I've been thinking a lot about how mass layoffs in tech affect the EA community. I got laid off early last year, and after job searching for 7 months and pivoting to trying to start a tech startup, I'm on a career break trying to recover from burnout and depression.

Many EAs are tech professionals, and I imagine that a lot of us have been impacted by layoffs and/or the decreasing number of job openings that are actually attainable for our skill level. The EA movement depends on a broad base of high earners to sustain high-impact orgs through relatively small donations (on the order of $300-3000)—this improves funding diversity and helps orgs maintain independence from large funders like Open Philanthropy. (For example, Rethink Priorities has repeatedly argued that small donations help them pursue projects "that may not align well with the priorities or constraints of institutional grantmakers.")

It's not clear that all of us will be able to continue sustaining the level of donations we historically have, especially if we're forced out of the job markets that we spent years training and getting degrees for. I think it's incumbent on us to support each other more to help each other get back to a place where we can earn to give or otherwise have a high impact again.

I'm concerned about the new terms of service for Giving What We Can, which will go into effect after August 31, 2024:

6.3 Feedback. If you provide us with any feedback or suggestions about the GWWC Sites or GWWC’s business (the “Feedback”), GWWC may use the Feedback without obligation to you, and you irrevocably assign to GWWC all right, title, and interest in and to the Feedback. (emphasis added)

This is a significant departure from the Effective Ventures' TOS (GWWC is spinning out of EV), which has users grant EV an unlimited but non-exclusive license to use feedback or suggestions they send, while retaining the right to do anything with it themselves. I've previously talked to GWWC staff about my ideas to help people give effectively, like a donation decision worksheet that I made. If this provision goes into effect, it would deter me from sharing my suggestions with GWWC in the future because I would risk losing the right to disseminate or continue developing those ideas or materials myself.

A hack to multiply your donations by up to 102%

Disclaimer: I'm a former PayPal employee. The following statements are my opinion alone and do not reflect PayPal's views. Also, this information is accurate as of 2024-10-14 and may become outdated in the future.

More donors should consider using PayPal Giving Fund to donate to charities. To do so, go to this page, search for the charity you want, and donate through the charity's page with your PayPal account. (For example, this is GiveDirectly's page.)

PayPal covers all processing fees on charitable donations made through their giving website, so you don't have to worry about the charity losing money to credit card fees. If you use a credit card that gives you 1.5 or 2% cash back (or 1.5-2x points) on all purchases, your net donation will be multiplied by ~102%. I don't know of any credit cards that offer elevated rewards for charitable donations as a category (like many do for restaurants, groceries, etc.), so you most likely can't do better than a 2% card for donations (unless you donate stocks).

For political donations, platforms like ActBlue and Anedot charge the same processing fees to organizations regardless of what payment method you use.^[1] So you should also donate using your 1.5-2% card.

1. ^{^}

   ActBlue: 3.95% on all transactions.

   Anedot: For non-501(c)(3) organizations, 4% + 30¢ on all transactions except Bitcoin and 1% on Bitcoin transactions. 501(c)(3) organizations are charged a much lower rate for ACH transactions.

I've heard from women I know in this community that they are often shunted into low-level or community-building roles rather than object-level leadership roles. Does anyone else have knowledge about and/or experience with this?

Crazy idea: A vegan hot dog eating contest

What if there was an effective giving org that partnered with movie studios to pledge a portion of their revenue to thematically relevant high-impact charities? e.g. "5% of gross box office revenue from Chicken Run Super 3D! will be donated to Legal Impact for Chickens"

Seeing the unexpected success of KPop Demon Hunters (released in June 2025) made me think of this idea again. Imagine how cool it would be if next year's summer blockbusters used their reach for a good cause? This initiative would also diversify the EA funding base away from finance and tech.

Crazy idea: When charities apply for funding from foundations, they should be required to list 3-5 other charities they think should receive funding. Then, the grantmaker can run a statistical analysis to find orgs that are mentioned a lot and haven't applied before, reach out to those charities, and encourage them to apply. This way, the foundation can get a more diverse pool of applicants by learning about charities outside their network.

Epistemic status: preliminary take, likely not considering many factors.

I'm starting to think that economic development and animal welfare go hand in hand. Since the end of the COVID pandemic, the plant-based meat industry has declined in large part because consumers' disposable incomes declined (at least in developed countries). It's good that GFI and others are trying to achieve price parity with conventional meat. However, finding ways to increase disposable incomes (or equivalently, reduce the cost of living) will likely accelerate the adoption of meat substitutes, even if price parity isn't reached.

Okay, so one thing I don't get about "common sense ethics" discourse in EA is, which common sense ethical norms prevail? Different people even in the same society have different attitudes about what's common sense.

For example, pretty much everyone agrees that theft and fraud in the service of a good cause - as in the FTX case - is immoral. But what about cases where the governing norms are ambiguous or changing? For example, in the United States, it's considered customary to tip at restaurants and for deliveries, but there isn't much consensus on when and how much to tip, especially with digital point-of-sale systems encouraging people to tip in more situations. (Just as an example of how conceptions of "common sense ethics" can differ: I just learned that apparently, you're supposed to tip the courier before you get a delivery now, otherwise they might refuse to take your order at all. I've grown up believing that you're supposed to tip after you get service, but many drivers expect you to tip beforehand.) You're never required to tip as a condition of service, so what if you just never tipped and always donated the equivalent amount to highly effective charities instead? That sounds unethical to me but technically it's legal and not a breach of contract.

Going further, what if you started a company, like a food delivery app, that hired contractors to do the important work and paid them subminimum wages^[1], forcing them to rely on users' generosity (i.e. tips) to make a living? And then made a 40% profit margin and donated the profits to GiveWell? That also sounds unethical - you're taking with one hand and giving with the other. But in a capitalist society like the U.S., it's just business as usual.

1. ^{^}

   Under federal law and in most U.S. states, employers can pay tipped workers less than the minimum wage as long as their wages and tips add up to at least the minimum wage. However, many employers get away with not ensuring that tipped workers earn the minimum wage, or outright stealing tips.

Are there currently any safety-conscious people on the OpenAI Board?

"Quality-adjusted civilization years"

We should be able to compare global catastrophic risks in terms of the amount of time they make global civilization significantly worse and how much worse it gets. We might call this measure "quality-adjusted civilization years" (QACYs), or the quality-adjusted amount of civilization time that is lost.

For example, let's say that the COVID-19 pandemic reduces the quality of civilization by 50% for 2 years. Then the QACY burden of COVID-19 is $0.5\times 2=1$ QACYs.

Another example: suppose climate change will reduce the quality of civilization by 80% for 200 years, and then things will return to normal. Then the total QACY burden of climate change over the long term will be $0.8\times 200=160$ QACYs.

In the limit, an existential catastrophe would have a near-infinite QACY burden.

Content warning: Israel/Palestine

Has there been research on what interventions are effective at facilitating dialogue between social groups in conflict?

I remember an article about how during the last Israel-Gaza flare-up, Israelis and Palestinians were using the audio chatroom app Clubhouse to share their experiences and perspectives. This was portrayed as a phenomenon that increased dialogue and empathy between the two groups. But how effective was it? Could it generalize to other ethnic/religious conflicts around the world?

Does anyone know what's going on with Apart Research's funding situation? I participated in one of their AI safety hackathons and it propelled me into the world of AIS research, so I'm sad to hear that they might be forced to shut down or downsize. They're trying to raise nearly a million dollars in the next month.

Does anyone remember which EA org lets people specify in their wills that they are leaving money to effective charities?

Utility of money is not always logarithmic

EA discussions often assume that the utility of money is logarithmic, but while this is a convenient simplification, it's not always the case. Logarithmic utility is a special case of isoelastic utility, a.k.a. power utility, where the elasticity of marginal utility is $\eta =1$. But $\eta$ can be higher or lower. The most general form of isoelastic utility is the following:

$u\left(c\right)=\{\begin{array}{cc}\frac{c^{1-\eta }-1}{1-\eta }& \eta \ge 0,\eta \ne 1\\ \ln \left(c\right)& \eta =1\end{array}$

Some special cases:

• When $\eta =0$, we get linear utility, or $u\left(c\right)=c$.
• When $\eta =0.5$, we get the square root utility function, $u\left(c\right)=2(\sqrt{c}-1)$.
• When $\eta =1$, we get the familiar logarithmic utility function, $u\left(c\right)=\ln \left(c\right)$.
• For any $\eta >1$, the utility function asymptotically approaches a constant as $c$ approaches infinity. When $\eta =2$, we get the utility function $u\left(c\right)=1-1/c$.

$\eta$ tells us how sharply marginal utility drops off with increasing consumption: if a person already has $k$ times as much money as the baseline, then giving them an extra dollar is worth $(1/k{)}^{\eta }$ times as much. Empirical studies have found that $\eta$ for most people is between 1 and 2. So if the average GiveDirectly recipient is 100x poorer than the average American, then it could be 100 to 10,000 times as valuable to give money to GiveDirectly than to spend the same money on yourself.

Source: "The value of money going to different groups" by Toby Ord

Nonprofit idea: YIMBY for energy

YIMBY groups in the United States (like YIMBY Action) systematically advocate for housing developments as well as rezonings and other policies to create more housing in cities. YIMBYism is an explicit counter-strategy to the NIMBY groups that oppose housing development; however, NIMBYism affects energy developments as well - everything from solar farms to nuclear power plants to power lines - and is thus an obstacle to the clean energy transition.

There should be groups that systematically advocate for energy projects (which are mostly in rural areas), borrowing the tactics of the YIMBY movement. Currently, when developers propose an energy project, they do an advertising campaign to persuade local residents of the benefits of the development, but there is often opposition as well.

I think an EA career fair would be a good idea. It could have EA orgs as well as non-EA orgs that are relevant to EAs (for gaining career capital or earning to give)

I think we separate causes and interventions into "neartermist" and "longtermist" causes too much.

Just as some members of the EA community have complained that AI safety is pigeonholed as a "long-term" risk when it's actually imminent within our lifetimes^[1], I think we've been too quick to dismiss conventionally "neartermist" EA causes and interventions as not valuable from a longtermist perspective. This is the opposite failure mode of surprising and suspicious convergence - instead of assuming (or rationalizing) that the spaces of interventions that are promising from neartermist and longtermist perspectives overlap a lot, we tend to assume they don't overlap at all, because it's more surprising if the top longtermist causes are all different from the top neartermist ones. If the cost-effectiveness of different causes according to neartermism and longtermism are independent from one another (or at least somewhat positively correlated), I'd expect at least some causes to be valuable according to both ethical frameworks.

I've noticed this in my own thinking, and I suspect that this is a common pattern among EA decision makers; for example, Open Phil's "Longtermism" and "Global Health and Wellbeing" grantmaking portfolios don't seem to overlap.

Consider global health and poverty. These are usually considered "neartermist" causes, but we can also tell a just-so story about how global development interventions such as cash transfers might also be valuable from the perspective of longtermism:

• People in extreme poverty who receive cash transfers often spend the money on investments as well as consumption. For example, a study by GiveDirectly found that people who received cash transfers owned 40% more durable goods (assets) than the control group. Also, anecdotes show that cash transfer recipients often spend their funds on education for their kids (a type of human capital investment), starting new businesses, building infrastructure for their communities, and healthcare (another type of human capital).
• According to most models of economic growth, labor and capital are the main determinants of an economy's output. Increasing a community's capital stock tends to increase its economic output, some of which is reinvested in more capital, which leads to more output, and so on. The virtuous cycle of capital accumulation and increasing economic output is a key driver of long-term economic growth.
• Economic growth is valuable from a longtermist perspective because it raises the living standards and well-being of present and future generations. Furthermore, economic growth in poor countries is more valuable in the near term because the marginal utility of consumption in poor countries is higher.^[2]
• Therefore, cash transfers to the global poor are highly valuable by longtermist lights because they promote economic growth in poor countries.

Note that I'm not claiming that cash transfers are the most valuable interventions for longtermists. They're probably not, since any trend of sustained economic growth would likely run up against physical limits of the universe within the next few thousand years,^[3] and AGI is likely to render all other interventions to promote economic growth moot in the next few decades anyway. Interventions to reduce existential risk would probably have more impact over the long term (although I'm sympathetic to the argument in "Existential risk pessimism and the time of perils").

1. ^{^}

   At least those of us who are 40 and under.

2. ^{^}

   Notwithstanding growth rates. Rich countries like the United States probably sustain higher rates of economic growth and capital accumulation than poor countries because of stronger institutions that encourage investment. I'd like to see an economic model that could tell us which type of growth is more valuable in the long term, but I don't have the training in economics that one would need to create one.

3. ^{^}

   See "This Can't Go On" by Holden Karnofsky for an argument against indefinite sustained growth, and "This Can Go On" by Dwarkesh Patel for a counterargument.

An idea I liked from Owen Cotton-Barratt's new interview on the 80K podcast: Defense in depth

If S, M, or L is any small, medium, or large catastrophe and X is human extinction, then the probability of human extinction is

$$Pr\left(X\right)=Pr\left(S\right)Pr(M\mid S)Pr(L\mid S,M)Pr(X\mid S,M,L).$$

So halving the probability of all small disasters, the probability of any small disaster becoming a medium-sized disaster, etc. would halve the probability of human extinction.

On the difference between x-risks and x-risk factors

I suspect there isn't much of a meaningful difference between "x-risks" and "x-risk factors," for two reasons:

1. We can treat them the same in terms of probability theory. For example, if $X$ is an "x-risk" and $Y$ is a "risk factor" for $X$, then $Pr(X\mid Y)>Pr\left(X\right)$. But we can also say that $Pr(Y\mid X)>Pr\left(Y\right)$, because both statements are equivalent to $Pr(X,Y)>Pr\left(X\right)Pr\left(Y\right)$. We can similarly speak of the total probability of an x-risk factor because of the law of total probability (e.g. $Pr\left(Y\right)=Pr(Y\mid X_1)+Pr(Y\mid X_2)+\dots$) like we can with an x-risk.
2. Concretely, something can be both an x-risk and a risk factor. Climate change is often cited as an example: it could cause an existential catastrophe directly by making all of Earth unable to support complex societies, or indirectly by increasing humanity's vulnerability to other risks. Pandemics might also be an example, as a pandemic could either directly cause the collapse of civilization or expose humanity to other risks.

I think the difference is that x-risks are events that directly cause an existential catastrophe, such as extinction or civilizational collapse, whereas x-risk factors are events that don't have a direct causal pathway to x-catastrophe. But it's possible that pretty much all x-risks are risk factors and vice versa. For example, suppose that humanity is already decimated by a global pandemic, and then a war causes the permanent collapse of civilization. We usually think of pandemics as risks and wars as risk factors, but in this scenario, the war is the x-risk because it happened last... right?

One way to think about x-risks that avoids this problem is that x-risks can have both direct and indirect causal pathways to x-catastrophe.

One thing the EA community should try doing is multinational op-ed writing contests. The focus would be op-eds advocating for actions or policies that are important, neglected, and tractable (although the op-eds themselves don't have to mention EA); and by design, op-eds could be submitted from anywhere in the world. To make judging easier, op-eds could be required to be in a single language, but op-ed contests in multiple languages could be run in parallel (such as English, Spanish, French, and Arabic, each of which is an official language in at least 20 countries).

This would have two benefits for the EA community:

• It would be a cheap way to spread EA-aligned ideas in multiple countries. Also, the people writing the op-eds would know more about the political climates of the countries for which they are publishing them than the organizers of the contest would, and we can encourage them to tailor their messaging accordingly.
• It would also be a way to measure countries' receptiveness to EA ideas. For example, if there were multiple submissions about immigration policy, we could use them to compare the receptiveness of different countries to immigration reforms that would increase global well-being.

Possible outline for a 2-3 part documentary adaptation of The Precipice:

Part 1: Introduction & Natural Risks

• Introduce the idea that we are in a time of unprecedented existential risk, but that the future could be very good (Introduction and Chapter 1)
• Discuss natural risks (Chapter 3)
• Argue that man-made risks are greater and use this to lead to the next episode (Chapter 3)

Part 2: Human-Made Risks

• Well-known anthropogenic risks - nuclear war, climate change, other environmental damage (Chapter 4)
• Emerging technological risks - pandemics, AI, dystopia (Chapter 5)
• Existential risk and security factors (Chapter 6)

Part 3: What We Can Do

• Discuss actions society can take to minimize its existential risk (Chapter 7)

What this leaves out:

• Chapter 2 - mostly a discussion of the moral arguments for x-risk's importance. Can assume that the audience will already care about x-risk at a less sophisticated level, and focus on making the case that x-risk is high and we sort of know what to do about it.
• The discussion of joint probabilities of x-risks in Chapter 6 - too technical for a general audience

Another way to do it would be to do an episode on each type of risk and what can be done about it, for example:

• Part 1: Introduction
• Part 2: Pandemic risk (timely because of COVID-19)
• Part 3: Risks from asteroids, comets, volcanoes, and supernovas
• Part 4: Climate change
• Part 5: Artificial intelligence
• and so on

Like the original book, I'd want the tone of the documentary to be authoritative and hopeful and not lean on fear.

I think partnering with local science museums to run events on EA topics could be a great way to get EA-related ideas out to the public.

I just listened to Andrew Critch's interview about "AI Research Considerations for Human Existential Safety" (ARCHES). I took some notes on the podcast episode, which I'll share here. I won't attempt to summarize the entire episode; instead, please see this summary of the ARCHES paper in the Alignment Newsletter.

• We need to explicitly distinguish between "AI existential safety" and "AI safety" writ large. Saying "AI safety" without qualification is confusing for both people who focus on near-term AI safety problems and those who focus on AI existential safety problems; it creates a bait-and-switch for both groups.
• Although existential risk can refer to any event that permanently and drastically reduces humanity's potential for future development (paraphrasing Bostrom 2013), ARCHES only deals with the risk of human extinction because it's easier to reason about and because it's not clear what other non-extinction outcomes are existential events.
• ARCHES frames AI alignment in terms of delegation from m ≥ 1 human stakeholders (such as individuals or organizations) to n ≥ 1 AI systems. Most alignment literature to date focuses on the single-single setting (one principal, one agent), but such settings in the real world are likely to evolve into multi-principal, multi-agent settings. Computer scientists interested in AI existential safety should pay more attention to the multi-multi setting relative to the single-single one for the following reasons:
  • There are commercial incentives to develop AI systems that are aligned with respect to the single-single setting, but not to make sure they won't break down in the multi-multi setting. A group of AI systems that are "aligned" with respect to single-single may still precipitate human extinction if the systems are not designed to interact well.
  • Single-single delegation solutions feed into AI capabilities, so focusing only on single-single delegation may increase existential risk.
  • What alignment means in the multi-multi setting is more ambiguous because the presence of multiple stakeholders engenders heterogeneous preferences. However, predicting whether humanity goes extinct in the multi-multi setting is easier than predicting whether a group of AI systems will "optimally" satisfy a group's preferences.
• Critch and Krueger coin the term "prepotent AI" to refer to an AI system that is powerful enough to transform Earth's environment at least as much as humans have and where humans cannot effectively stop or reverse these changes. Importantly, a prepotent AI need not be an artificial general intelligence.

April Fools' Day is in 11 days! Get yer jokes ready 🎶

Not sure who to alert to this, but: when filling out the EA Organization Survey, I noticed that one of the fields asks for a date in DD/MM/YYYY format. As an American this tripped me up and I accidentally tried to enter a date in MM/DD/YYYY format because I am more used to seeing it.

I suggest using the ISO 8601 (YYYY-MM-DD) format on forms that are used internationally to prevent confusion, or spelling out the month (e.g. "1 December 2023" or "December 1, 2023").

I'm excited about Open Phil's new cause area, global aid advocacy. Development aid from rich countries could be used to serve several goals that many EAs care about:

• Economic development and poverty reduction
• Public health and biosecurity, including drug liberalization
• Promoting liberal democracy
• Climate change mitigation and adaptation

Also, development aid can fund a combination of randomista-style and systemic interventions (such as building infrastructure to promote growth).

The United States has two agencies that provide development aid: USAID, which provides grants and technical assistance, and the U.S. International Development Finance Corporation (DFC), which provides equity and debt financing. I'd like to see both of these agencies strengthened and expanded with more funding.

I'm especially excited about the DFC because it was created in 2019 as a counterweight to China's Belt and Road Initiative (BRI). China has used the BRI to buy influence in regions such as Africa, so aid-receiving countries don't criticize its authoritarian policies. Development finance institutions run by liberal democracies, like the DFC, can be used to peel developing countries away from China and make them more likely to support global democracy promotion initiatives.

Big O as a cause prioritization heuristic

When estimating the amount of good that can be done by working on a given cause, a good first approximation might be the asymptotic behavior of the amount of good done at each point in time (the trajectory change).

Other important factors are the magnitude of the trajectory change (how much good is done at each point in time) and its duration (how long the trajectory change lasts).

For example, changing the rate of economic growth (population growth * GDP/capita growth)  has an $O\left(t^2\right)$ trajectory change in the short run, as long as humanity doesn't expand into space. We break it down into a population growth component, which grows linearly, and a component for the change in average welfare due to GDP per capita growth^[1]. GDP per capita typically grows exponentially, and welfare is a logarithmic function of GDP per capita. These two trends cancel out, resulting in a linear trend, or $O\left(t\right)$.

If humanity expands into space, then economic growth becomes a quartic trend. Population growth becomes cubic, since humanity can fill an $O\left(t^3\right)$ amount of space in $t$ time.^[2] The average welfare due to GDP per capita growth is still linear. Multiplying the cubic and linear trends yields a quartic trend. So increasing the probability that space colonization goes well looks more important than changing the rate of economic growth on Earth.

Surprisingly, the trajectory change caused by reducing existential risk is not an exponential trend. Existential risk reduces expected welfare at each point in time by a factor of $(1-r{)}^t$, where $r$ is the probability of catastrophe per unit of time. This exponential trend causes all trajectory changes to wash out as $t$ becomes very large.

The amount of good done by reducing x-risk from $r_i$ to $r_f$ is

$\sum _{t=1}^{\infty }y\left(t\right)\left(\right(1-r_f{)}^t-(1-r_i{)}^t)$,

where $y\left(t\right)$ is the welfare trajectory conditional on no existential catastrophe. The factor $\left(\right(1-r_f{)}^t-(1-r_i{)}^t)$ increases to a maximum value and then decays to 0 as $t$ approaches infinity, so its asymptotic behavior is $O\left(1\right)$. So the trajectory change caused by reducing x-risk is $O\left(y\right(t\left)\right)$, or whatever the asymptotic behavior of $y\left(t\right)$ is.

On the other hand, the amount of good done by changing the trajectory from $y_i\left(t\right)$ to $y_f\left(t\right)$ is

$\sum _{t=1}^{\infty }\left(y_f\right(t)-y_i(t\left)\right)(1-r{)}^t$.

So if you can change the trajectory to a $y_f\left(t\right)$ that will grow asymptotically faster than $y_i\left(t\right)$, e.g. by colonizing space, then this may be more important than reducing x-risk while holding the trajectory constant.

1. ^{^}

   Change in GDP per capita may underestimate the amount of welfare created by technological progress. I'm not sure if this makes the change in average welfare a super-linear growth trend, though.

2. ^{^}

   The Epistemic Challenge to Longtermism (Tarsney, 2020)

EA Forum Wrapped 2024 👀

NYC is adopting ranked-choice voting for the 2021 City Council election. One challenge will be explaining the new voting system, though.

Maybe EA philanthropists should be invest more conservatively, actually

The pros and cons of unusually high risk tolerance in EA philanthropy have been discussed a lot, e.g. here. One factor that may weigh in favor of higher risk aversion is that nonprofits benefit from a stable stream of donations, rather than one that goes up and down a lot with the general economy. This is for a few reasons:

• Funding stability in a cause area makes it easier for employees to advance their careers because they can count on stable employment. It also makes it easier for nonprofits to hire, retain, and develop talent. This allows both nonprofits and their employees to have greater impact in the long run. Whereas a higher but more volatile stream of funding might not lead to as much impact.
• It becomes more politically difficult to make progress in some causes during a recession. For example, politicians may have lower appetite for farm animal welfare regulations and might even be more willing to repeal existing regulations if they believe the regulations stifle economic growth. This makes it especially important for animal welfare orgs to retain funding.

Some rough thoughts on cause prioritization

• I've been tying myself up in knots about what causes to prioritize. I originally came back to effective altruism because I realized I had gotten interested in 23 different causes and needed to prioritize them. But looking at the 80K problem profile page (I am fairly aligned with their worldview), I see at least 17 relatively unexplored causes that they say could be as pressing as the top causes they've created profiles for. I've taken a stab at one of them: making surveillance compatible with privacy, civil liberties, and public oversight.
• I'm sympathetic to this proposal for how to prioritize given cluelessness. But I'm not sure it should dominate my decision making. It also stops feeling like altruism when it's too abstracted away from the object-level problems (other than x-risk and governance).
• I've been seriously considering just picking causes from the 80K list "at random."
  • By this, I mean could just pick a cause from the list that seems more neglected, "speaks to me" in some meaningful way, and that I have a good personal fit for. Many of the more unexplored causes on the 80K list seem more neglected, like one person worked on it just long enough to write one forum post (e.g. risks from malevolent actors).
  • It feels inherently icky because it's not really taking into account knowledge of the scale of impact, and it's the exact thing that EA tells you not to do. But: MIRI calls it quantilizing, or picking an action at random from the top x% of actions one could do. They think it's a promising alternative to expected utility maximization for AI agents, which makes me more confident that it might be a good strategy for clueless altruists too.
  • Some analogies that I think support this line of thinking:
    • In 2013, the British newspaper The Observer  ran a contest between professional investment managers and... a cat throwing a toy at a dartboard to pick stocks. The cat won. According to the efficient market hypothesis, investors are clueless about what investing opportunities will outperform the pack, so they're unlikely to outperform an index fund or a stock-picking cat. If we're similarly clueless about what's effective in the long term, then maybe the stochastic approach is fine.
    • One strategy for dimensionality reduction in machine learning and statistics is to compress a high-dimensional dataset into a lower-dimensional space that's easier to compute with by creating a random projection. Even though the random projection doesn't take into account any information in the dataset (like PCA does), it still preserves most of the information in the dataset most of the time.
• I've also been thinking about going into EA community building activities (such as setting up an EA/public interest tech hackathon) so I can delegate, in expectation, the process of thinking about which causes are promising to other people who are better suited to doing it. If I did this, I would most likely still be thinking about cause prioritization, but it would allow me to stretch that thinking over a longer time scale than if I had to do it all at once before deciding on an object-level cause to work on.
• Even though I think AI safety is a potentially pressing problem, I don't emphasize it as much because it doesn't seem constrained by CS talent. The EA community currently encourages people with CS skills to go into either AI technical safety or earning to give. Direct work applying CS to other pressing causes seems more neglected, and it's the path I'm exploring.

Making specialty meats like foie gras using cellular agriculture could be especially promising. Foie gras traditionally involves fattening ducks or geese by force-feeding them, which is especially ethically problematic (although alternative production methods exist). It could probably be produced by growing liver and fat cells in a medium without much of a scaffold, which would make it easier to develop.

Asking for a friend - there's no dress code for EAG, right?

I lowkey miss the name "shortform" 🙁

It seems like decibels (dB) are a natural unit for perceived pleasure and pain, since they account for the fact that humans and other beings mostly perceive sensations in proportion to the logarithm of their actual strength. (This is discussed at length in "Logarithmic Scales of Pleasure and Pain".)

Decibels are a relative quantity: they express the intensity of a signal relative to another. A 10x difference is 10 dB, a 100x difference is 20 dB, and so on. The "just noticeable difference" in amplitude of sound is ~1 dB, or a ~25% increase. But decibels can also be used in an "absolute" sense by quantifying the ratio of the signal to a reference value. In the case of sound, the reference value is the smallest value that most humans can hear (a sound pressure of 20 micropascals).^[1]

Since pleasure and pain are perceived according to a log scale, the utility of a sensation could be approximated by:

$U\left(S\right)=U_{0}max(0,\log (S/S_0\left)\right)$

where $S$ is the intensity of the sensation, $S_0$ is the smallest perceptible sensation, and $U_0$ is a constant that is positive for pleasurable sensations and negative for painful ones. (This is only an approximation because Fechner's law, the principle that governs logarithmic perception of signals, breaks down for very strong and very weak signals.)

It seems very natural, therefore, to use decibels as the main unit for pleasure and pain, alongside utils for the utility of perceived sensations, as the relationship between decibels and utils is linear. For example, if a utility function is given by

$U\left(D\right)=0.5max(0,D)$

where $D$ is the decibel amount, then we have 10 dB of pleasure = 5 utils, 20 dB = 10 utils, 30 dB = 15 utils, and so on.

1. ^{^}

   https://en.wikipedia.org/wiki/Decibel#Acoustics_2

There should be a directory of EA co-living spaces, if there isn't already. The EA Hub would be a good place for it.

I think some of us really need to create op-eds, videos, etc. for a mainstream audience defending longtermism. The Phil Torres pieces have spread a lot (people outside the EA community have shared them in a Discord server I moderate, and Timnit Gebru has picked them up) and thus far I haven't seen an adequate response.

AOC's Among Us stream on Twitch nets $200K for coronavirus relief

"We did it! $200k raised in one livestream (on a whim!) for eviction defense, food pantries, and more. This is going to make such a difference for those who need it most right now." — AOC's Tweet

Video game streaming is a popular way to raise money for causes. We should use this strategy to fundraise for EA organizations.

We're probably surveilling poor and vulnerable people in developing and developed countries too much in the name of aiding them, and we should give stronger consideration to the privacy rights of aid recipients. Personal data about these people collected for benign purposes can be weaponized against them by malicious actors, and surveillance itself can deter people from accessing vital services.

"Stop Surveillance Humanitarianism" by Mark Latonero

Automating Inequality by Virginia Eubanks makes a similar argument regarding aid recipients in developed countries.

Back-of-the-envelope calculations for improving efficiency of public transit spending

The cost of building and maintaining public transportation varies widely across municipalities due to inefficiencies - for example, the NYC Second Avenue Subway has cost $2.14 billion per kilometer to build, whereas it costs an average of $80.22 million to build a kilometer of tunnel in Spain (Transit Costs Project). While many transit advocacy groups advocate for improving quality of public transit service (e.g. Straphangers Campaign in NYC), few advocate for reducing wasteful infrastructure spending.

BOTEC for operating costs

• Uday Schultz writes: "bringing NYCT’s [the NYC subway agency] facility maintenance costs down to the national average could save $1.3 billion dollars per year."
• With a 6% discount rate, this equates to a $21.7 billion net present value. So an advocacy campaign that spent $21.7 million to reduce NYCT's maintenance costs to the national average would yield a 1000x return.
• Things that would make the cost-effectiveness of this campaign higher or lower:
  • (Higher) A lower discount rate would increase the net present value of the benefits
  • (Higher) In theory, we can reduce maintenance costs to even lower levels than the US national average; Western European levels are (I think) lower.
  • (Lower) We might not realize all of the potential efficiency gains for political reasons - e.g. if contractors and labor unions block the best possible reforms.

BOTEC for capital construction costs

• The NYC Second Avenue Subway will be 13.7 km long when completed and cost over $17 billion. Phase 1 of the subway line has been completed, consists of 2.9 km of tunnel, and cost $4.45 billion.
• So the rest of the planned subway line (yet to be built) consists of 10.8 km of tunnel and is expected to cost $12.55 billion, for an average of $1.16 billion per km of tunnel.
• Phase 2 of the subway will be 2.4 km long and cost $6 billion, for an average of $2.5 billion per km of tunnel.
• There will likely be cost overruns in the future, so let's take the average of these two numbers and assume that the subway will cost an average of $1.83 billion/km to build.
• As I stated before, the average cost per km of new tunnel in Spain is $80.22 million (Transit Costs Project). If NYCT could build the rest of the Second Avenue Subway at this cost, it would save $1.75 billion per km of new tunnel, or $18.9 billion overall (since there are 10.8 km of tunnel left to build).

N.B.: These are BOTECs for individual aspects of transit spending, and a transit spending advocacy project would benefit from economies of scope because it would be lobbying for cost reductions across all aspects of public transit, not just e.g. the Second Avenue Subway or operating costs.

See also: "So You Want to Do an Infrastructure Package," Alon Levy's whitepaper on reducing transit infrastructure costs (Niskanen Center, 2021)

I think improving bus systems in the United States (and probably other countries) could be a plausible Cause X.

Importance: Improving bus service would:

• Increase economic output in cities
• Dramatically improve quality of life for low-income residents
• Reduce cities' carbon footprint, air pollution, and traffic congestion

Neglectedness: City buses probably don't get much attention because most people don't think very highly of them, and focus much more on novel transportation technologies like electric vehicles.

Tractability: According to Higashide, improving bus systems is a matter of improving how the bus systems are governed. Right now, I think a nationwide movement to improve bus transit would be less polarizing than the YIMBY movement has been. While YIMBYism has earned a reputation as elitist due to some of its early advocates' mistakes, a pro-bus movement could be seen as aligned with the interests of low-income city dwellers provided that it gets the messaging right from the beginning.

Also, bus systems are less costly to roll out, upgrade, and alter than other public transportation options like trains.