Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: The Low-Hanging Fruit Prior and sloped valleys in the loss landscape, published by Dmitry Vaintrob on August 24, 2023 on LessWrong.

You can find code for the referenced experiments in this GitHub repository

Many have postulated that training large neural networks will enforce a simplicity, or Solomonoff prior. This is grounded in the idea that simpler solutions occupy expansive regions in the weight space (there exist more generalization directions in weight space along which loss does not increase or increases very little), translating to a broad attractor basin where perturbations in weight adjustments have a marginal impact on the loss.

However, stochastic gradient descent (SGD), the workhorse of deep learning optimization, operates in a manner that challenges this simplicity-centric view. SGD is, by design, driven by the immediate gradient on the current batch of data. The nature of this process means that SGD operates like a greedy heuristic search, progressively inching towards solutions that may be incrementally better but not necessarily the simplest.

Part of this process can be understood as a collection of "grokking" steps, or phase transitions, where the network learns and "solidifies" a new circuit corresponding to correctly identifying some relationships between weights (or, mathematically, finding a submanifold). This circuit then (often) remains "turned on" (i.e., this relationship between weights stays in force) throughout learning.

From the point of view of the loss landscape, this can be conceptualized as recursively finding a valley corresponding to a circuit, then executing search within that valley until it meets another valley (corresponding to discovering a second circuit), then executing search in the joint valley of the two found circuits, and so on. As the number of circuits learned starts to saturate the available weight parameters (in the underparametrized case), old circuits may get overwritten (i.e., the network may leave certain shallow valleys while pursuing new, deeper ones). However, in small models or models not trained to convergence, we observe that large-scale circuits associated with phase transitions largely survive to the end.

A greedier picture

This idea aligns with what we call the low-hanging fruit prior concept. Once a solution that reduces loss reasonably is identified, it becomes more computationally efficient to incrementally refine this existing strategy than to overhaul it in search of an entirely new solution, even if the latter might be simpler. This is analogous to continuously picking the lowest-hanging fruit / cheapest way to reduce loss at each stage of the gradient descent optimization search process.

This model predicts that SGD training processes are more likely to find solutions that look like combinations of shallow circuits and heuristics working together rather than simpler but less decomposable algorithms. In a mathematical abstraction, suppose that we have an algorithm that consists of two circuits, each of which requires getting 10 parameters right (note that this corresponds to a measure of complexity), and each of which independently reduces the loss. Then the algorithm resulting from learning both circuits has a "complexity measure" of 20, but is more likely to be learned than a "complexity 15" algorithm with the same loss if it cannot be learned sequentially (as it is exponentially harder to correctly "guess" 20 parameters than to correctly "guess" 10 parameters twice).

Note that in general, the picture is more complicated: even when learning a single "atomic" circuit that cannot be further decomposed, the question of how easy it is to learn is not equivalent to the information content (how many parameters need to be learned), but incorporates more qualitative phenomena like basin shallowness or, m...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Diet Experiment Preregistration: Long-term water fasting + seed oil removal, published by lc on August 24, 2023 on LessWrong.

I am intrigued by exfatloss's recent theory that the primary cause of the obesity epidemic is slow, subtle damage from linoleic acid over many years. Specifically, I want to test the broader possibility that the composition of adipose tissues is what's causing the problem. Since we do not have a decade to test his hypothesis naturally, I'll retcon my current eating habits for a modified experiment.

I am just now transitioning from a very low calorie diet to a water fast, having lost around fifteen pounds in the last few weeks. I am going to attempt to continue my water fast for the next thirty days, or until I reach a BMI of <25. After the end of my water fast, I am going to rigorously exclude linoleic acid and other PUFAs from my diet; my meals will consist mostly of vegetables, rice, specially ordered low-PUFA pork, and fish. Then I'm going to eat to satiety for another month.

Normally what is expected to happen after a water fast is that you regain all of the weight you just lost, because you haven't modified your body's dietary "set point" and you wasted all of your mental strength trying to keep yourself from eating. However, if the linoleic acid hypothesis is true, all of the PUFA I've consumed over my many years as a good American are what's causing my set point to remain so high in the first place. So for this self-experiment I'm going to deliberately purge most of my remaining fat stores, fill them up with a traditional Japanese diet, and see if my "natural weight" changes.

It won't invalidate the lineolic acid hypothesis if the damage is just irreversible, but I can at least test out the hypothesis that the running composition of my fat is what causes my metabolism to freak out.

Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Why Is No One Trying To Align Profit Incentives With Alignment Research?, published by Prometheus on August 23, 2023 on LessWrong.

A whole lot of Alignment work seems to be resource-constrained. Many funders have talked about how they were only able to give grants to a small percentage of projects and work they found promising. Many researchers also receive a small fraction of what they could make in the for-profit sector (Netflix recently offered $900k for an ML position). The pipeline of recruiting talent, training, and hiring could be greatly accelerated if it wasn't contingent on continuing to receive nonprofit donations.

Possible Ideas

AI Auditing Companies

We've already seen a bit of this with ARC's eval of GPT4, but why isn't there more of this? Many companies will/are training their own models, or else using existing models in a way beyond what they were intended. Even starting with non-cutting-edge models could provide insight and train people to have the proper Security Mindset and understanding to audit larger ones. Furthermore, there has been a push to regulate and make this a required practice. The possibility of this regulation being made into law will likely be contingent on the infrastructure for it already existing. And it makes sense to take action toward this now, if we want those auditing teams to be as useful as possible, and not merely satisfy a governmental requirement. Existential concerns would also be taken more seriously by a company that has already built a reputation for auditing models.

Evals reporting

Companies don't want to see their models doing things that weren't intended (example, giving people credit card information, as was just recently demonstrated). And as time goes on, companies will want some way of showcasing their models have been rigorously tested. Audit reports covering a large, diverse set of vulnerabilities is something many will probably want.

Red teaming

Jailbreaking has been a common practice, done by a wide number of people after a model is released. Like an Evals Report, many will want a separate entity that can red team their models, the same way many tech companies hire an external cybersecurity company to provide a similar service.

Alignment as a service

This could bring in new talent and incentives toward building better understanding and talent to handle alignment. These services would be smaller scale, and would not tackle some of the "core problems" of alignment, but might provide pieces to the puzzle. Solving alignment may not be one big problem, but actually a thousand smaller problems. This gives market feedback, where the better approaches succeed more often than the worse approaches. Over time, this might steer us in a direction of actually coming up with solutions that can be scaled.

Offer procedures to better align models

Many companies will likely not know how to get their models to do the things they want them to, and they will want assistance to do it. This could start by assisting companies with basic RLHF, but might evolve to developing better methods. The better methods would be adopted by competing Alignment providers, who would also search for even better methods to provide.

Caveat: might accelerate surface-level alignment, but just further a false sense of security.

Alignment as a Product

This isn't the ideal approach, but one still worth considering. Develop new proprietary strategies for aligning models, but don't release them to the public. Instead, show the results of what these new strategies can do to companies, and sell them the strategy as a product. This might involve NDAs, which is why it is not an ideal approach. But an alignment strategy existing under an NDA is better than no strategy at all.

Mech Interp as a Service

This is perhaps not yet in reach, but might be in time. Many w...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: A Theory of Laughter, published by Steven Byrnes on August 23, 2023 on LessWrong.

1. tl;dr

There should be parallel explanations for laughter at two levels.

At the brain level, there should be some mechanism / algorithm that produces laughter, and it should fit the data of when people laugh in practice.

At the evolution level, there should be some explanation for why this mechanism exists in the first place. Why was it adaptive in our ancestors? And where did it come from - are there homologues in other animals?

I'll summarize my proposals for both of these, in the opposite order:

1.1 First half of the tl;dr: Laughter in terms of evolution

I endorse the popular theory that laughter is an indicator of "play", homologous to the play-related vocalizations and body language in other animals (e.g. the dog's "play bow").

The evolutionary purpose of play is "practice for future dangerous situations". For example, a wolf pup that engages in play-fighting and play-chasing would presumably be more skilled in its future real-life fights and chases.

The evolutionary purpose of innate communicative play signals, like laughter in humans and play-bows in dogs, is to reduce the probability of accidental escalation from practice to serious. For example, if a play-fight between two wolf-pups escalates into a real fight between the pups, that's dangerous for both pups. If the pups are emitting and responding to communicative play signals, then that kind of escalation is much less likely to happen. It's kinda the same idea as "safewords" in fight-related sports (among other places).

1.2 Second half of the tl;dr: Laughter in terms of brain algorithms

My (oversimplified) pseudocode brain "business logic" for laughter is something like:

PROPOSED BRAIN PSEUDOCODE FOR LAUGHTER:

(A) IF my hypothalamus & brainstem are getting some evidence that I'm in danger

(the "evidence" here would presumably be some of the same signals that, by themselves, would tend to activate the sympathetic nervous system)

(B) AND my hypothalamus & brainstem are simultaneously getting stronger evidence that I'm safe

(the "evidence" here would presumably be some of the same signals that, by themselves, would tend to activate the parasympathetic nervous system)

(C) AND my hypothalamus & brainstem have evidence that I'm in a social situation

(D) THEN I will emit innate play signals (e.g. laughter in humans), and also I will feel more energetic (on the margin), and more safe, less worried, etc.

Indeed, I expect that there is some genetically-specified neuron group in the hypothalamus or brainstem (or more generally, what I call the Steering Subsystem), and that when future scientists look at its various connections and their functional properties, it will be straightforwardly obvious that this neuron group and its connections are implementing the pseudocode above.

(Side note: These scientists will also find that this neuron group has various other inputs that make laughing more or less likely on the margin - inputs related to mood etc. - which I omitted from the box for simplicity.)

Note that nothing in this box is particularly tied to humans. If we're talking about 50kHz rat laughter instead of human laughter, I wouldn't change a single word in the box above. However, later in the post, I will talk about human laughter in particular, including humor, and I'll argue that this pseudocode box is a plausible match to the circumstances in which people laugh.

Also, the path by which I initially came to guess this pseudocode box (namely, introspection) was independent of how I came to believe the evolutionary story (namely, I read it in a book and it seemed obviously right). But I claim that the two stories match up beautifully - that the pseudocode box above is the natural, straightforward way to implement the "spec" associated...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Empowering Numbers: FEM since 2021, published by hhart on August 23, 2023 on The Effective Altruism Forum.

∼250,000 new contraceptive users, ∼20 million new listeners and up to ∼60x cost effectiveness

TL;DR:

In 2021, FEM launched our pilot: a 3-month family planning radio campaign, educating listeners about maternal health and effective contraception. Our ads and shows played up to 860 times in Kano state in northern Nigeria, reaching ~5.6 million. An independent survey showed that in 11 months, contraceptive use increased by ~75% among all women in the state. Since then, we launched a 9-month campaign, aired proofs of concept in three new locations and re-aired in 8 more states, reaching an estimated 20 million new listeners. We were recommended by Giving What We Can and Founders Pledge, who assess our campaigns as ~22 times as effective as cash transfers. We developed new technologies that will allow us to customise health advice to listeners and run a randomised controlled trial. We continue to build organisational capacity to scale across Nigeria and new cause areas in the years ahead.

Studio host during show recording in Kano

Promising results from our pilot (Oct - Dec 2021)

In Nigeria, 1 in 22 women dies of pregnancy-related complications.

In 2020, this meant ~1,047 mothers died for every 100,000 births. For contrast, in the United Kingdom, there were 9.8 maternal deaths per 100,000 births.

Across Nigeria, about half of women have an unmet need for effective contraception. Increasing access to contraception reduces unwanted pregnancies and increases space between births, making pregnancies much safer. It can save the lives of thousands of women and save others from devastating health conditions such as obstetric fistula, and postpartum anaemia and depression. When women and girls understand and access family planning services, they receive substantial positive effects: get more education, make more money, and enjoy the ability to take better care of their children.

Launched in September 2020, Family Empowerment Media develops evidenced-based radio campaigns to improve maternal and child health by raising awareness about contraception. In our Proof of Concept phase, we chose Kano, partnered with a Nigerian organisation, and conducted a short radio campaign.

In September 2021, on World Contraception Day, we launched a pilot campaign in collaboration with the Nigerian Ministry of Health. We produced radio programmes, including dramas, storytelling shows, Q&A programmes with health experts, and interviews with prominent religious leaders on the topic of childbirth spacing.

Our programmes reached a total of 5.6 million listeners up to 860 times and it looks like they had a substantial impact.

Data from pre-post survey by PMA, intervention period is indicated by the shaded area. Over 11 months, we saw an increase from 10% to 19% among married women and from 8% to 14% among all women.

An external survey on contraceptive use in Kano saw a 75% increase (among all women) between February 2021 and January 2022, corresponding to approximately 250,000 new contraceptive users. This is twice as many new users as over the last 5 years combined.

A similar trend was not seen in Lagos or other Nigerian states, suggesting the change was specific to Kano. We were the only new large scale family planning demand generation activity in the state at that time. We are working with researchers to better understand how much impact can be attributed to FEM' campaigns. But these results are promising.

New technology to evaluate our impact (Jan -Dec 2022)

FEM is committed to collecting robust, scientific data about the effects of our work. With funding from Grand Challenges Canada (Stars in Global Health), FEM developed a new technology to evaluate our impact: a transmitter that identifies w...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Impact obsession: Feeling like you never do enough good, published by David Althaus on August 23, 2023 on The Effective Altruism Forum.

Summary

Impact obsession is a potentially unhelpful way of relating to doing good which we've observed among effective altruists, including ourselves. (More)

What do we mean by impact obsession?

One can distinguish unhealthy and healthy forms of impact obsession. (More)

Common characteristics include an overwhelming desire for doing the most good one can do, basing one's self-worth on one's own impact, judging it by increasingly demanding standards ("impact treadmill"), overexerting oneself, neglecting or subjugating non-altruistic interests, and anxiety about having no or negative impact. (More)

Is impact obsession good or bad?

Many aspects of impact obsession are reasonable and even desirable. (More)

Others can have detrimental consequences like depression, anxiety, guilt, exhaustion, burnout, and disillusionment. (More)

What to do about (unhealthy) impact obsession?

Besides useful standard (mental) health advice, potentially helpful strategies involve, for example: reflecting on our relationship with and motives for having impact, integrating conflicting desires, shifting from avoidance to approach motivation, cultivating additional sources of meaning and self-worth, reducing resistance and non-acceptance, leaning into absurdity when being overwhelmed, and learning skills (e.g., exposure therapy, positive reframing, self-compassion) for managing common negative thoughts & emotions accompanying impact obsession. (More)

Introduction

We've noticed that many EAs, including ourselves, sometimes relate to effective altruism and impact in an unhealthy way. In this post, we describe this phenomenon, which we call 'impact obsession'. While its specifics vary from person to person, certain common patterns emerge (as others have also pointed out). Here's a (first-person) description of how this can feel, based on our own experiences and that of others we've spoken to:

By far my most important goal in life is to do as much good as I can. I connect with the logic of EA on a visceral level: seeing a human or animal suffering inevitably reminds me of just how much awfulness exists in this world, how much brighter the future could be, and that what ultimately matters is helping as many sentient beings as best I can. I feel a lot of responsibility because I might be able to make a big difference to the lives of others. Heck, the stakes might literally be astronomical once you consider how your actions might affect the long-term trajectory of Earth-originating life. This is why maximizing positive impact is so important to me.

Exactly how to go about doing the most good is a very difficult question and requires lots of strategizing and experimentation. Picking the right project in the right career path in the right cause area could easily mean having orders of magnitude more impact than if I'd settle on the first decent option. So getting this type of prioritization right is extremely important to me. I therefore spend a lot of time and emotional energy on questions like "is this cause area really the most important one?", "is this project really the best one I can do?", "is there some creative out-of-the-box idea that I'm missing and which would allow me to have way more impact?", or "am I falling prey to some cognitive bias or undue social influence which leads me down the wrong path?"

Unfortunately, the world is extremely complicated, cluelessness is often a huge problem, and many smart people - most of them way smarter than myself! - disagree about what to do. So trying to figure out what I should be doing can feel overwhelming and often fills me with despair.

My self-esteem and my happiness are greatly influenced by how much impact I think I'm...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Walk while you talk: don't balk at "no chalk", published by dkl9 on August 23, 2023 on LessWrong.

Some days ago, after a long conversation with a friend, I took retrospective notes, recalling what we discussed. (I wrote over 50 items. This is not a small sample.) For parts of the time, we walked; for other parts, we stayed in roughly the same place. Writing the notes, I could tell I was missing up to half of what we said (and forgot the order) for those stationary parts. I had no such issues recalling the walking-concurrent parts.

I've noticed this kind of discrepancy a bit before, but this was when it became obvious. Walking as you talk can help a lot towards remembering what's said.

This may be a quirk of how my mind works and wouldn't apply to everyone. I doubt I'm totally unique, so it probably at least applies to some other people, even if not everyone.

Why would this work?

This part is justified speculation, not explicitly confirmed, and not important if you're just using the method.

What seems to be going on here is that my episodic memory encodes the surroundings I see together with the conversation I hear. Later, when I recall the event, I think of what I saw around me to help bring the event to mind, and what I heard (or interpreted at the time from what I heard) comes with it.

I see two ways that the walking helps:

Walking means I'll be in different places at different stages of the conversation, so each visual memory (exact place and its surroundings) associates to fewer auditory/semantic memories (things spoken), and thus associates more strongly.

Moving in space enforces a continuous path (unless I can teleport, so maybe refrain from teleporting during conversations), which is easy to interpolate and "walk thru" from just a few points - much easier than the unpredictable transitions of conversation.

I don't know which is more important.

Corollaries and extensions

The mechanism here depends on you moving, not those with whom you speak. If you only care about your own memory (or the circumstances otherwise demand it), you could call them while walking alone and get the same effect. (I have tried this. It works.)

The method should also apply equally well to one-sided speeches to which you listen, tho those tend to be more predictable anyway.

The mechanism here depends on movement and changing surroundings, not specifically walking. I expect you'd get the same effect if you're cycling/driving/riding a vehicle, so long as you're observing what you pass by as you do so. That happens to be easy in the case of walking: you have no vehicle to obstruct your vision, and you're exposed to mild, attention-demanding risk from every direction.

If you care about the order in which things were said, don't go over the same place twice. Repeating locations makes your path overlap with itself, complicating sequential recall. (This mistake caused a bit of confusion in my example at the beginning.)

Apparently, the brain models some kinds of abstract spaces similarly to how it navigates in real life. (Citation needed. Relevant keywords "hippocampus" and "have you ever played a modern video game?") You might get the same effect if you move in an intricate video game while you talk. That might require full VR. (I have not tried this.)

Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Biosafety in BSL-3, BSL-3+ and BSL-4 Laboratories: Mapping and Recommendations for Latin America, published by JorgeTorresC on August 23, 2023 on The Effective Altruism Forum.

Executive summary

This article addresses biosafety and biosecurity in high-containment level laboratories (BSL-3 and BSL-4) in Latin America, with a focus on classification, current status, and regulatory frameworks. In the region, the lack of uniformity in data collection makes it difficult to accurately understand the infrastructure of high-containment laboratories. Regulatory frameworks vary across the region and present challenges in terms of standardization. Although countries like Colombia have made progress in this area, there is a need to establish updated and centralized regulatory frameworks in each country.

To improve biosafety and biosecurity, we make a series of recommendations such as the implementation of biological risk management systems in laboratories, the promotion of non-punitive incident reports, the standardization of supervision processes, collaboration between institutions, and the exchange of best practices.

Introduction

Biosafety and biosecurity in high-containment level laboratories (BSL-3 and BSL-4) are of vital importance for the protection of public health. These laboratories work with dangerous biological agents, so it is essential to ensure that practices, equipment, and security measures are adequate and rigorous. In this context, this article focuses on analyzing the current situation of BSL-3, BSL-3+, and BSL-4 laboratories in Latin America. We explore the increase in the construction of these laboratories at a global level, the regulatory frameworks by which they are governed, and the challenges that some Latin American countries face in their implementation. In addition, we propose several recommendations to improve biosafety and biosecurity in these laboratories.

To consult the complete map follow the link:

Classification of laboratories by biosafety levels

In 1974, the United States Centers for Disease Control and Prevention (CDC) published a document titled "Classification of etiologic agents on the basis of hazard", proposing the classification of pathogens into four risk groups. Subsequently, both the National Institutes of Health (NIH) of the United States and the World Health Organization (WHO) updated this system, thus establishing the bases for the classification of laboratories according to the risk group of the pathogens they handle (Villegas et al. al., 2007). Out of the classification of risk groups, four levels of biosafety in laboratories have been established. These levels are determined taking into account several factors, such as the infectious capacity of the pathogen, the existence of treatments or vaccines for it, the severity of the disease it causes, its transmissibility, whether it is of exotic origin or not, and the nature of the work carried out in the laboratory (Lara-Villegas et al., 2008).

Level 1 (BSL-1) laboratories use elemental equipment and practices for teaching purposes. They work with well-defined and characterized strains of microorganisms that do not cause disease in healthy people. The use of special protective equipment is not required.

Level 2 (BSL-2) laboratories adopt appropriate practices, equipment, and measures to realize clinical analysis, diagnoses, and pathology. These laboratories handle microorganisms of moderate risk that are present in the community and are associated with human diseases of variable severity.

Level 3 (BSL-3) laboratories implement appropriate practices, equipment, and measures to realize clinical analysis, diagnoses, and research. These laboratories handle known or unknown agents that have the potential to be transmitted by aerosol or splash and that can cause life-threatening infections...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Select examples of adverse selection in longtermist grantmaking, published by Linch on August 23, 2023 on The Effective Altruism Forum.

Sometimes, there is a reason other grantmakers aren't funding a fairly well-known EA (-adjacent) project.

This post is written in a professional capacity, as a volunteer/sometimes contractor for EA Funds' Long-Term Future Fund (LTFF), which is a fiscally sponsored project of Effective Ventures Foundation (UK) and Effective Ventures Foundation USA Inc. I am not and have never been an employee at either Effective Ventures entity. Opinions are my own and do not necessarily represent that of any of my employers or of either Effective Ventures entity. I originally wanted to make this post a personal shortform, but Caleb Parikh encouraged me to make it a top-level post instead.

There is an increasing number of new grantmakers popping up, and also some fairly rich donors in longtermist EA that are thinking of playing a more active role in their own giving (instead of deferring). I am broadly excited about the diversification of funding in longtermist EA. There are many advantages of having a diverse pool of funding:

Potentially increases financial stability of projects and charities

Allows for a diversification of worldviews

Encourages accountability, particularly of donors and grantmakers - if there's only one or a few funders, people might be scared of offering justified criticisms

Access to more or better networks - more diverse grantmakers might mean access to a greater diversity of networks, allowing otherwise overlooked and potentially extremely high-impact projects to be funded

Greater competition and race to excellence and speed among grantmakers - I've personally been on both sides of being faster and much slower than other grantmakers, and it's helpful to have a competitive ecosystem to improve grantee and/or donor experience

However, this comment will mostly talk about the disadvantages. I want to address adverse selection: In particular, if a project that you've heard of through normal EA channels hasn't been funded by existing grantmakers like LTFF, there is a decently high likelihood that other grantmakers have already evaluated the grant and (sometimes for sensitive private reasons) have decided it is not worth funding.

Reasons against broadly sharing reasons for rejection

From my perspective as an LTFF grantmaker, it is frequently imprudent, impractical, or straightforwardly unethical to directly make public our reasons for rejection. For example:

Our assessments may include private information that we are not able to share with other funders.

Writing up our reasons for rejection of specific projects may be time-consuming, politically unwise, and/or encourage additional ire ("punching down").

We don't want to reify our highly subjective choices too much, and public writeups of rejections can cause informational cascades.

Often other funders don't even think to ask about whether the project has already been rejected by us, and why (and/or rejected grantees don't pass on that they've been rejected by another funder).

Sharing negative information about applicants would make applying to EA Funds more costly and could discourage promising applicants.

Select examples

Here are some (highly) anonymized examples of grants I have personally observed being rejected by a centralized grantmaker. For further anonymization, in some cases I've switched details around or collapsed multiple examples into one. Most, although not all, of the examples are personal experiences from working on the LTFF. Many of these examples are grants that have later been funded by other grantmakers or private donors.

An academic wants funding for a promising sounding existential safety research intervention in an area of study that none of the LTFF grantmakers ...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: High-leverage opportunity to fund an EA org -- vote for CATF (takes 15 seconds)!, published by Ann Garth on August 23, 2023 on The Effective Altruism Forum.

tl;dr: You can move significant amounts of non-EA funding toward an EA-aligned charity in only 15 seconds by voting for Clean Air Task Force in the Charity Navigator Community Choice Awards. Vote!

Clean Air Task Force is an EA-aligned climate mitigation nonprofit which is Founders Pledge's top climate charity and has been recommended by them since 2018. CATF has also been recommended by Giving Green since 2021. CATF is doing exceptional work pushing policy solutions to enable decarbonization through new technologies, such as next-generation geothermal energy, nuclear energy (both fission and fusion), and carbon capture and storage. For example, we were influential in many of the key provisions of the IRA, as well as in keeping open the Diablo Canyon nuclear power plant in California.

CATF has been nominated for the Community Choice Awards from Charity Navigator, the largest and most-utilized charity evaluator in the United States (almost 11 million unique site visitors per year). Winning this award will earn CATF prominent visibility on Charity Navigator's website for the next year, an email announcement to Charity Navigator's audience of donors, dedicated posts on Charity Navigator's social media, and a webinar speaking opportunity. We expect that this extra visibility and engagement could drive significant funding for CATF from non-EA donors. CATF appears to be the only EA-aligned organization nominated for the Community Choice Awards, so voting is potentially very high-leverage.

You can vote for CATF at this link (#organization=Clean%20Air%20Task%20Force). Voting takes only about 15 seconds and you can vote every day through August 27th, when the first round of the contest closes. Please enable more effective giving by supporting CATF!

Disclaimer: I work at CATF (on the Superhot Rock Energy team). I joined CATF because of the Founders Pledge recommendation and love working at an organization that's so aligned with my values. I'm always happy to talk about CATF with anyone who's interested in learning more -- [email protected].

Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org