Note that donations to some of the organizations below may not be tax deductible.

I apologize in advance for any errors, any out of date information, and for anyone who I included who I did not realize would not want to be included. I did my best to verify information, and to remove any organizations that do not wish to be included.

If you wish me to issue a correction of any kind, or to update your information, I will be happy to do that at least through the end of the year.

If you wish me to remove your organization entirely, for any reason, I will do that, too.

What I unfortunately cannot do, in most cases, is take the time to analyze or debate beyond that. I also can’t consider additional organizations for inclusion. My apologies.

The same is true for the website version.

I am giving my full opinion on all organizations listed, but where I feel an organization would be a poor choice for marginal dollars even within its own cause and intervention area, or I anticipate my full opinion would not net help them, they are silently not listed.

Listen to arguments and evidence. But do not let me, or anyone else, tell you any of:

What is important.
What is a good cause.
What types of actions are best to make the change you want to see in the world.
What particular strategies are most promising.
That you have to choose according to some formula or you’re an awful person.

This is especially true when it comes to policy advocacy, and especially in AI.

If an organization is advocating for what you think is bad policy, or acting in a way that does bad things, don’t fund them!

If an organization is advocating or acting in a way you think is ineffective, don’t fund them!

Only fund people you think advance good changes in effective ways.

Not cases where I think that. Cases where you think that.

During SFF, I once again in 2025 chose to deprioritize all meta-level activities and talent development. I see lots of good object-level work available to do, and I expected others to often prioritize talent and meta activities.

The counterargument to this is that quite a lot of money is potentially going to be freed up soon as employees of OpenAI and Anthropic gain liquidity, including access to DAFs (donor advised funds). This makes expanding the pool more exciting.

I remain primarily focused on those who in some form were helping ensure AI does not kill everyone. I continue to see highest value in organizations that influence lab or government AI policies in the right ways, and continue to value Agent Foundations style and other off-paradigm technical research approaches.

I believe that the best places to give are the places where you have local knowledge.

If you know of people doing great work or who could do great work, based on your own information, then you can fund and provide social proof for what others cannot.

The less legible to others the cause, and the harder it is to fit it into the mission statements and formulas of various big donors, the more excited you should be to step forward, if the cause is indeed legible to you. This keeps you grounded, helps others find the show (as Tyler Cowen says), is more likely to be counterfactual funding, and avoids information cascades or looking under streetlights for the keys.

Most importantly it avoids adverse selection. The best legible opportunities for funding, the slam dunk choices? Those are probably getting funded. The legible things that are left are the ones that others didn’t sufficiently fund yet.

If you know why others haven’t funded, because they don’t know about the opportunity? That’s a great trade.

The process of applying for grants, raising money, and justifying your existence sucks.

A lot.

It especially sucks for many of the creatives and nerds that do a lot of the best work.

It also sucks to have to worry about running out of money, or to have to plan your work around the next time you have to justify your existence, or to be unable to be confident in choosing ambitious projects.

If you have to periodically go through this process, and are forced to continuously worry about making your work legible and how others will judge it, that will substantially hurt your true productivity. At best it is a constant distraction. By default, it is a severe warping effect. A version of this phenomenon is doing huge damage to academic science.

As I noted in my AI updates, the reason this blog exists is that I received generous, essentially unconditional, anonymous support to ‘be a public intellectual’ and otherwise pursue whatever I think is best. My benefactors offer their opinions when we talk because I value their opinions, but they never try to influence my decisions, and I feel zero pressure to make my work legible in order to secure future funding.

If you have money to give, and you know individuals who should clearly be left to do whatever they think is best without worrying about raising or earning money, who you are confident would take advantage of that opportunity and try to do something great, then giving them unconditional grants is a great use of funds, including giving them ‘don’t worry about reasonable expenses’ levels of funding.

This is especially true when combined with ‘retrospective funding,’ based on what they have already done. It would be great if we established a tradition and expectation that people who make big contributions can expect such rewards.

Not as unconditionally, it’s also great to fund specific actions and projects and so on that you see not happening purely through lack of money, especially when no one is asking you for money.

This includes things that you want to exist, but that don’t have a path to sustainability or revenue, or would be importantly tainted if they needed to seek that. Fund the project you want to see in the world. This can also be purely selfish, often in order to have something yourself you need to create it for everyone, and if you’re tempted there’s a good chance that’s a great value.

Resist the temptation to think purely on the margin, asking only what one more dollar can do. The incentives get perverse quickly. Organizations are rewarded for putting their highest impact activities in peril. Organizations that can ‘run lean’ or protect their core activities get punished.

If you always insist on being a ‘funder of last resort’ that requires key projects or the whole organization otherwise be in trouble, you’re defecting. Stop it.

Also, you want to do some amount of retrospective funding. If people have done exceptional work in the past, you should be willing to give them a bunch more rope in the future, above and beyond the expected value of their new project.

Don’t make everyone constantly reprove their cost effectiveness each year, or at least give them a break. If someone has earned your trust, then if this is the project they want to do next, presume they did so because of reasons, although you are free to disagree with those reasons.

This especially goes for AI lab employees. There’s no need for everyone to do all of their own research, you can and should compare notes with those who you can trust, and this is especially great when they’re people you know well.

What I do worry about is too much outsourcing of decisions to larger organizations and institutional structures, including those of Effective Altruism but also others, or letting your money go directly to large foundations where it will often get captured.

Jaan Tallinn created SFF in large part to intentionally take his donation decisions out of his hands, so he could credibly tell people those decisions were out of his hands, so he would not have to constantly worry that people he talked to were attempting to fundraise.

This is a huge deal. Communication, social life and a healthy information environment can all be put in danger by this.

Time to talk about the organizations themselves.

Rather than offer precise rankings, I divided by cause category and into three confidence levels.

High confidence means I have enough information to be confident the organization is at least a good pick.
Medium or low confidence means exactly that – I have less confidence that the choice is wise, and you should give more consideration to doing your own research.
If my last investigation was in 2024, and I haven’t heard anything, I will have somewhat lower confidence now purely because my information is out of date.

Low confidence is still high praise, and very much a positive assessment! Most organizations would come nowhere close to making the post at all.

If an organization is not listed, that does not mean I think they would be a bad pick – they could have asked not to be included, or I could be unaware of them or their value, or I could simply not have enough confidence to list them.

I know how Bayesian evidence works, but this post is not intended as a knock on anyone, in any way. Some organizations that are not here would doubtless have been included, if I’d had more time.

I try to give a sense of how much detailed investigation and verification I was able to complete, and what parts I have confidence in versus not. Again, my lack of confidence will often be purely about my lack of time to get that confidence.

Unless I already knew them from elsewhere, assume no organizations here got as much attention as they deserve before you decide on what for you is a large donation.

I’m tiering based on how I think about donations from you, from outside SFF.

I think the regranting organizations were clearly wrong choices from within SFF, but are reasonable picks if you don’t want to do extensive research, especially if you are giving small.

In terms of funding levels needed, I will similarly divide into three categories.

They roughly mean this, to the best of my knowledge:

Low: Could likely be fully funded with less than ~$250k.

Medium: Could plausibly be fully funded with between ~$250k and ~$2 million.

High: Could probably make good use of more than ~$2 million.

These numbers may be obsolete by the time you read this. If you’re giving a large amount relative to what they might need, check with the organization first, but also do not be so afraid of modest amounts of ‘overfunding’ as relieving fundraising pressure is valuable and as I noted it is important not to only think on the margin.

A lot of organizations are scaling up rapidly, looking to spend far more money than they have in the past. This was true in 2024, and 2025 has only accelerated this trend. A lot more organizations are in ‘High’ now but I decided not to update the thresholds.

Everyone seems eager to double their headcount. I’m not putting people into the High category unless I am confident they can scalably absorb more funding after SFF.

The person who I list as the leader of an organization will sometimes accidentally be whoever was in charge of fundraising rather than strictly the leader. Partly the reason for listing it is to give context and some of you can go ‘oh right, I know who that is,’ and the other reason is that all organization names are often highly confusing – adding the name of the organization’s leader allows you a safety check, to confirm that you are indeed pondering the same organization I am thinking of!

This is my post, so I get to list Balsa Research first. (I make the rules here.)

If that’s not what you’re interested in, you can of course skip the section.

Focus: Groundwork starting with studies to allow repeal of the Jones Act

Leader: Zvi Mowshowitz

Funding Needed: Medium

Confidence Level: High

Our first target continues to be the Jones Act. With everything happening in 2025, it is easy to get distracted. We have decided to keep eyes on the prize.

We’ve commissioned two studies. Part of our plan is to do more of them, and also do things like draft model repeals and explore ways to assemble a coalition and to sell and spread the results, to enable us to have a chance at repeal.

We also are networking, gathering information, publishing findings where there are information holes or where we can offer superior presentations, planning possible collaborations, and responding quickly in case of a crisis in related areas. We believe we meaningfully reduced the probability that certain very damaging additional maritime regulations could have become law, as described in this post.

Other planned cause areas include NEPA reform and federal housing policy (to build more housing where people want to live).

We have one full time worker on the case and are trying out a potential second one.

I don’t intend to have Balsa work on AI or assist with my other work, or to take personal compensation, unless I get substantially larger donations than we have had previously, that are either dedicated to those purposes or that at least come with the explicit understanding I should consider doing that.

Further donations would otherwise be for general support.

The pitch for Balsa, and the reason I am doing it, is in two parts.

I believe Jones Act repeal and many other abundance agenda items are neglected, tractable and important, and that my way of focusing on what matters can advance them. That the basic work that needs doing is not being done, it would be remarkably cheap to do a lot of it and do it well, and that this would give us a real if unlikely chance to get a huge win if circumstances break right. Chances for progress currently look grim, but winds can change quickly, we need to be ready, and also we need to stand ready to mitigate the chance things get even worse.

I also believe that if people do not have hope for the future, do not have something to protect and fight for, or do not think good outcomes are possible, that people won’t care about protecting the future. And that would be very bad, because we are going to need to fight to protect our future if we want to have one, or have a good one.

You got to give them hope.

I could go on, but I’ll stop there.

Donate here, or get in touch at [email protected].

Focus: Zvi Mowshowitz writes a lot of words, really quite a lot.

Leader: Zvi Mowshowitz

Funding Needed: None, but it all helps, could plausibly absorb a lot

Confidence Level: High

You can also of course always donate directly to my favorite charity.

By which I mean me. I always appreciate your support, however large or small.

The easiest way to help on a small scale (of course) is a Substack subscription or Patreon. Paid substack subscriptions punch above their weight because they assist with the sorting algorithm, and also for their impact on morale.

If you want to go large then reach out to me.

Thanks to generous anonymous donors, I am able to write full time and mostly not worry about money. That is what makes this blog possible.

I want to as always be 100% clear: I am totally, completely fine as is, as is the blog.

Please feel zero pressure here, as noted throughout there are many excellent donation opportunities out there.

Additional funds are still welcome. There are levels of funding beyond not worrying.

Such additional support is always highly motivating.

Also there are absolutely additional things I could and would throw money at to improve the blog, potentially including hiring various forms of help or even expanding to more of a full news operation or startup.

As a broad category, these are organizations trying to figure things out regarding AI existential risk, without centrally attempting to either do technical work or directly to influence policy and discourse.

Lightcone Infrastructure is my current top pick across all categories. If you asked me where to give a dollar, or quite a few dollars, to someone who is not me, I would tell you to fund Lightcone Infrastructure.

Focus: Rationality community infrastructure, LessWrong, the Alignment Forum and Lighthaven.

Leaders: Oliver Habryka and Rafe Kennedy

Funding Needed: High

Confidence Level: High

Disclaimer: I am on the CFAR board which used to be the umbrella organization for Lightcone and still has some lingering ties. My writing appears on LessWrong. I have long time relationships with everyone involved. I have been to several reliably great workshops or conferences at their campus at Lighthaven. So I am conflicted here.

With that said, Lightcone is my clear number one. I think they are doing great work, both in terms of LessWrong and also Lighthaven. There is the potential, with greater funding, to enrich both of these tasks, and also for expansion.

There is a large force multiplier here (although that is true of a number of other organizations I list as well).

They made their 2024 fundraising pitch here, I encourage reading it.

Where I am beyond confident is that if LessWrong, the Alignment Forum or the venue Lighthaven were unable to continue, any one of these would be a major, quite bad unforced error.

LessWrong and the Alignment Forum a central part of the infrastructure of the meaningful internet.

Lighthaven is miles and miles away the best event venue I have ever seen. I do not know how to convey how much the design contributes to having a valuable conference, designed to facilitate the best kinds of conversations via a wide array of nooks and pathways designed with the principles of Christopher Alexander. This contributes to and takes advantage of the consistently fantastic set of people I encounter there.

The marginal costs here are large (~$3 million per year, some of which is made up by venue revenue), but the impact here is many times that, and I believe they can take on more than ten times that amount and generate excellent returns.

If we can go beyond short term funding needs, they can pay off the mortgage to secure a buffer, and buy up surrounding buildings to secure against neighbors (who can, given this is Berkeley, cause a lot of trouble) and to secure more housing and other space. This would secure the future of the space.

I would love to see them then expand into additional spaces. They note this would also require the right people.

Donate through every.org, or contact [email protected].

Focus: AI forecasting research projects, governance research projects, and policy engagement, in that order.

Leader: Daniel Kokotajlo, with Eli Lifland

Funding Needed: None Right Now

Confidence Level: High

Of all the ‘shut up and take my money’ applications in the 2024 round where I didn’t have a conflict of interest, even before I got to participate in their tabletop wargame exercise, I judged this the most ‘shut up and take my money’-ist. At The Curve, I got to participate in the exercise and participate in discussions around it, I’ve since done several more, and I’m now even more confident this is an excellent pick.

I continue to think it is a super strong case for retroactive funding as well. Daniel walked away from OpenAI, and what looked to be most of his net worth, to preserve his right to speak up. That led to us finally allowing others at OpenAI to speak up as well.

This is how he wants to speak up, and try to influence what is to come, based on what he knows. I don’t know if it would have been my move, but the move makes a lot of sense, and it has already paid off big. AI 2027 was read by the Vice President, who took it seriously, along with many others, and greatly informed the conversation. I believe the discourse is much improved as a result, and the possibility space has improved.

Note that they are comfortably funded through the medium term via private donations and their recent SFF grant.

Donate through every.org, or contact Jonas Vollmer.

Focus: AI governance, advisory and research, finding how to change decision points

Leader: Ian David Moss

Funding Needed: Medium

Confidence Level: High

EIP operates on two tracks. They have their flagship initiatives and attempts to intervene directly. They also serve as donation advisors, which I discuss in that section.

Their current flagship initiative plans are to focus on the intersection of AI governance and the broader political and economic environment, especially risks of concentration of power and unintentional power shifts from humans to AIs.

Can they indeed identify ways to target key decision points, and make a big difference? One can look at their track record. I’ve been asked to keep details confidential, but based on my assessment of private information, I confirmed they’ve scored some big wins including that they helped improve safety practices at a major AI lab, and will plausibly continue to be able to have high leverage and punch above their funding weight. You can read about some of the stuff that they can talk about here in a Founders Pledge write up.

It seems important that they be able to continue their work on all this.

I also note that in SFF I allocated less funding to EIP than I would in hindsight have liked to allocate, due to quirks about the way matching funds worked and my attempts to adjust my curves to account for it.

Donate through every.org, or contact [email protected].

Focus: Primarily polls about AI, also lobbying and preparing for crisis response.

Leader: Daniel Colson.

Also Involved: Mark Beall and Daniel Eth

Funding Needed: High

Confidence Level: High

Those polls about how the public thinks about AI, including several from last year around SB 1047 including an adversarial collaboration with Dean Ball?

Remarkably often, these are the people that did that. Without them, few would be asking those questions. Ensuring that someone is asking is super helpful. With some earlier polls I was a bit worried that the wording was slanted, and that will always be a concern with a motivated pollster, but I think recent polls have been much better at this, and they are as close to neutral as one can reasonably expect.

There are those who correctly point out that even now in 2025 the public’s opinions are weakly held and low salience, and that all you’re often picking up is ‘the public does not like AI and it likes regulation.’

Fair enough. Someone still has to show this, and show it applies here, and put a lie to people claiming the public goes the other way, and measure how things change over time. We need to be on top of what the public is thinking, including to guard against the places it wants to do dumb interventions.

They don’t only do polling. They also do lobbying and prepare for crisis responses.

Donate here, or use their contact form to get in touch.

Focus: Monitoring the AI safety record and plans of the frontier AI labs

Leader: Zach Stein-Perlman

Funding Needed: Low

Confidence Level: High

Zach has consistently been one of those on top of the safety and security plans, the model cards and other actions of the major labs, both writing up detailed feedback from a skeptical perspective and also compiling the website and its scores in various domains. Zach is definitely in the ‘demand high standards that would actually work and treat everything with skepticism’ school of all this, which I feel is appropriate, and I’ve gotten substantial benefit of his work several times.

However, due to uncertainty about whether this is the best thing for him to work on, and thus not being confident he will have this ball, Zach is not currently accepting funding, but would like people who are interested in donations to contact him via Intercom on the AI Lab Watch website.

Focus: AI capabilities demonstrations to inform decision makers on capabilities and loss of control risks

Leader: Jeffrey Ladish

Funding Needed: High

Confidence Level: High

This is clearly an understudied approach. People need concrete demonstrations. Every time I get to talking with people in national security or otherwise get closer to decision makers who aren’t deeply into AI and in particular into AI safety concerns, you need to be as concrete and specific as possible – that’s why I wrote Danger, AI Scientist, Danger the way I did. We keep getting rather on-the-nose fire alarms, but it would be better if we could get demonstrations even more on the nose, and get them sooner, and in a more accessible way.

Since last time, I’ve had a chance to see their demonstrations in action several times, and I’ve come away feeling that they have mattered.

I have confidence that Jeffrey is a good person to continue to put this plan into action.

To donate, click here or email [email protected].

Focus: Visceral demos of AI risks

Leader: Sid Hiregowdara

Funding Needed: High

Confidence Level: Medium

I was impressed by the demo I was given (so a demo demo?). There’s no question such demos fill a niche and there aren’t many good other candidates for the niche.

The bear case is that the demos are about near term threats, so does this help with the things that matter? It’s a good question. My presumption is yes, that raising situational awareness about current threats is highly useful. That once people notice that there is danger, that they will ask better questions, and keep going. But I always do worry about drawing eyes to the wrong prize.

To donate, click here or email [email protected].

Focus: Making YouTube videos about AI safety, starring Rob Miles

Leader: Rob Miles

Funding Needed: Low

Confidence Level: High

I think these are pretty great videos in general, and given what it costs to produce them we should absolutely be buying their production. If there is a catch, it is that I am very much not the target audience, so you should not rely too much on my judgment of what is and isn’t effective video communication on this front, and you should confirm you like the cost per view.

To donate, join his patreon or contact him at [email protected].

Focus: Facilitation of the AI scenario roleplaying exercises including Intelligence Rising

Leader: Shahar Avin

Funding Needed: Low

Confidence Level: High

I haven’t had the opportunity to play Intelligence Rising, but I have read the rules to it, and heard a number of strong after action reports (AARs). They offered this summary of insights in 2024. The game is clearly solid, and it would be good if they continue to offer this experience and if more decision makers play it, in addition to the AI Futures Project TTX.

To donate, reach out to [email protected].

Focus: A series of sociotechnical reports on key AI scenarios, governance recommendations and conducting AI awareness efforts.

Leader: David Kristoffersson

Funding Needed: High (combining all tracks)

Confidence Level: Low

They do a variety of AI safety related things. Their Scenario Planning continues to be what I find most exciting, although I’m also somewhat interested in their modeling cooperation initiative as well. It’s not as neglected as it was a year ago, but we could definitely use more work than we’re getting. For track record you check out their reports from 2024 in this area, and see if you think that was good work, and the rest of their website has more.

Their donation page is here, or you can contact [email protected].

Focus: Grab bag of AI safety actions, research, policy, community, conferences, standards

Leader: Mark Nitzberg

Funding Needed: High

Confidence Level: Low

There are some clearly good things within the grab bag, including some good conferences and it seems substantial support for Geoffrey Hinton, but for logistical reasons I didn’t do a close investigation to see if the overall package looked promising. I’m passing the opportunity along.

Donate here, or contact them at [email protected].

Focus: Whistleblower advising and resources for those in AI labs warning about catastrophic risks, including via Third Opinion.

Leader: Larl Koch

Funding Needed: High

Confidence Level: Medium

I’ve given them advice, and at least some amount of such resourcing is obviously highly valuable. We certainly should be funding Third Opinion, so that if someone wants to blow the whistle they can have help doing it. The question is whether if it scales this loses its focus.

Donate here, or reach out to [email protected].

Focus: Advocating for a pause on AI, including via in-person protests

Leader: Holly Elmore (USA) and Joep Meindertsma (Global)

Funding Level: Low

Confidence Level: Medium

Some people say that those who believe we should pause AI would be better off staying quiet about it, rather than making everyone look foolish.

I disagree.

I don’t think pausing right now is a good idea. I think we should be working on the transparency, state capacity, technical ability and diplomatic groundwork to enable a pause in case we need one, but that it is too early to actually try to implement one.

But I do think that if you believe we should pause? Then you should say that we should pause. I very much appreciate people standing up, entering the arena and saying what they believe in, including quite often in my comments. Let the others mock all they want.

If you agree with Pause AI that the right move is to Pause AI, and you don’t have strong strategic disagreements with their approach, then you should likely be excited to fund this. If you disagree, you have better options.

Either way, they are doing what they, given their beliefs, should be doing.

Donate here, or reach out to [email protected].

Focus: At this point, primarily AI policy advocacy, letting everyone know that If Anyone Builds It, Everyone Dies and all that, plus some research

Leaders: Malo Bourgon, Eliezer Yudkowsky

Funding Needed: High

Confidence Level: High

MIRI, concluding that it is highly unlikely alignment will make progress rapidly enough otherwise, has shifted its strategy to largely advocate for major governments coming up with an international agreement to halt AI progress and to do communications, although research still looks to be a large portion of the budget, and they have dissolved its agent foundations team. Hence the book.

That is not a good sign for the world, but it does reflect their beliefs.

They have accomplished a lot. The book is at least a modest success on its own terms in moving things forward.

I strongly believe they should be funded to continue to fight for a better future however they think is best, even when I disagree with their approach.

This is very much a case of ‘do this if and only if this aligns with your model and preferences.’

Donate here, or reach out to [email protected].

Focus: Pause-relevant research

Leader: Otto Barten

Funding Needed: Low

Confidence Level: Medium

Mostly this is the personal efforts of Otto Barten, ultimately advocating for a conditional pause. For modest amounts of money, in prior years he’s managed to have a hand in some high profile existential risk events and get the first x-risk related post into TIME magazine. He’s now pivoted to pause-relevant research (as in how to implement one via treaties, off switches, evals and threat models).

The track record and my prior investigation is less relevant now, so I’ve bumped them down to low confidence, but it would definitely be good to have the technical ability to pause and not enough work is being done on that.

To donate, click here, or get in touch at [email protected].

Some of these organizations also look at bio policy or other factors, but I judge those here as being primarily concerned with AI.

In this area, I am especially keen to rely on people with good track records, who have shown that they can build and use connections and cause real movement. It’s so hard to tell what is and isn’t effective, otherwise. Often small groups can pack a big punch, if they know where to go, or big ones can be largely wasted – I think that most think tanks on most topics are mostly wasted even if you believe in their cause.

Focus: AI research, field building and advocacy

Leaders: Dan Hendrycks

Funding Needed: High

Confidence Level: High

They did the CAIS Statement on AI Risk, helped SB 1047 get as far as it did, and have improved things in many other ways. Some of these other ways are non-public. Some of those non-public things are things I know about and some aren’t. I will simply say the counterfactual policy world is a lot worse. They’ve clearly been punching well above their weight in the advocacy space. The other arms are no slouch either, lots of great work here. Their meaningful rolodex and degree of access is very strong and comes with important insight into what matters.

They take a lot of big swings and aren’t afraid of taking risks or looking foolish. I appreciate that, even when a given attempt doesn’t fully work.

If you want to focus on their policy, then you can fund their 501(c)(4), the Action Fund, since 501c(3)s are limited in how much they can spend on political activities, keeping in mind the tax implications of that. If you don’t face any tax implications I would focus first on the 501(c)(4).

We should definitely find a way to fund at least their core activities.

Donate to the Action Fund for funding political activities, or the 501(c)(3) for research. They can be contacted at [email protected].

Focus: Tech policy research, thought leadership, educational outreach to government, fellowships.

Leader: Grace Meyer

Funding Needed: High

Confidence Level: High

FAI is centrally about innovation. Innovation is good, actually, in almost all contexts, as is building things and letting people do things.

AI is where this gets tricky. People ‘supporting innovation’ are often using that as an argument against all regulation of AI, and indeed I am dismayed to see so many push so hard on this exactly in the one place I think they are deeply wrong, when we could work together on innovation (and abundance) almost anywhere else.

FAI and resident AI studiers Samuel Hammond and Dean Ball are in an especially tough spot, because they are trying to influence AI policy from the right and not get expelled from that coalition or such spaces. There’s a reason we don’t have good alternative options for this. That requires striking a balance.

I’ve definitely had my disagreements with Hammond, including strong disagreements with his 95 theses on AI although I agreed far more than I disagreed, and I had many disagreements with his AI and Leviathan as well. He’s talked on the Hill about ‘open model diplomacy.’

I’ve certainly had many strong disagreements with Dean Ball as well, both in substance and rhetoric. Sometimes he’s the voice of reason and careful analysis, other times (from my perspective) he can be infuriating, most recently in discussions of the Superintelligence Statement, remarkably often he does some of both in the same post. He was perhaps the most important opposer of SB 1047 and went on to a stint at the White House before joining FAI.

Yet here is FAI, rather high on the list. They’re a unique opportunity, you go to war with the army you have, and both Ball and Hammond have stuck their neck out in key situations. Hammond came out opposing the moratorium. They’ve been especially strong on compute governance.

I have private reasons to believe that FAI has been effective and we can expect that to continue, and its other initiatives also mostly seem good. We don’t have to agree on everything else, so long as we all want good things and are trying to figure things out, and I’m confident that is the case here.

I am especially excited that they can speak to the Republican side of the aisle in the R’s native language, which is difficult for most in this space to do.

An obvious caveat is that if you are not interested in the non-AI pro-innovation part of the agenda (I certainly approve, but it’s not obviously a high funding priority for most readers) then you’ll want to ensure it goes where you want it.

To donate, click here, or contact them using the form here.

Focus: Youth activism on AI safety issues

Leader: Sneha Revanur

Funding Needed: Medium

Confidence Level: High

They started out doing quite a lot on a shoestring budget by using volunteers, helping with SB 1047 and in several other places. Now they are turning pro, and would like to not be on a shoestring. I think they have clearly earned that right. The caveat is risk of ideological capture. Youth organizations tend to turn to left wing causes.

The risk here is that this effectively turns mostly to AI ethics concerns. It’s great that they’re coming at this without having gone through the standard existential risk ecosystem, but that also heightens the ideological risk.

I continue to believe it is worth the risk.

To donate, go here. They can be contacted at [email protected].

Focus: AI governance standards and policy.

Leader: Nicolas Moës

Funding Needed: High

Confidence Level: High

I’ve seen credible sources saying they do good work, and that they substantially helped orient the EU AI Act to at least care at all about frontier general AI. The EU AI Act was not a good bill, but it could easily have been a far worse one, doing much to hurt AI development while providing almost nothing useful for safety.

We should do our best to get some positive benefits out of the whole thing. And indeed, they helped substantially improve the EU Code of Practice, which was in hindsight remarkably neglected otherwise.

They’re also active around the world, including the USA and China.

Donate here, or contact them here.

Focus: Specifications for good AI safety, also directly impacting EU AI policy

Leader: Henry Papadatos

Funding Needed: Medium

Confidence Level: Low

I’ve been impressed by Simeon and his track record, including here. Simeon is stepping down as leader to start a company, which happened post-SFF, so they would need to be reevaluated in light of this before any substantial donation.

Donate here, or contact them at [email protected].

Focus: Papers and projects for ‘serious’ government circles, meetings with same, policy research

Leader: Peter Wildeford

Funding Needed: Medium

Confidence Level: High

I have a lot of respect for Peter Wildeford, and they’ve clearly put in good work and have solid connections down, including on the Republican side where better coverage is badly needed, and the only other solid lead we have is FAI. Peter has also increasingly been doing strong work directly via Substack and Twitter that has been helpful to me and that I can observe directly. They are strong on hardware governance and chips in particular (as is FAI).

Given their goals and approach, funding from outside the traditional ecosystem sources would be extra helpful, ideally such efforts are fully distinct from OpenPhil.

With the shifting landscape and what I’ve observed, I’m moving them up to high confidence and priority.

Donate here, or contact them at [email protected].

Focus: Accelerating the writing of AI safety standards

Leaders: Koen Holtman and Chin Ze Shen

Funding Needed: Medium

Confidence Level: High

They help facilitate the writing of AI safety standards, for EU/UK/USA, including on the recent EU Code of Practice. They have successfully gotten some of their work officially incorporated, and another recommender with a standards background was impressed by the work and team.

This is one of the many things that someone has to do, and where if you step up and do it and no one else does that can go pretty great. Having now been involved in bill minutia myself, I know it is thankless work, and that it can really matter, both for public and private standards, and they plan to pivot somewhat to private standards.

I’m raising my confidence to high that this is at least a good pick, if you want to fund the writing of standards.

To donate, go here or reach out to [email protected].

Focus: International AI safety conferences

Leader: Fynn Heide and Sophie Thomson

Funding Needed: Medium

Confidence Level: Low

They run the IDAIS series of conferences, including successful ones involving China. I do wish I had a better model of what makes such a conference actually matter versus not mattering, but these sure seem like they should matter, and certainly well worth their costs to run them.

To donate, contact them using the form at the bottom of the page here.

Focus: UK Policy Think Tank focusing on ‘extreme AI risk and biorisk policy.’

Leader: Angus Mercer

Funding Needed: High

Confidence Level: Low

The UK has shown promise in its willingness to shift its AI regulatory focus to frontier models in particular. It is hard to know how much of that shift to attribute to any particular source, or otherwise measure how much impact there has been or might be on final policy.

They have endorsements of their influence from philosopher Toby Ord, Former Special Adviser to the UK Prime Minister Logan Graham, and Senior Policy Adviser Nitarshan Rajkumar.

I reached out to a source with experience in the UK government who I trust, and they reported back they are a fan and pointed to some good things they’ve helped with. There was a general consensus that they do good work, and those who investigated where impressed.

However, I have concerns. Their funding needs are high, and they are competing against many others in the policy space, many of which have very strong cases. I also worry their policy asks are too moderate, which might be an advantage for others.

My lower confidence this year is a combination of worries about moderate asks, worry about organizational size, and worries about the shift in governments in the UK and the UK’s ability to have real impact elsewhere. But if you buy the central idea of this type of lobbying through the UK and are fine with a large budget, go for it.

Donate here, or reach out to [email protected].

Focus: Foundations and demand for international cooperation on AI governance and differential tech development

Leader: Konrad Seifert and Maxime Stauffer

Funding Needed: High

Confidence Level: Low

As with all things diplomacy, hard to tell the difference between a lot of talk and things that are actually useful. Things often look the same either way for a long time. A lot of their focus is on the UN, so update either way based on how useful you think that approach is, and also that makes it even harder to get a good read.

They previously had a focus on the Global South and are pivoting to China, which seems like a more important focus.

To donate, scroll down on this page to access their donation form, or contact them at [email protected].

Focus: Legal team for lawsuits on catastrophic risk and to defend whistleblowers.

Leader: Tyler Whitmer

Funding Needed: Medium

Confidence Level: Medium

I wasn’t sure where to put them, but I suppose lawsuits are kind of policy by other means in this context, or close enough?

I buy the core idea of having a legal team on standby for catastrophic risk related legal action in case things get real quickly is a good idea, and I haven’t heard anyone else propose this, although I do not feel qualified to vet the operation. They were one of the organizers of the NotForPrivateGain.org campaign against the OpenAI restructuring.

I definitely buy the idea of an AI Safety Whistleblower Defense Fund, which they are also doing. Knowing there will be someone to step up and help if it comes to that changes the dynamics in helpful ways.

Donors who are interested in making relatively substantial donations or grants should contact [email protected], for smaller amounts click here.

Focus: Legal research on US/EU law on transformational AI, fellowships, talent

Leader: Moritz von Knebel

Involved: Gabe Weil

Funding Needed: High

Confidence Level: Low

I’m confident that they should be funded at all, the question is if this should be scaled up quite a lot, and what aspects of this would scale in what ways. If you can be convinced that the scaling plans are worthwhile this could justify a sizable donation.

Donate here, or contact them at [email protected].

Focus: Amplify Nick Bostrom

Leader: Toby Newberry

Funding Needed: High

Confidence Level: Low

If you think Nick Bostrom is doing great work and want him to be more effective, then this is a way to amplify that work. In general, ‘give top people support systems’ seems like a good idea that is underexplored.

Get in touch at [email protected].

Focus: Advocacy for public safety and security protocols (SSPs) and related precautions

Leader: Nick Beckstead

Funding Needed: High

Confidence Level: High

I’ve had the opportunity to consult and collaborate with them and I’ve been consistently impressed. They’re the real deal, they pay attention to detail and care about making it work for everyone, and they’ve got results. I’m a big fan.

Donate here, or contact them at [email protected].

This category should be self-explanatory. Unfortunately, a lot of good alignment work still requires charitable funding. The good news is that (even more than last year when I wrote the rest of this introduction) there is a lot more funding, and willingness to fund, than there used to be, and also the projects generally look more promising.

The great thing about interpretability is that you can be confident you are dealing with something real. The not as great thing is that this can draw too much attention to interpretability, and that you can fool yourself into thinking that All You Need is Interpretability.

The good news is that several solid places can clearly take large checks.

I didn’t investigate too deeply on top of my existing knowledge here in 2024, because at SFF I had limited funds and decided that direct research support wasn’t a high enough priority, partly due to it being sufficiently legible.

We should be able to find money previously on the sidelines eager to take on many of these opportunities. Lab employees are especially well positioned, due to their experience and technical knowledge and connections, to evaluate such opportunities, and also to provide help with access and spreading the word.

Formerly ARC Evaluations.

Focus: Model evaluations

Leaders: Beth Barnes, Chris Painter

Funding Needed: High

Confidence Level: High

Originally I wrote that we hoped to be able to get large funding for METR via non-traditional sources. That happened last year, and METR got major funding. That’s great news. Alas, they once again have to hit the fundraising trail.

METR has proven to be the gold standard for outside evaluations of potentially dangerous frontier model capabilities, and has proven its value even more so in 2025.

We very much need these outside evaluations, and to give the labs every reason to use them and no excuse not to use them, and their information has been invaluable. In an ideal world the labs would be fully funding METR, but they’re not.

So this becomes a place where we can confidently invest quite a bit of capital, make a legible case for why it is a good idea, and know it will probably be well spent.

If you can direct fully ‘square’ ‘outside’ funds that need somewhere legible to go and are looking to go large? I love METR for that.

To donate, click here. They can be contacted at [email protected].

Focus: Theoretically motivated alignment work

Leader: Jacob Hilton

Funding Needed: Medium

Confidence Level: High

There’s a long track record of good work here, and Paul Christiano remained excited as of 2024. If you are looking to fund straight up alignment work and don’t have a particular person or small group in mind, this is certainly a safe bet to put additional funds to good use and attract good talent.

Donate here, or reach out to [email protected].

Focus: Scheming, evaluations, and governance

Leader: Marius Hobbhahn

Funding Needed: Medium

Confidence Level: High

This is an excellent thing to focus on, and one of the places we are most likely to be able to show ‘fire alarms’ that make people sit up and notice. Their first year seems to have gone well, one example would be their presentation at the UK safety summit that LLMs can strategically deceive their primary users when put under pressure. They will need serious funding to fully do the job in front of them, hopefully like METR they can be helped by the task being highly legible.

They suggest looking at this paper, and also this one. I can verify that they are the real deal and doing the work.

To donate, reach out to [email protected].

Focus: Support for Roman Yampolskiy’s lab and work

Leader: Roman Yampolskiy

Funding Needed: Low

Confidence Level: High

Roman Yampolskiy is the most pessimistic known voice about our chances of not dying from AI, and got that perspective on major platforms like Joe Rogan and Lex Fridman. He’s working on a book and wants to support PhD students.

Supporters can make a tax detectable gift to the University, specifying that they intend to fund Roman Yampolskiy and the Cyber Security lab.

Focus: Interpretability research

Leader:Jesse Hoogland, Daniel Murfet, Stan van Wingerden

Funding Needed: High

Confidence Level: High

Timaeus focuses on interpretability work and sharing their results. The set of advisors is excellent, including Davidad and Evan Hubinger. Evan, John Wentworth and Vanessa Kosoy have offered high praise, and there is evidence they have impacted top lab research agendas. They’re done what I think is solid work, although I am not so great at evaluating papers directly.

If you’re interested in directly funding interpretability research, that all makes this seem like a slam dunk. I’ve confirmed that this all continues to hold true in 2025.

To donate, get in touch with Jesse at [email protected]. If this is the sort of work that you’re interested in doing, they also have a discord at http://devinterp.com/discord.

Focus: Mechanistic interpretability of how inference breaks down

Leader: Paul Riechers and Adam Shai

Funding Needed: Medium

Confidence Level: High

I am not as high on them as I am on Timaeus, but they have given reliable indicators that they will do good interpretability work. I’d (still) feel comfortable backing them.

Donate here, or contact them via webform.

Focus: Interpretability and other alignment research, incubator, hits based approach

Leader: Adam Gleave

Funding Needed: High

Confidence Level: Medium

They take the hits based approach to research, which is correct. I’ve gotten confirmation that they’re doing the real thing here. In an ideal world everyone doing the real thing would get supported, and they’re definitely still funding constrained.

To donate, click here. They can be contacted at [email protected].

Focus: AI alignment research on hierarchical agents and multi-system interactions

Leader: Jan Kulveit

Funding Needed: Medium

Confidence Level: High

I liked ACS last year, and since then we’ve seen Gradual Disempowerment and other good work, which means this now falls into the category ‘this having funding problems would be an obvious mistake.’ I ranked them very highly in SFF, and there should be a bunch more funding room.

To donate, reach out to [email protected], and note that you are interested in donating to ACS specifically.

Focus: AI safety hackathons, MATS-style programs and AI safety horizon scanning.

Leaders: Esben Kran, Jason Schreiber

Funding Needed: Medium

Confidence Level: Low

I’m (still) confident in their execution of the hackathon idea, which was the central pitch at SFF although they inform me generally they’re more centrally into the MATS-style programs. My doubt for the hackathons is on the level of ‘is AI safety something that benefits from hackathons.’ Is this something one can, as it were, hack together usefully? Are the hackathons doing good counterfactual work? Or is this a way to flood the zone with more variations on the same ideas?

As with many orgs on the list, this one makes sense if and only if you buy the plan, and is one of those ‘I’m not excited but can see it being a good fit for someone else.’

To donate, click here. They can be reached at [email protected].

Focus: Specialized superhuman systems for understanding and overseeing AI

Leaders: Jacob Steinhardt, Sarah Schwettmann

Funding Needed: High

Confidence Level: Medium

Last year they were a new org. Now they have now grown to 14 people and now have a solid track record and want to keep growing. I have confirmation the team is credible. The plan for scaling themselves is highly ambitious, with planned scale well beyond what SFF can fund. I haven’t done anything like the investigation into their plans and capabilities you would need before placing a bet that big, as AI research of all kinds gets expensive quickly.

If there is sufficient appetite to scale the amount of privately funded direct work of this type, then this seems like a fine place to look. I am optimistic on them finding interesting things, although on a technical level I am skeptical of the larger plan.

To donate, reach out to [email protected].

Focus: Developing ‘AI analysts’ that can assist policy makers.

Leaders: John Coughlan

Funding Needed: High

Confidence Level: Medium

This is a thing that RAND should be doing and that should exist. There are obvious dangers here, but I don’t think this makes them substantially worse and I do think this can potentially improve policy a lot. RAND is well placed to get the resulting models to be actually used. That would enhance state capacity, potentially quite a bit.

The problem is that doing this is not cheap, and while funding this shouldn’t fall to those reading this, it plausibly does. This could be a good place to consider sinking quite a large check, if you believe in the agenda.

Donate here.

Right now it looks likely that AGI will be based around large language models (LLMs). That doesn’t mean this is inevitable. I would like our chances better if we could base our ultimate AIs around a different architecture, one that was more compatible with being able to get it to do what we would like it to do.

One path for this is agent foundations, which involves solving math to make the programs work instead of relying on inscrutable giant matrices.

Even if we do not manage that, decision theory and game theory are potentially important for navigating the critical period in front of us, for life in general, and for figuring out what the post-transformation AI world might look like, and thus what choice we make now might do to impact that.

There are not that many people working on these problems. Actual Progress would be super valuable. So even if we expect the median outcome does not involve enough progress to matter, I think it’s still worth taking a shot.

The flip side is you worry about people ‘doing decision theory into the void’ where no one reads their papers or changes their actions. That’s a real issue. As is the increased urgency of other options. Still, I think these efforts are worth supporting, in general.

Focus: AI alignment via agent foundations

Leaders: Tamsin Leake

Funding Needed: Medium

Confidence Level: High

I have funded Orthogonal in the past. They are definitely doing the kind of work that, if it succeeded, might actually amount to something, and would help us get through this to a future world we care about. It’s a long shot, but a long shot worth trying. They very much have the ‘old school’ Yudkowsky view that relatively hard takeoff is likely and most alignment approaches are fools errands. My sources are not as enthusiastic as they once were, but there are only a handful of groups trying that have any chance at all, and this still seems like one of them.

Donate here, or get in touch at [email protected].

Focus: Math for AI alignment

Leaders: Brendan Fong and David Spivak.

Funding Needed: High

Confidence Level: High

Topos is essentially Doing Math to try and figure out what to do about AI and AI Alignment. I’m very confident that they are qualified to (and actually will) turn donated money (partly via coffee) into math, in ways that might help a lot. I am also confident that the world should allow them to attempt this.

They’re now working with ARIA. That seems great.

Ultimately it all likely amounts to nothing, but the upside potential is high and the downside seems very low. I’ve helped fund them in the past and am happy about that.

To donate, go here, or get in touch at [email protected].

Focus: Two people doing research at MIRI, in particular Sam Eisenstat

Leader: Sam Eisenstat

Funding Needed: Medium

Confidence Level: High

Given Sam Eisenstat’s previous work, including from 2025, it seems worth continuing to support him, including supporting researchers. I still believe in this stuff being worth working on, obviously only support if you do as well. He’s funded for now but that’s still only limited runway.

To donate, contact [email protected].

Focus: Johannes Mayer does agent foundations work

Leader: Johannes Mayer

Funding Needed: Low

Confidence Level: Medium

Johannes Mayer does solid agent foundations work, and more funding would allow him to hire more help.

To donate, contact [email protected].

Focus: Examining intelligence

Leader: Vanessa Kosoy

Funding Needed: Medium

Confidence Level: High

This is Vanessa Kosoy and Alex Appel, who have another research agenda formerly funded by MIRI that now needs to stand on its own after their refocus. I once again believe this work to be worth continuing even if the progress isn’t what one might hope. I wish I had the kind of time it takes to actually dive into these sorts of theoretical questions, but alas I do not, or at least I’ve made a triage decision not to.

To donate, click here. For larger amounts contact directly at [email protected]

Focus: Searching for a mathematical basis for metaethics.

Leader: Alex Zhu

Funding Needed: Low

Confidence Level: Low

Alex Zhu has run iterations of the Math & Metaphysics Symposia, which had some excellent people in attendance, and intends partly to do more things of that nature. He thinks eastern philosophy contains much wisdom relevant to developing a future ‘decision-theoretic basis of metaethics’ and plans on an 8+ year project to do that.

I’ve seen plenty of signs that the whole thing is rather bonkers, but also strong endorsements from a bunch of people I trust that there is good stuff here, and the kind of crazy that is sometimes crazy enough to work. So there’s a lot of upside. If you think this kind of approach has a chance of working, this could be very exciting. For additional information, you can see this google doc.

To donate, message Alex at [email protected].

Focus: Game theory for cooperation by autonomous AI agents

Leader: Vincent Conitzer

Funding Needed: Medium

Confidence Level: Low

This is an area MIRI and the old rationalist crowd thought about a lot back in the day. There are a lot of ways for advanced intelligences to cooperate that are not available to humans, especially if they are capable of doing things in the class of sharing source code or can show their decisions are correlated with each other.

With sufficient capability, any group of agents should be able to act as if it is a single agent, and we shouldn’t need to do the game theory for them in advance either. I think it’s good things to be considering, but one should worry that even if they do find answers it will be ‘into the void’ and not accomplish anything. Based on my technical analysis I wasn’t convinced Focal was going to sufficiently interesting places with it, but I’m not at all confident in that assessment.

They note they’re also interested in the dynamics prior to Ai becoming superintelligent, as the initial conditions plausibly matter a lot.

To donate, reach out to Vincent directly at [email protected] to be guided through the donation process.

This section is the most fun. You get unique projects taking big swings.

Focus: Feeding people with resilient foods after a potential nuclear war

Leaders: David Denkenberger

Funding Needed: High

Confidence Level: Medium

As far as I know, no one else is doing the work ALLFED is doing. A resilient food supply ready to go in the wake of a nuclear war (or other major disaster with similar dynamics) could be everything. There’s a small but real chance that the impact is enormous. In my 2021 SFF round, I went back and forth with them several times over various issues, ultimately funding them, you can read about those details here.

I think all of the concerns and unknowns from last time essentially still hold, as does the upside case, so it’s a question of prioritization, how likely you view nuclear war scenarios and how much promise you see in the tech.

If you are convinced by the viability of the tech and ability to execute, then there’s a strong case that this is a very good use of funds.

I think that this is a relatively better choice if you expect AI to remain a normal technology for a while or if your model of AI risks includes a large chance of leading to a nuclear war or other cascading impacts to human survival, versus if you don’t think this.

Research and investigation on the technical details seems valuable here. If we do have a viable path to alternative foods and don’t fund it, that’s a pretty large miss, and I find it highly plausible that this could be super doable and yet not otherwise done.

Donate here, or reach out to [email protected].

Focus: Collaborations for tools to increase civilizational robustness to catastrophes

Leader: Colby Thompson

Funding Needed: High

Confident Level: High

The principle of ‘a little preparation now can make a huge difference to resilience and robustness in a disaster later, so it’s worth doing even if the disaster is not so likely’ generalizes. Thus, the Good Ancestor Foundation, targeting nuclear war, solar flares, internet and cyber outages, and some AI scenarios and safety work.

A particular focus is archiving data and tools, enhancing synchronization systems and designing a novel emergency satellite system (first one goes up in June) to help with coordination in the face of disasters. They’re also coordinating on hardening critical infrastructure and addressing geopolitical and human rights concerns.

They’ve also given out millions in regrants.

One way I know they make good decisions is they continue to help facilitate the funding for my work, and make that process easy. They have my sincerest thanks. Which also means there is a conflict of interest, so take that into account.

Donate here, or contact them at [email protected].

Focus: Building charter cities

Leader: Kurtis Lockhart

Funding Needed: Medium

Confidence Level: Medium

I do love charter cities. There is little question they are attempting to do a very good thing and are sincerely going to attempt to build a charter city in Africa, where such things are badly needed. Very much another case of it being great that someone is attempting to do this so people can enjoy better institutions, even if it’s not the version of it I would prefer that would focus on regulatory arbitrage more.

Seems like a great place for people who don’t think transformational AI is on its way but do understand the value here.

Donate to them here, or contact them via webform.

Focus: Whole brain emulation

Leader: Randal Koene

Funding Needed: Medium

Confidence Level: Low

At this point, if it worked in time to matter, I would be willing to roll the dice on emulations. What I don’t have is much belief that it will work, or the time to do a detailed investigation into the science. So flagging here, because if you look into the science and you think there is a decent chance, this becomes a good thing to fund.

Donate here, or contact them at [email protected].

Focus: Scanning DNA synthesis for potential hazards

Leader: Kevin Esvelt, Andrew Yao and Raphael Egger

Funding Needed: Medium

Confidence Level: Medium

It is certainly an excellent idea. Give everyone fast, free, cryptographically screening of potential DNA synthesis to ensure no one is trying to create something we do not want anyone to create. AI only makes this concern more urgent. I didn’t have time to investigate and confirm this is the real deal as I had other priorities even if it was, but certainly someone should be doing this.

There is also another related effort, Secure Bio, if you want to go all out. I would fund Secure DNA first.

To donate, contact them at [email protected].

Focus: Increasing capability to respond to future pandemics, Next-gen PPE, Far-UVC.

Leader: Jake Swett

Funding Needed: Medium

Confidence Level: Medium

There is no question we should be spending vastly more on pandemic preparedness, including far more on developing and stockpiling superior PPE and in Far-UVC. It is rather a shameful that we are not doing that, and Blueprint Biosecurity plausibly can move substantial additional investment there. I’m definitely all for that.

To donate, reach out to [email protected] or head to the Blueprint Bio PayPal Giving Fund.

Focus: EU policy for AI enabled biorisks, among other things.

Leader: Patrick Stadler

Funding Needed: Low

Confidence Level: Low

Everything individually looks worthwhile but also rather scattershot. Then again, who am I to complain about a campaign for e.g. improved air quality? My worry is still that this is a small operation trying to do far too much, some of it that I wouldn’t rank too high as a priority, and it needs more focus, on top of not having that clear big win yet. They are a French nonprofit.

Donation details are at the very bottom of this page, or you can contact them at [email protected].

Focus: AI safety and biorisk for Israel

Leader: David Manheim

Funding Needed: Low

Confidence Level: Medium

Israel has Ilya’s company SSI (Safe Superintelligence) and otherwise often punches above its weight in such matters but is getting little attention. This isn’t where my attention is focused but David is presumably choosing this focus for good reason.

To support them, get in touch at [email protected].

The first best solution, as I note above, is to do your own research, form your own priorities and make your own decisions. This is especially true if you can find otherwise illegible or hard-to-fund prospects.

However, your time is valuable and limited, and others can be in better positions to advise you on key information and find opportunities.

Another approach to this problem, if you have limited time or actively want to not be in control of these decisions, is to give to regranting organizations, and take the decisions further out of your own hands.

Focus: AI governance, advisory and research, finding how to change decision points

Leader: Ian David Moss

Confidence Level: High

I discussed their direct initiatives earlier. This is listing them as a donation advisor and in their capacity of attempting to be a resource to the broader philanthropic community.

They report that they are advising multiple major donors, and would welcome the opportunity to advise additional major donors. I haven’t had the opportunity to review their donation advisory work, but what I have seen in other areas gives me confidence. They specialize in advising donors who have brad interests across multiple areas, and they list AI safety, global health, democracy and (peace and security).

To donate, click here. If you have further questions or would like to be advised, contact them at [email protected].

Focus: Conferences and advice on x-risk for those giving >$1 million per year

Leader: Simran Dhaliwal

Funding Needed: None

Confidence Level: Low

Longview is not seeking funding, instead they are offering support to large donors, and you can give to their regranting funds, including the Emerging Challenges Fund on catastrophic risks from emerging tech, which focuses non-exclusively on AI.

I had a chance to hear a pitch for them at The Curve and check out their current analysis and donation portfolio. It was a good discussion. There were definitely some areas of disagreement in both decisions and overall philosophy, and I worry they’ll be too drawn to the central and legible (a common issue with such services).

On the plus side, they’re clearly trying, and their portfolio definitely had some good things in it. So I wouldn’t want to depend on them or use them as a sole source if I had the opportunity to do something higher effort, but if I was donating on my own I’d find their analysis useful. If you’re considering relying heavily on them or donating to the funds, I’d look at the fund portfolios in detail and see what you think.

I pointed them to some organizations they hadn’t had a chance to evaluate yet.

They clearly seem open to donations aimed at particular RFPs or goals.

To inquire about their services, contact them at [email protected].

There were lots of great opportunities in SFF in both of my recent rounds. I was going to have an embarrassment of riches I was excited to fund.

Thus I decided quickly that I would not be funding any regrating organizations. If you were in the business of taking in money and then shipping it out to worthy causes, well, I could ship directly to highly worthy causes.

So there was no need to have someone else do that, or expect them to do better.

That does not mean that others should not consider such donations.

I see three important advantages to this path.

Regranters can offer smaller grants that are well-targeted.
Regranters save you a lot of time.
Regranters avoid having others try to pitch on donations.

Thus, if you are making a ‘low effort’ donation, and think others you trust that share your values to invest more effort, it makes more sense to consider regranters.

In particular, if you’re looking to go large, I’ve been impressed by SFF itself, and there’s room for SFF to scale both its amounts distributed and level of rigor.

Focus: Give out grants based on recommenders, primarily to 501c(3) organizations

Leaders: Andrew Critch and Jaan Tallinn

Funding Needed: High

Confidence Level: High

If I had to choose a regranter right now to get a large amount of funding, my pick would be to partner with and participate in the SFF process as an additional funder. The applicants and recommenders are already putting in their effort, with plenty of room for each round to scale. It is very clear there are plenty of exciting places to put additional funds.

With more funding, the decisions could improve further, as recommenders would be better motivated to devote more time, and we could use a small portion of additional funds to make them better resourced.

The downside is that SFF can’t ‘go small’ efficiently on either funders or causes.

SFF does not accept donations but they are interested in partnerships with people or institutions who are interested in participating as a Funder in a future S-Process round. The minimum requirement for contributing as a Funder to a round is $250k. They are particularly interested in forming partnerships with American donors to help address funding gaps in 501(c)(4)’s and other political organizations.

This is a good choice if you’re looking to go large and not looking to ultimately funnel towards relatively small funding opportunities or individuals.

Focus: Regranters to AI safety, existential risk, EA meta projects, creative mechanisms

Leader: Austin Chen (austin at manifund.org).

Funding Needed: Medium

Confidence Level: Medium

This is a regranter that gives its money to its own regranters, one of which was me, for unrestricted grants. They’re the charity donation offshoot of Manifold. They’ve played with crowdfunding, and with impact certificates, and ACX grants. They help run Manifest.

You’re essentially hiring these people to keep building a website and trying alternative funding allocation mechanisms, and for them to trust the judgment of selected regranters. That seems like a reasonable thing to do if you don’t otherwise know where to put your funds and want to fall back on a wisdom of crowds of sorts. Or, perhaps, if you actively want to fund the cool website.

Manifold itself did not apply, but I would think that would also be a good place to invest or donate in order to improve the world. It wouldn’t even be crazy to go around subsidizing various markets. If you send me manna there, I will set aside and use that manna to subsidize markets when it seems like the place to do that.

If you want to support Manifold itself, you can either donate or buy a SAFE by contacting Austin at [email protected].

Also I’m a regranter at Manifund, so if you wanted to, you could use that to entrust me with funds to regrant. As you can see I certainly feel I have plenty of good options here if I can’t find a better local one, and if it’s a substantial amount I’m open to general directions (e.g. ensuring it happens relatively quickly, or a particular cause area as long as I think it’s net positive, or the method of action or theory of impact). However, I’m swamped for time, so I’d probably rely mostly on what I already know.

Focus: Spinoff of LTFF, grants for AI safety projects

Leader: Thomas Larsen

Funding Needed: Medium

Confidence Level: High

Seems very straightforwardly exactly what it is, a regranter that is usually in the low six figure range. Fellow recommenders were high on Larsen’s ability to judge projects. If you think this is better than you can do on your own and you want to fund such projects, then go for it.

I’ve talked to them on background about their future plans and directions, and without sharing details their plans make me more excited here.

Donate here or contact them at [email protected].

Focus: Grants of 4-6 figures mostly to individuals, mostly for AI existential risk

Leader: Caleb Parikh (among other fund managers)

Funding Needed: High

Confidence Level: Low

The pitch on LTFF is that it is a place for existential risk people who need modest cash infusions to ask for them, and to get them without too much overhead or distortion. Looking over the list of grants, there is at least a decent hit rate.

One question is, are the marginal grants a lot less effective than the average grant?

My worry is that I don’t know the extent to which the process is accurate, fair, favors insiders or extracts a time or psychic tax on participants, favors legibility, or rewards ‘being in the EA ecosystem’ or especially the extent to which the net effects are distortionary and bias towards legibility and standardized efforts. Or the extent to which people use the system to extract funds without actually doing anything.

That’s not a ‘I think this is bad,’ it is a true ‘I do not know.’ I doubt they know either.

What do we know? They say applications should take 1-2 hours to write and between 10 minutes and 10 hours to evaluate, although that does not include time forming the plan, and this is anticipated to be an ~yearly process long term. And I don’t love that this concern is not listed under reasons not to choose to donate to the fund (although the existence of that list at all is most welcome, and the reasons to donate don’t consider the flip side either).

Given their current relationship to EA funds, you likely should consider LTFF if and only if you both want to focus on AI existential risk via regrants and also want to empower and strengthen the existing EA formal structures and general ways of being.

That’s not my preference, but it could be yours.

Donate here, or contact the fund managers at [email protected].

Focus: Regrants, fellowships and events

Leader: Allison Duettmann

Funding Needed: Medium

Confidence Level: Low

Foresight also does other things. I’m focusing here on their AI existential risk grants, which they offer on a rolling basis. I’ve advised them on a small number of potential grants, but they rarely ask.

The advantage on the regrant side would be to get outreach that wasn’t locked too tightly into the standard ecosystem. The other Foresight activities all seem clearly like good things, but the bar these days is high and since they weren’t the topic of the application I didn’t investigate.

Donate here, or reach out to [email protected].

Focus: Strategic incubator and launchpad for EA talent, research, and high-impact initiatives, with emphasis on AI safety, GCR reduction, and longtermist work

Leader: Attila Ujvari

Funding Needed: High

Confidence Level: Low

I loved the simple core concept of a ‘catered hotel’ where select people can go to be supported in whatever efforts seem worthwhile. They are now broadening their approach, scaling up and focusing on logistical and community supports, incubation and a general infrastructure play on top of their hotel. This feels less unique to me now and more of a typical (EA UK) community play, so you should evaluate it on that basis.

Donate here, or reach out to [email protected].

I am less skeptical of prioritizing AI safety talent funnels than I was last year, but I remain skeptical.

The central reason remains simple. If we have so many good organizations already, in need of so much funding, why do we need more talent funnels? Is talent our limiting factor? Are we actually in danger of losing important talent?

The clear exception is leadership and management. There remains, it appears, a clear shortage of leadership and management talent across all charitable space, and startup space, and probably flat out all of space.

Which means if you are considering stepping up and doing leadership and management, then that is likely more impactful than you might at first think.

If there was a strong talent funnel specifically for leadership or management, that would be a very interesting funding opportunity. And yes, of course there still need to be some talent funnels. Right now, my guess is we have enough, and marginal effort is best spent elsewhere.

What about for other talent? What about placements in government, or in the AI labs especially Anthropic of people dedicated to safety? What about the prospects for much higher funding availability by the time we are ready to put people to work?

If you can pull it off, empowering talent can have a large force multiplier, and the opportunity space looks better than a year ago. It seems plausible that frontier labs will soak up every strong safety candidate they can find, since the marginal returns there are very high and needs are growing rapidly.

Secondary worries include the danger you end up feeding capability researchers to AI labs, and the discount for the time delays involved.

My hunch is this will still receive relatively more attention and funding than is optimal, but marginal funds here will still be useful if deployed in places that are careful to avoid being lab talent funnels.

Focus: Learning by doing, participants work on a concrete project in the field

Leaders: Remmelt Ellen and Linda Linsefors and Robert Kralisch

Funding Needed: Low

Confidence Level: High

By all accounts they are the gold standard for this type of thing. Everyone says they are great, I am generally a fan of the format, I buy that this can punch way above its weight or cost. If I was going to back something in this section, I’d start here.

Donors can reach out to Remmelt at [email protected], or leave a matched donation to support next projects.

Focus: Paying academics small stipends to move into AI safety work

Leaders: Peter Salib (psalib @ central.uh.edu), Yonathan Arbel (yarbel @ law.ua.edu) and Kevin Frazier (kevin.frazier @ law.utexas.edu).

Funding Needed: Low

Confidence Level: High

This strategy is potentially super efficient. You have an academic that is mostly funded anyway, and they respond to remarkably small incentives to do something they are already curious about doing. Then maybe they keep going, again with academic funding. If you’re going to do ‘field building’ and talent funnel in a world short on funds for those people, this is doubly efficient. I like it. They’re now moving into hiring an academic fellow, the theory being ~1 year of support to create a permanent new AI safety law professor.

To donate, message one of leaders at the emails listed above.

Focus: Enabling ambitious research programs that are poor fits for both academia and VC-funded startups including but not limited to Drexlerian functional nanomachines, high-throughput tools and discovering new superconductors.

Leader: Benjamin Reinhardt

Funding Needed: Medium

Confidence Level: Medium

I have confirmation that Reinhardt knows his stuff, and we certainly could use more people attempting to build revolutionary hardware. If the AI is scary enough to make you not want to build the hardware, it would figure out how to build the hardware anyway. You might as well find out now.

If you’re looking to fund a talent funnel, this seems like a good choice.

To donate, go here or reach out to [email protected].

Focus: Fellowships to other organizations, such as Future Society, Safer AI and FLI.

Leader: Chiara Gerosa

Funding Needed: Medium

Confidence Level: Low

They run two fellowship cohorts a year. They seem to place people into a variety of solid organizations, and are exploring the ability to get people into various international organizations like the OECD, UN or European Commission or EU AI Office.

The more I am convinced people will actually get inside meaningful government posts, the more excited I will be.

To donate, contact [email protected].

Focus: Researcher mentorship for those new to AI safety.

Leaders: Ryan Kidd and Christian Smith.

Funding Needed: High

Confidence Level: Medium

MATS is by all accounts very good at what they do and they have good positive spillover effects on the surrounding ecosystem. The recruiting classes they’re getting are outstanding.

If (and only if) you think that what they do, which is support would-be alignment researchers starting out and especially transitioning from other professions, is what you want to fund, then you should absolutely fund them. That’s a question of prioritization.

Donate here, or contact them via webform.

Focus: X-risk residencies, workshops, coworking in Prague, fiscal sponsorships

Leader: Irena Kotikova

Funding Needed: Medium

Confidence Level: Medium

I see essentially two distinct things here.

First, you have the umbrella organization, offering fiscal sponsorship for other organizations. Based on what I know from the charity space, this is a highly valuable service – it was very annoying getting Balsa a fiscal sponsor while we waited to become a full 501c3, even though we ultimately found a very good one that did us a solid, and also annoying figuring out how to be on our own going forward.

Second, you have various projects around Prague, which seem like solid offerings in that class of action of building up EA-style x-risk actions in the area, if that is what you are looking for. So you’d be supporting some mix of those two things.

To donate, contact [email protected].

Focus: Small grants to individuals to help them develop their talent

Leader: Tyler Cowen

Funding Needed: Medium

Confidence Level: High

Emergent Ventures are not like the other talent funnels in several important ways.

It’s not about AI Safety. You can definitely apply for an AI Safety purpose, he’s granted such applications in the past, but it’s rare and topics run across the board, well beyond the range otherwise described in this post.
Decisions are quick and don’t require paperwork or looking legible. Tyler Cowen makes the decision, and there’s no reason to spend much time on your end either.
There isn’t a particular cause area this is trying to advance. He’s not trying to steer people to do any particular thing. Just to be more ambitious, and be able to get off the ground and build connections and so on. It’s not prescriptive.

I strongly believe this is an excellent way to boost the development of more talent, as long as money is serving as a limiting factor on the project, and that it is great to develop talent even if you don’t get to direct or know where it is heading. Sure, I get into rhetorical arguments with Tyler Cowen all the time, around AI and also other things, and we disagree strongly about some of the most important questions where I don’t understand how he can continue to have the views he expresses, but this here is still a great project, an amazingly cost-efficient intervention.

Donate here (specify “Emergent Ventures” in notes), or reach out to [email protected].

Focus: AI safety community building and research in South Africa

Leaders: Leo Hyams and Benjamin Sturgeon

Funding Needed: Low

Confidence Level: Low

This is a mix of AI research and building up the local AI safety community. One person whose opinion I value gave the plan and those involved in it a strong endorsement, so including it based on that.

To donate, reach out to [email protected].

Focus: Talent for AI safety in Africa

Leaders: Cecil Abungu

Funding Needed: Low

Confidence Level: Low

I have a strong endorsement in hand in terms of their past work, if you think this is a good place to go in search of talent.

To donate, reach out to [email protected].

Focus: Global talent accelerator and hiring partner for technical AI safety, supporting worker transitions into AI safety.

Leader: Roy Hagemann and Varun Agarwal

Funding Needed: Medium

Confidence Level: Low

They previously focused on India, one place with lots of talent, they’re now global. A lot has turned over in the last year, so you’ll want to check them out anew.

To donate, contact [email protected].

Focus: Mapping & creating missing orgs for AI safety (aka Charity Entrepreneurship for AI risk)

Leaders: Evan Miyazono

Funding Needed: Medium

Confidence Level: Low

There was a pivot this past year from technical research to creating ‘missing orgs’ in the AI risk space. That makes sense as a strategy if and only if you expect the funding necessary to come in, or you think they can do especially strong targeting. Given the change they will need to be reevaluated.

They receive donations from here, or you can email them at [email protected].

Focus: Fellowships and affiliate programs for new alignment researchers

Leader: Lucas Teixeira and Dusan D. Nesic

Funding Needed: High

Confidence Level: Low

There are some hits here. Gabriel Weil in particular has impressed me in our interactions and with his work and they cite a good technical paper. But also that’s with a lot of shots on goal, and I’d have liked to see some bigger hits by now.

A breakdown revealed that, largely because they start with relatively senior people, most of them get placed in a way that doesn’t require additional support. That makes them a better bet than many similar rivals.

To donate, reach out to [email protected], or fund them through Manifund here.

Focus: Journalism fellowships for oversight of AI companies.

Leader: Cillian Crosson (Ex-Talos Network; still on their board.)

Funding Needed: High

Confidence Level: Medium

They offer fellowships to support journalism that helps society navigate the emergence of increasingly advanced AI, and a few other journalism ventures. They have sponsored at least one person who went on to do good work in the area. They also sponsor article placement, which seems reasonably priced in the grand scheme of things, I think?

I am not sure this is a place we need to do more investment, or if people trying to do this even need fellowships. Hard to say. There’s certainly a lot more tech reporting and more every day, if I’m ever short of material I have no trouble finding more.

It is still a small amount of money per person that can meaningfully help people get on their feet and do something useful. We do in general need better journalism. They seem to be in a solid place but also I’d be fine with giving a bunch more funding to play with, they seem pretty unique.

Donate here, or reach out to them via webform.

Focus: Incubation of AI safety organizations

Leader: Alexandra Bos

Funding Needed: Medium

Confidence Level: Low

Why funnel individual talent when you can incubate entire organizations? I am not convinced that on the margin we currently need more of either, but I’m more receptive to the idea of an incubator. Certainly incubators can be high leverage points for getting valuable new orgs and companies off the ground, especially if your model is that once the org becomes fundable it can unlock additional funding.

If you think an incubator is worth funding, then the question is whether this is the right team. The application was solid all around, and their track record includes Timaeus and Carma, although counterfactuals are always difficult. Beyond that I don’t have a differentiator on why this is the team.

To donate, contact them at [email protected].

Focus: New AI safety org in Paris, discourse, R&D collaborations, talent pipeline

Leaders: Charbel-Raphael Segerie, Florent Berthet

Funding Needed: Low

Confidence Level: Low

They’re doing all three of discourse, direct work and talent funnels. They run the only university AI safety course in Europe, maintain the AI Safety Atlas, and have had their recommendations integrated verbatim into the EU AI Act’s Code of Practice. Their two main priorities are supporting the enforcement of the EU AI Act, and driving international agreements on AI red lines.

To donate, go here, or contact them at [email protected].

Focus: Recruitment for existential risk causes

Leader: Steve Luby

Funding Needed: Medium

Confidence Level: Low

Stanford students certainly are one place to find people worth educating about existential risk. It’s also an expensive place to be doing it, and a place that shouldn’t need extra funding. And that hates fun. And it’s not great that AI is listed third on their existential risk definition. So I’m not high on them, but it sure beats giving unrestricted funds to your Alma Mater.

Interested donors should contact Steve Luby directly at [email protected].

Focus: Talent funnel directly to AI safety and biosecurity out of high school

Leader: Peter McIntyre

Funding Needed: Low

Confidence Level: Low

Having high school students jump straight to research and placement sounds good to me, and plausibly the best version of a talent funnel investment. I haven’t confirmed details but I like the theory.

To donate, get in touch at [email protected].

Focus: Teaching rationality skills, seeking to make sense of the world and how to think

Leader: Anna Salamon

Funding Needed: High

Confidence Level: High

I am on the board of CFAR, so there is a direct and obvious conflict. Of course, I am on the board of CFAR exactly because I think this is a worthwhile use of my time, and also because Anna asked me. I’ve been involved in various ways since the beginning, including the discussions about whether and how to create CFAR in the first place.

CFAR is undergoing an attempted revival. There weren’t workshops for many years, for a variety of reasons including safety concerns and also a need to reorient. The workshops are now starting up again, with a mix of both old and new units, and I find much of the new material interesting and potentially valuable. I’d encourage people to consider attending workshops, and also donating.

To donate, click here, or reach out to [email protected].

Focus: Workshops in the style of CFAR but focused on practical courage, forming high value relationships between attendees with different skill sets and learning to care for lineages, in the hopes of repairing the anglosphere and creating new capable people to solve our problems including AI in more grounded ways.

Leader: Anna Salamon

Funding Needed: Low

Confidence Level: High

LARC is kind of a spin-off of CFAR, a place to pursue a different kind of agenda. I absolutely do not have high confidence that this will succeed, but I do have high confidence that this is a gamble worth taking, and that if those involved here (especially Anna Salamon but also others that I know) want to devote their time to trying this, that we should absolutely give them that opportunity.

Donate here.

If an organization was not included here, or was removed for the 2025 edition, again, that does not mean they aren’t good, or even that I wouldn’t endorse them if asked.

It could be because I am not aware of the organization, or lack sufficient knowledge at this point to be confident in listing them, or I fear my knowledge is obsolete.

It could be that they asked to be excluded, which happened in several cases.

If by accident I included you and you didn’t want to be included and I failed to remove you, or you don’t like the quote here, I sincerely apologize and will edit you out right away, no questions asked.

If an organization is included here, that is a good thing, but again, it does not mean you should donate without checking if it makes sense based on what you think is true, how you think the world works, what you value and what your priorities are. There are no universal right answers.

Discussion about this post

The Big Nonprofits Post 2025 Read More »

RFK Jr.’s new CDC deputy director prefers “natural immunity” over vaccines

anti-vaccine, CDC, health, public health, ralph abraham, Robert F Kenned Jr., vaccine / Kris Guyer / November 26, 2025

Under ardent anti-vaccine Health Secretary Robert F. Kennedy Jr., the Centers for Disease Control and Prevention has named Louisiana Surgeon General Ralph Abraham as its new principal deputy director—a choice that was immediately called “dangerous” and “irresponsible,” yet not as bad as it could have been, by experts.

Physician Jeremy Faust revealed the appointment in his newsletter Inside Medicine yesterday, which was subsequently confirmed by journalists. Faust noted that a CDC source told him, “I heard way worse names floated,” and although Abraham’s views are “probably pretty terrible,” he at least has had relevant experience running a public health system, unlike other current leaders of the agency.

But Abraham hasn’t exactly been running a health system the way most public health experts would recommend. Under Abraham’s leadership, the Louisiana health department waited months to inform residents about a deadly whooping cough (pertussis) outbreak. He also has a clear record of anti-vaccine views. Earlier this year, he told a Louisiana news outlet he doesn’t recommend COVID-19 vaccines because “I prefer natural immunity.” In February, he ordered the health department to stop promoting mass vaccinations, including flu shots, and barred staff from running seasonal vaccine campaigns.

RFK Jr.’s new CDC deputy director prefers “natural immunity” over vaccines Read More »