Frequently asked questions

It's difficult to model reality empirically, and harder still to reach certainty on the philosophical questions involved, like how to weigh different types of outcomes, how to compare welfare across species, or how to handle uncertain long-term effects. These aren't questions with obvious answers, and reasonable people disagree. Our approach is to make these uncertainties explicit rather than paper over them with false confidence.

Very little. Most effective altruism prioritization research focuses on comparisons within causes rather than across them. GiveWell provides rigorous cost-effectiveness estimates for global health charities, but there's no equivalent that systematically compares global health interventions to animal welfare or catastrophic risk interventions. In practice, many allocation decisions across causes are made on a surprisingly unsystematic basis, which is part of why we built our cross-cause model. Learn more in our post on why and how to prioritize.

"Common sense" is excellent for identifying that global poverty is bad or that animal torture is wrong, but it fails completely at making trade-offs. Common sense cannot tell you how many resources to shift from malaria prevention to pandemic preparedness. These decisions require explicit weights and measures, because without them, you are likely relying on "My Favorite Theory", acting as if your most plausible guess is 100% certain, which is a significant strategic error. Learn more on how cause prioritization can go wrong.

This approach has intuitive appeal but faces several problems. First, cost-effectiveness varies enormously within cause areas, a cause can score highly on scale, tractability, and neglectedness yet contain interventions ranging from highly effective to net-negative. Second, diminishing returns mean the best opportunity at $100,000 may not be the best at $500,000. Third, meeting the bar for confidently declaring one cause definitively "best" requires the same careful modeling we advocate, you can't escape the hard work through sweeping theoretical arguments. Learn more in our post on why and how to prioritize.

Quantifying beliefs, even when highly uncertain, prevents "vague thinking." For example, there is a massive mathematical difference between a 1% and a 0.001% chance of success in a cause like wild animal welfare. Until we put a number on it, we don't truly know what we believe or what the implications of those beliefs are. Prior research shows that this discipline significantly improves the accuracy of thinking compared to purely qualitative "gut feelings." Learn more about benefits of modeling.

Research shows that in many fields, simple shortcuts (like "Take the Best") work well. However, philanthropy lacks fast feedback loops. If you make a bad cause-prioritization decision, you might not know for decades. Cause prioritization involves many complex factors that interact in counterintuitive ways that are hard to account for without an explicit model. Learn more on how can cause prioritization go wrong.

Deciding how to allocate resources based on high-level characteristics of causes is likely to mislead. Cause prioritization involves many interacting complexities: cost-effectiveness across different funding levels, how to compare very different outcomes (saving a life vs. increasing income, helping humans vs. helping animals), and how to weigh high-probability, modest impact against low-probability, large impact. Holding all of this in your head isn't plausible. Explicit models can't capture everything, but they are an essential input for decision-making when allocating across or within areas. Importantly, implicit models are less transparent–obscuring weaknesses–and have less capacity to account for the large number of complex, interacting considerations that high-stakes resource allocation decisions involve. Learn more about benefits of modeling.

No. Our ultimate recommendations reflect our best model, informed by our best judgment, nearly a decade of cause prioritization research at Rethink Priorities, and input from domain experts. While the model rigorously accounts for many relevant factors there are important caveats and factors not captured by the model, which you can view below. Generally, we think explicit modeling and applying careful judgment are likely to produce significantly better decisions than relying on informal estimates.

We think several widely-used shortcuts often mislead donors. These include: assuming that a single factor about an intervention means it should win; treating scale as the decisive consideration; acting as though AI uncertainty reduces the value of all non-AI interventions to zero; and categorizing very different interventions as equivalent based on simple metrics like tractability or neglectedness. These heuristics feel intuitive but often produce worse results than careful modeling.

We explicitly model four types of uncertainty:

• Empirical uncertainty (uncertainty about facts): How effective is an intervention? What does it actually cost per outcome? Rather than relying on single point estimates, we model ranges that reflect our confidence in the underlying data. See Jamie Elsey's post on incorporating uncertainty into GiveWell's analysis.
• Moral uncertainty (uncertainty about values): Which ethical framework is correct? Rather than betting everything on one worldview (e.g., prioritizing only humans, or only future generations), the model weights multiple moral perspectives according to how plausible you find each. See our work on the moral parliament.
• Meta-normative / decision-theoretic uncertainty (how to handle uncertainty): Should you maximize expected value? Be risk-averse? Weight worst-case scenarios more heavily? The model allows for different attitudes toward risk and time (e.g., how much to discount distant future impacts). See Causes and uncertainty: Rethinking value in expectation and Philosophical Uncertainty: Aggregation Methods Disagree.

Key limitations include:

• Dependence on fund inputs: The model relies on information from funds themselves, which may change as their plans evolve.
• Simplified diminishing returns: The model doesn't distinguish between interventions that decline in effectiveness because they might backfire versus those where success probability decreases.
• Difficulty modeling far-future impact: Assessing long-term or extinction risk reduction is inherently uncertain, and input choices can significantly swing GCR allocations.
• Sensitivity to animal welfare parameters: Small changes in moral weights for invertebrates can substantially shift the animal welfare allocation due to the large numbers involved.
• Limited modeling of interaction effects: The model doesn't fully capture how donors, grantmakers, and grant recipients influence each other's impact.

The model doesn't currently cover all interventions or inputs we consider relevant. On the intervention side, we hope to include political giving to democracy preservation later this year, but it's not yet incorporated. Giving to political candidates is excluded entirely, as a 501(c)(3), Rethink Priorities is legally prohibited from supporting or opposing candidates or parties, including any advice that could be construed as favoring a particular candidate. The model also doesn't cover "meta" causes aimed at improving giving decisions or expanding the number of people working on priority causes. On the philosophical and empirical side, several inputs aren't yet incorporated: the model doesn't account for virtue ethics, flow-through effects (the indirect or downstream impacts of interventions), or interaction effects between donors, grantmakers, and grant recipients. Most inputs currently use point estimates rather than full probability distributions. We're prioritizing the addition of flow-through effects and distributions later this year, and we plan to expand the model's coverage more broadly as our research develops.

Intuition often fails to account for how the 100th million dollar does less good than the 1st million. Our model allows us to test specific mathematical curves, like logarithmic diminishing returns, to see exactly where cost-effectiveness might drop (e.g., a 100x drop after $100M). This level of precision is impossible to hold in your head while simultaneously weighing other factors like moral weights or success probabilities. Learn more about benefits of modeling.

We recognize the risk of "misleading precision." However, a good model doesn't just give a single number (a point estimate); it shows a range of possibilities. If a model's output is highly uncertain, it's usually because the reality it is tracking is complex. Knowing exactly "what we don't know" is more valuable for high-stakes giving than using a simple heuristic that provides a clear but potentially wrong answer. Read In defence of modeling to learn more.

We've built internal models to estimate the cost per unit of impact for major activities supported by each fund. We'll use these to estimate the impact achieved with granted resources. We recognize this approach has limitations. Not all activities will be captured, and impact tracking is more straightforward for some areas (like global health) than others. We're committed to improving our methods over time.

Because different ethical worldviews (like Utilitarianism vs. Deontology) often disagree on how to measure "good," we use a "weighted average" of several aggregation methods. Rather than picking one "favorite" theory or a simple majority vote, which can lead to "fanaticism" where one extreme view hijacks the budget, we prioritize Bargaining Methods (like Nash Bargaining). This approach treats different moral theories like delegates in a "moral parliament" who must negotiate a fair split of resources. By weighting these different methods, we ensure the fund's allocations remain stable, ethically humble, and resistant to any single philosophical outlier. Learn more about aggregation methods.

We embrace a default attitude of intellectual skepticism. Unlike science, philosophical "experiments" cannot be verified in a lab, and historically, many philosophical positions have eventually been revised or discarded. Because of this, we avoid "betting the farm" on any single contentious claim about the far future or specific decision theories. Our fund management strategy is designed to accept uncertainty. We provide a diversified portfolio that achieves high impact under many different plausible versions of the truth, rather than chasing a single, fragile "optimum" that might rely on a flawed assumption. Learn more on why we should be uncertain about cause prioritization.

No. Our cross-cause model is a living tool. While it represents our best current methodology, it doesn't yet capture every factor relevant to estimating impact. For example, we're not currently modeling certain interventions, such as alternative proteins, which we intend to add.

The pace and depth of refinement will scale with the fund's size. As donation volume increases, the value of marginal improvements grows correspondingly. We'll invest in model development in proportion to the resources we steward. If you're interested in supporting this work, contact us at fund@rethinkpriorities.org.

We aim to disburse donations quickly when we're confident that recipient funds can deploy them to high-impact opportunities. When a fund has strong near-term grantmaking opportunities with reasonably well-characterized cost-effectiveness, we disburse within four weeks of receiving your donation.

If total donations reach a level where one or more recipient funds would face significant diminishing returns, we won't disburse the full amount at once. Instead, we disburse in tranches, consult with fund managers, and update our cost-effectiveness data before proceeding. This ensures funds are directed to the right places in the right amounts.

For some funds, the recommended allocation could be several multiples of their current budget, and they couldn't deploy the full amount within a few months. Holding back a portion allows us to gather information about actual opportunities rather than relying solely on projections of future spending. We strike a balance between capturing near-term opportunities and making better-informed decisions.

We aim to avoid two key failure modes: overfunding (directing resources to funds that have raised more than they can deploy effectively) and crowding out aligned donors (granting to funds that would otherwise raise equivalent funds from less mission-aligned sources). Where we anticipate either concern, we adjust the timing and size of our grant recommendations.

While we update our strategy based on economic returns and philanthropic opportunities, we suggest three general "rules of thumb" for donors to avoid the failure modes of extreme strategies:

• Do not donate (to a specific opportunity, not simply a DAF) less than 4% per year.
• Do not plan on a full depletion within 10 years.
• Do not plan on a >95% depletion within 25 years.

Learn more on how to determine the optimal pace of charitable spending.

Saving for the future isn't a "risk-free" move. We explicitly model several threats to long-term capital, including:

• Economic Volatility: The risk of market crashes or AI-specific stock bubbles.
• Appropriation: The non-trivial risk that unspent money could be lost to political actors or misaligned foundations over decades.
• Value Drift: The psychological risk that a donor's motivations or effectiveness might decline over time.
• Diminishing Returns: The possibility that in a future where everyone has saved, the marginal value of your dollar will be much lower than it is today.

Learn more on how to determine the optimal pace of charitable spending.

Our models explore several AI scenarios. If we expect a "Productivity Shock", where AI breakthroughs suddenly make saving a life 10x cheaper, it is rational to save more now to deploy a much larger "pot" of money once that efficiency kicks in. However, if we expect AI to disrupt the economy and lower financial returns, the model recommends spending sooner. Learn more on spending optimally in a changing world.

Yes. We report on disbursement timing and rationale so donors can see how pooled funds have been allocated.

If you want your contribution deployed within a specific window, contact us at fund@rethinkpriorities.org to discuss options.

Across 501(c)(3) funds:

• Global Health: 18%
  • GiveWell All Grants Fund: 13%
  • Lead Exposure Action Fund: 5%
• Animal Welfare: 46%
  • EA Animal Welfare Fund: 13%
  • TNF - General Farm Animal Fund: 17%
  • TNF - Cage-Free Fund: 16%
• Global Catastrophic Risk: 36%
  • Sentinel Bio: 9%
  • Longview Philanthropy’s Nuclear Weapons Policy Fund: 4%
  • Longview Philanthropy’s Frontier AI Fund: 23%

This reflects our best all-things-considered judgment based on a decade of cause prioritization research.

We’re preparing a detailed sensitivity analysis that explores how different answers affect the recommended allocations. Check back soon.

The allocation is based on explicit modeling of empirical, moral, and aggregation uncertainties across different ethical theories. We believe rigorous modeling produces significantly better decisions than relying on informal estimates.

While the potential scale of AI disruption is massive, "Scale" is only one of many factors. Using scale alone ignores:

• Probability of Success: Many AI interventions are highly speculative.
• Immediacy: Programs like GiveDirectly or Against Malaria Foundation help real people today. Even if AI transforms the world in 10 years, it doesn't solve the suffering occurring right now.
• Risk Aversion: A pure "Expected Value" play might suggest going all-in on AI, but even a small amount of risk aversion or "cluelessness" about the far future suggests a more diversified portfolio is more rational.

Learn more on how cause prioritization can go wrong.

Actually, it's the opposite. Qualitative reasoning makes it very easy to subconsciously adjust your logic to reach a preferred conclusion. An explicit model is disciplined; once the numbers and formulas are set, you have to follow where they lead. This makes it much harder to "hide" biases and much easier for others to point out exactly where they think a calculation is wrong. Read more about benefits of modeling.

The model currently excludes: political giving to candidates (legally prohibited for 501(c)(3)s), democracy preservation (we hope to add this), "meta" causes aimed at improving giving decisions, flow-through effects, and virtue ethics. We plan to expand coverage over time.

We maintain ongoing communication with each fund we support to gather updated data on: diminishing returns (how marginal effectiveness changes with recent grantmaking), intervention mix (what types of interventions they're funding), and the funder landscape (other funders in their space and how this affects strategy). We've established relationships with all currently included funds, and each has indicated a willingness to share this information.

We’re continuously updating, improving, and incorporating new information into our cross-cause model. Check this latest updates and improvements doc for more information.

Our approach rests on four pillars: continuously updating our model inputs and grant decisions using current information, expanding coverage to additional funds and cause areas, refining and improving our allocation methodology, and tracking and reporting impact to donors.

We recognize that past performance doesn't guarantee future results. We continuously monitor grantmaker behavior, and if a fund's allocation decisions or organizational conduct diverge from our expectations, or if relevant experts raise concerns, we will reassess our recommendation and, if necessary, adjust the allocation or remove the fund from our portfolio.

We're actively working to expand the Cross-Cause Fund's coverage. We're exploring democracy protection (non-partisan) and climate change over the next few months. Longer term, we're considering areas such as digital minds and AI welfare, economic growth interventions, wild animal welfare, additional biosecurity interventions, and "meta" causes aimed at increasing resources flowing to high-impact charities. We'll communicate any additions through regular updates. For a full list of funds we intend to vet for inclusion, see the Methodology section.

For donors in the RP Cross-Cause DAF, your personal account will automatically track: when the DAF receives funds, when funds are sent to grant recipients, when funds reach recipients' bank accounts, and overall transaction activity (anonymous by default).

We'll make as much information as possible public, including: annual reports summarizing impact and spending, cost-effectiveness estimates for grants made (with transparent methodology), clear benchmarks for success and failure (with public acknowledgment of any mistakes), and a grant pipeline showing recently completed grants, current grants, and forthcoming recommendations.

For RP Cross-Cause DAF donors, we'll aim to provide a personal dashboard that shows the estimated impact attributable to your contributions. For the RP Cross-Cause Fund, we'll publish aggregated impact reporting for the fund as a whole.

RP has received grants from Longview Philanthropy, EA Animal Welfare Fund, Navigation Fund, and Coefficient Giving. RP has done commissioned research for GiveWell and Coefficient Giving. In addition, one RP staff member serves as an advisor to the EA Animal Welfare Fund but is recused from grantmaking decisions involving RP. RP is not involved in any grant-making decisions of these organizations, and RP carries out independent cost-effectiveness analyses of any funds we decide to include in our Cross-Cause Fund.

No. Rethink Priorities does not charge any management fees. 100% of your donation is directed to the recommended grantees.

When allocating hundreds of millions of dollars, the "Value of Information" is extremely high. Even a 0.001% improvement in how effectively that money is spent could justify a million-dollar investment in modeling. Given that different causes can vary in effectiveness by orders of magnitude, the risk of not modeling is far more expensive than the research required to build the model. Read In defence of modelling to learn more.

Contact us at fund@rethinkpriorities.org. We welcome scrutiny and are committed to continuous improvement.