The Nonlinear Library

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: Vote on Interesting Disagreements, published by Ben Pace on November 7, 2023 on LessWrong. Do you have a question you'd like to see argued about? Would you like to indicate your position and discuss it with someone who disagrees? Add poll options to the thread below to find questions with lots of interest and disagreement. How to use the poll Reacts: Click on the agree/disagree reacts to help people see how much disagreement there is on the topic. Karma: Upvote positions that you'd like to read dialogues about. New Poll Option: Add new positions for people to take sides on. Please add the agree/disagree reacts to new poll options you make. The goal is to show people where a lot of interesting disagreement lies. This can be used to find discussion and dialogue topics in the future. 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: Announcing Athena - Women in AI Alignment Research, published by Claire Short on November 7, 2023 on LessWrong. Athena is a new research mentorship program fostering diversity of ideas in AI safety research. We aim to get more women and marginalized genders into technical research and offer the support needed to thrive in this space. Applications for scholars are open until December 3rd, 2023 Apply as a scholar: here Apply as a mentor or speaker: here Financial aid is available for travel expenses for the in-person retreat to those otherwise unable to attend without it Program Structure A 2-month hybrid mentorship program for women looking to strengthen their research skills and network in technical AI safety research beginning in January 2024. This includes 1-week in-person retreat in Oxford, UK followed by a 2-month remote mentorship by established researchers in the field, with networking and weekly research talks. Athena aims to equip women with the knowledge, skills, and network they need to thrive in AI safety research. We believe that diversity is a strength, and hope to see this program as a stepping stone towards a more diverse and inclusive AI safety research field. This program is designed to offer mentorship opportunities to technically qualified women who are early in their AI safety research careers or looking to transition into the field by connecting them with experienced mentors, resources to upskill, networking, and a supportive community. Who should apply? Women and people of other marginalized genders that have some research or technical industry experience, and are interested in transitioning to AI Alignment research or have a bit of experience in the Alignment field but are looking for more support. We encourage those with a non-traditional background to apply and welcome interdisciplinary work in this field. Application process Submit the online application questions here Complete an interview with the founder and one other AI safety researcher Additional possible interviews with mentor Questions? Email: claire@researchathena.org Why are we doing thisThe current culture requires a shift to retain a diverse set of qualified researchers Athena aims to increase the number of women pursuing careers in AI alignment, which is currently a male-dominated field with a very specific culture that can initially come across as unwelcoming to those that aren't traditionally represented here. Women may have different hurdles to cross than their male counterparts such as implicit and explicit bias, different family and home obligations, unwanted approaches for romantic relationships by male colleagues, isolation, and a lack of representation. We want to take steps to shift the current culture to one that values diversity and inclusivity through recruiting qualified women into the field through extended outreach, providing technical mentorship with an experienced researcher, creating a targeted support structure during the program, and continued role and grant placement support after the program. There are also opportunities for networking and collaboration within the larger research ecosystem. Having diverse research teams and ideas is valuable for AI Alignment research Research has consistently shown that diverse teams produce more innovative solutions. When we have a diverse group of people, including women, working on AI alignment, we are more likely to come up with comprehensive and holistic solutions that consider a wide range of perspectives and people. When more women participate in traditionally male-dominated fields like the sciences, the breadth of knowledge in that area usually grows, a surge in female involvement directly correlates with advancements in understanding [ 1]. Since there is a lack of women in this field, Athena aims to prepare women ...

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: AMA: Earning to Give, published by jefftk on November 7, 2023 on LessWrong. This week the Effective Altruism Forum is running an Effective Giving Spotlight, and they asked if I could post an Ask Me Anything (AMA) on my experience earning to give. Some background: I was earning to give from 2009 to 2022, except for a few months in 2017 when I worked on expanding access to the financial system in Ethiopia and looking into AI risk disagreements. I've been a Giving What We Can member since 2013, making a public pledge to continue with effective giving. For most of this time my wife and I were donating 50% of our pre-tax income, for a total of $2.1M. This has been about 50-50 between trying to help the EA community grow into the best version of itself and funding global poverty reduction (details, thoughts, more recent but still obsolete thoughts). In 2016 I gave a EA Global talk (transcript) on earning to give, which gives more background on the idea and how I've neen thinking about it. That's a lot of links, and it's fine to ask questions even if you haven't read any of them! I'm happy to take questions on earning to give, or anything else within EA. Here are some example questions I'd be happy to answer if there's interest: Where do individual donors earning to give have an advantage over foundations and funds? How should you decide whether to use a fund? How have I thought about how much to donate? How much is enough? Why did I stop earning to give? Why am I still donating some even though I'm funded by EA donors? Feel free to comment on any platform, but if you're having trouble deciding then the EA Forum post is ideal. Comment via: the EA Forum 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: Varieties of minimalist moral views: Against absurd acts, published by Teo Ajantaival on November 7, 2023 on The Effective Altruism Forum. (A standalone part of Minimalist Axiologies: Alternatives to 'Good Minus Bad' Views of Value .) 1. Introduction What are minimalist views? Minimalist views of value (axiologies) are evaluative views that define betterness solely in terms of the absence or reduction of independent bads. For instance, they might roughly say, "the less suffering, violence, and violation, the better". They reject the idea of weighing independent goods against these bads, as they deny that independent goods exist in the first place. Minimalist moral views are views about how to act and be that include a minimalist view of value, instead of an offsetting ('good minus bad') view of value. They reject the concept of independently positive moral value, such as positive virtue or pleasure that could independently counterbalance bads.[1] Minimalist views are sometimes alleged - at least in their purely consequentialist versions - to recommend absurd acts in practice, such as murdering individuals to prevent their suffering, or opposing life-extending interventions lest we prolong suffering. My aim in this essay is to broadly outline the various reasons why the most plausible and well-construed versions of minimalist moral views - including their purely consequentialist versions - do not recommend such acts.Sequence recap For context, below is a chronological recap of the present series on minimalist views so far. 1. " Positive roles of life and experience in suffering-focused ethics": Even if we assume a purely suffering-focused view, it's wise to recognize the highly positive and often necessary roles that various other things may have for the overall goal of reducing suffering. These include the positive roles of autonomy, cooperation, nonviolence, as well as our personal wellbeing and valuable skills and experiences. Suffering-focused moral views may value these things for different reasons, but not necessarily any less, than do other moral views. 2. " Minimalist axiologies and positive lives": Minimalist axiologies define goodness in entirely relational or 'instrumental' terms, namely in terms of the minimization of bads such as suffering. These views avoid many problems in population ethics, yet the minimalist notion of (relationally) positive value is entirely excluded by the standard, restrictive assumption of treating lives as isolated value-containers. Minimalist views become more intuitive when we adopt a relational view of the overall value of individual lives, that is, when we don't track only the causally isolated "contents" of these lives, but also their (often far more significant) causal roles. 3. " Peacefulness, nonviolence, and experientialist minimalism": For purely experience-focused and consequentialist versions of minimalist views, an ideal world would be any perfectly peaceful world, including an empty world. When it comes to theoretical implications about the cessation and replacement of worlds, one can reasonably argue that offsetting ('good minus bad') views have worse implications than do minimalist views. Zooming out from unrealistic thought experiments, it's crucial to be mindful of the gap between theory and practice, of the pitfalls of misconceived consequentialism, and of how minimalist consequentialists have strong practical reasons to pursue a nonviolent approach and to cooperate with people who hold different values. 4. " Minimalist extended very repugnant conclusions are the least repugnant": It has been argued that certain "repugnant conclusions" are an inevitable feature of any plausible axiology. Yet based on a 'side-by-side' comparison of different views, it appears that offsetting views share all the most "repugnant" ...

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: AI Alignment Research Engineer Accelerator (ARENA): call for applicants, published by TheMcDouglas on November 7, 2023 on The Effective Altruism Forum. TL;DR Apply here for the third iteration of ARENA (Jan 8th - Feb 2nd)! Introduction We are excited to announce the third iteration of ARENA (Alignment Research Engineer Accelerator), a 4-week ML bootcamp with a focus on AI safety. Our mission is to prepare participants for full-time careers as research engineers in AI safety, e.g. at leading organizations or as independent researchers. The program will run from January 8th - February 2nd 2024[1], and will be held at the offices of the London Initiative for Safe AI. These offices are also being used by several safety orgs (BlueDot, Apollo, Leap Labs), as well as the current London MATS cohort, and several independent researchers. We expect this to bring several benefits, e.g. facilitating productive discussions about AI safety & different agendas, and allowing participants to form a better picture of what working on AI safety can look like in practice. ARENA offers a unique opportunity for those interested in AI safety to learn valuable technical skills, work in their own projects, and make open-source contributions to AI safety-related libraries. The program is comparable to MLAB or WMLB, but extends over a longer period to facilitate deeper dives into the content, and more open-ended project work with supervision. For more information, see our website. Outline of Content The 4-week program will be structured as follows: Chapter 0 - Fundamentals Before getting into more advanced topics, we first cover the basics of deep learning, including basic machine learning terminology, what neural networks are, and how to train them. We will also cover some subjects we expect to be useful going forwards, e.g. using GPT-3 and 4 to streamline your learning, good coding practices, and version control.Note - participants can optionally not attend the program during this week, and instead join us at the start of Chapter 1, if they'd prefer this option and if we're confident that they are already comfortable with the material in this chapter. Topics include: PyTorch basics CNNs, Residual Neural Networks Optimization (SGD, Adam, etc) Backpropagation Hyperparameter search with Weights and Biases GANs & VAEsDuration: 5 days Chapter 1 - Transformers & Interpretability In this chapter, you will learn all about transformers, and build and train your own. You'll also study LLM interpretability, a field which has been advanced by Anthropic's Transformer Circuits sequence, and open-source work by Neel Nanda. This chapter will also branch into areas more accurately classed as "model internals" than interpretability, e.g. recent work on steering vectors. Topics include: GPT models (building your own GPT-2) Training and sampling from transformersTransformerLensIn-context Learning and Induction HeadsIndirect Object IdentificationSuperpositionSteering VectorsDuration: 5 days Chapter 2 - Reinforcement Learning In this chapter, you will learn about some of the fundamentals of RL, and work with OpenAI's Gym environment to run their own experiments. Topics include: Fundamentals of RLVanilla Policy GradientProximal Policy Gradient RLHF (& finetuning LLMs with RLHF) Gym & Gymnasium environmentsDuration: 5 days Chapter 3 - Paper Replications We will conclude this program with paper replications, where participants will get guidance and mentorship while they replicate a paper containing material relevant to this course. This should draw on much of the skills and knowledge participants will have accumulated over the last 3 weeks.Duration: 5 days Below is a diagram of the curriculum as a whole, and the dependencies between sections. Note that this may change slightly in the lead-up to the program.Here is som...

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: Scalable And Transferable Black-Box Jailbreaks For Language Models Via Persona Modulation, published by Soroush Pour on November 7, 2023 on The AI Alignment Forum. Paper coauthors: Rusheb Shah, Quentin Feuillade--Montixi, Soroush J. Pour, Arush Tagade, Stephen Casper, Javier Rando. Motivation Our research team was motivated to show that state-of-the-art (SOTA) LLMs like GPT-4 and Claude 2 are not robust to misuse risk and can't be fully aligned to the desires of their creators, posing risk for societal harm. This is despite significant effort by their creators, showing that the current paradigm of pre-training, SFT, and RLHF is not adequate for model robustness. We also wanted to explore & share findings around "persona modulation"[1], a technique where the character-impersonation strengths of LLMs are used to steer them in powerful ways. Summary We introduce an automated, low cost way to make transferable, black-box, plain-English jailbreaks for GPT-4, Claude-2, fine-tuned Llama. We elicit a variety of harmful text, including instructions for making meth & bombs. The key is *persona modulation*. We steer the model into adopting a specific personality that will comply with harmful instructions.We introduce a way to automate jailbreaks by using one jailbroken model as an assistant for creating new jailbreaks for specific harmful behaviors. It takes our method less than $2 and 10 minutes to develop 15 jailbreak attacks. Meanwhile, a human-in-the-loop can efficiently make these jailbreaks stronger with minor tweaks. We use this semi-automated approach to quickly get instructions from GPT-4 about how to synthesise meth . Abstract Despite efforts to align large language models to produce harmless responses, they are still vulnerable to jailbreak prompts that elicit unrestricted behaviour. In this work, we investigate persona modulation as a black-box jailbreaking method to steer a target model to take on personalities that are willing to comply with harmful instructions. Rather than manually crafting prompts for each persona, we automate the generation of jailbreaks using a language model assistant. We demonstrate a range of harmful completions made possible by persona modulation, including detailed instructions for synthesising methamphetamine, building a bomb, and laundering money. These automated attacks achieve a harmful completion rate of 42.5% in GPT-4, which is 185 times larger than before modulation (0.23%). These prompts also transfer to Claude 2 and Vicuna with harmful completion rates of 61.0% and 35.9%, respectively. Our work reveals yet another vulnerability in commercial large language models and highlights the need for more comprehensive safeguards. Full paper You can find the full paper here on arXiv https://arxiv.org/abs/2311.03348 . Safety and disclosure We have notified the companies whose models we attacked We did not release prompts or full attack details We are happy to collaborate with researchers working on related safety work - please reach out via correspondence emails in the paper. Acknowledgements Thank you to Alexander Pan and Jason Hoelscher-Obermaier for feedback on early drafts of our paper. ^ Credit goes to @Quentin FEUILLADE--MONTIXI for developing the model psychology and prompt engineering techniques that underlie persona modulation. Our research built upon these techniques to automate and scale them as a red-teaming method for jailbreaks. 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: On the UK Summit, published by Zvi on November 7, 2023 on LessWrong. In the eyes of many, Biden's Executive Order somewhat overshadowed the UK Summit. The timing was unfortunate. Both events were important milestones. Now that I have had time, here is my analysis of what happened at the UK Summit. As is often the case with such events, there was a lot of talk relative to the amount of action. There was a lot of diplomatic talk, talk of that which everyone agrees upon, relative to the amount of talk of real substance. There were days of meetings that resulted in rather unspicy summaries and resolutions. The language around issues that matter most was softened, the actual mission in danger of being compromised. And as usual, the net result was reason for optimism, a net highly positive event versus not having it, while also in some ways being disappointing when compared to what might have been. A declaration was signed including by China, but it neglected existential risk. Sunak's words on AI were not as strong as his words have been previously. We got promises for two additional summits, in South Korea and France. Given that, I am willing to declare this a success. One area of strong substance was the push for major AI labs to give substantive safety policies addressing a variety of issues, sometimes largely called Responsible Scaling Policies (RSPs). The biggest labs all did so, even Meta. Now we can examine their responses, know who is being how responsible, and push for better in the future or for government action to fix issues or enshrine progress. This was an excellent development. This post will look at the rest of what happened at the Summit. I will be writing about the RSPs and other safety policies of the labs in a distinct post next week. Looking Back at People's Goals for the Summit and TaskforceJack Clark's proposal from July 5 for what the Foundation Model taskforce might do to evaluate frontier models as its priority, and how it might prioritize that, Simeon's response emphasizing the need for a good way to know whether a proposal is safe enough to allow it to proceed.Navigating AI Risks asked on July 17 what the taskforce should do, advising focus on interventions to impact policy at labs and other governments. Suggested focus was risk assessment methodology, demonstrating current risks and assessing current state of the art models, and to avoid direct alignment work.Lennart Heim's (GovAI) July 10 proposal of what the summit should try to accomplish, which he reviewed after the summit.Matt Clifford from the PM's office shared on September 10 their objectives for the summit: A shared understanding of the risks posed by frontier AI and the need for action, a forward process for international collaboration, measures for organizations, finding areas for safety collaboration and showcasing how safe AI development can enhance global good. AI Safety Summit AgendaWhat has the UK Taskforce been up to in advance of the summit (report)? Ian Hogarth (Chair UK AI Frontier Model Taskforce): The Taskforce is a start-up inside government, delivering on the mission given to us by the Prime Minister: to build an AI research team that can evaluate risks at the frontier of AI. We are now 18 weeks old and this is our second progress report. The frontier is moving very fast. On the current course, in the first half of 2024, we expect a small handful of companies to finish training models that could produce another significant jump in capabilities beyond state-of-the-art in 2023. As these AI systems become more capable they may augment risks. An AI system that advances towards expert ability at writing software could increase cybersecurity threats. An AI system that becomes more capable at modelling biology could escalate biosecurity threats. We believe it is critical that f...

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: Numerical Breakdown of 47 1-on-1s as an EAG First-Timer (Going All Out Strategy), published by Harry Luk on November 7, 2023 on The Effective Altruism Forum. tl;dr Just attended my first ever EA Global conference (EAG Boston last week) and I have nothing but positive things to say. In total, I had about 47 one-on-one conversations depending on how you count the informal 1:1s (43 scheduled via SwapCard, while the other noteworthy conversations happened at meetups, the organization fair, office hours and unofficial satellite events). I came into the conference with an open mind, wanting to talk to others who are smarter than me, more experienced than me, and experts in their own domain. I invited redteaming of our nonprofit StakeOut.AI's mission/TOC, and gathered both positive and negative feedback throughout EAG. I came out of the conference with new connections, a refined strategy for our nonprofit startup going forward and lots of resources. I am so grateful for everyone that met with me (as I'm a small potato who at many times felt out of his depth during EAG, and likely one of the most junior EAs attending). I thank all the organizers, volunteers, helpers, speakers and attendees who made the event a huge success. The post below goes over The Preparation, the Statistics and Breakdown, why consider going all out at an EAG, 12 Practical Tips for Doing 30+ 1:1s and potential future improvements. The Preparation To be honest, as a first-time attendee, I really didn't know what to expect nor how to prepare for the conference. I had heard good things and was recommended to go by fellow EAs, but I had my reservations. Luckily, an email titled "Join us for an EAG first-timers online workshop!" by the EA Global Team came to the rescue. Long story short, I highly recommend anyone new to EAG to attend the online workshop prior to the conference if you want to make your first EAG a success. Few highlights I will note here: Watch this presentation from 2022's San Francisco EAG that outlines how you can get the most out of the event Take your time and fill out this EA Conference: Planning Worksheet for a step-by-step guide on conference planning, including setting your EAG goals and expectations Also fill out the career planning worksheet (if relevant): EA Conference: Career Plan Requesting 1:1s Pre-conference I was quite hesitant at first about introducing myself on SwapCard and trying to schedule 1:1s. This all changed after watching the presentation and attending the "Join us for an EAG first-timers online workshop!" virtual event. Something that was repeated over and over again from this presentation, the online workshop, and talking to others is the value of the 1:1s. People told me most sessions will be recorded and hence can be watched later, but having the 1:1s is where the true value is at EAG. After hearing it from so many people, I made 1:1s a core part of my conference planning and did not regret it. As I'm writing this after the conference, I can see why 1:1s are said to be the true value of EAG. I estimate that 80% (maybe even closer to 90%, I would know better after I sort through the notes) of the 1:1 conversations I had were beneficial and had a positive impact on either me or the direction of our nonprofit, StakeOut.AI. How Many 1:1s? In terms of how many 1:1s, here is the range I gathered from different sources: Attendees will typically have four to ten 1:1s Getting to 20 1:1s is a great number Having 30 1:1s is amazing but very tiring Someone reached 35 1:1s once, and that was insane Since I wanted to maximize my EAG experience, I set the goal of 30 and started reaching out via SwapCard one week before the conference. Reach Out Early The main reason for starting early is because everyone is busy at the conferences, and everyone is trying to optimize their sch...

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: How Rethink Priorities is Addressing Risk and Uncertainty, published by Marcus A Davis on November 7, 2023 on The Effective Altruism Forum. This post is part of Rethink Priorities' Worldview Investigations Team's CURVE Sequence: "Causes and Uncertainty: Rethinking Value in Expectation." The aim of this sequence is twofold: first, to consider alternatives to expected value maximization for cause prioritization; second, to evaluate the claim that a commitment to expected value maximization robustly supports the conclusion that we ought to prioritize existential risk mitigation over all else. Introduction RP has committed itself to doing good. Given the limits of our knowledge and abilities, we won't do this perfectly but we can do this in a principled manner. There are better and worse ways to work toward our goal. In this post, we discuss some of the practical steps that we're taking to navigate uncertainty, improve our reasoning transparency, and make better decisions. In particular, we want to flag the value of three changes we intend to make: Incorporating multiple decision theories into Rethink Priorities' modeling More rigorously quantifying the value of different courses of action Adopting transparent decision-making processes Using Multiple Decision Theories Decision theories are frameworks that help us evaluate and make choices under uncertainty about how to act.[1] Should you work on something that has a 20% chance of success and a pretty good outcome if success is achieved, or work on something that has a 90% chance of success but only a weakly positive outcome if achieved? Expected value theory is the typical choice to answer that type of question. It calculates the expected value (EV) of each action by multiplying the value of each possible outcome by its probability and summing the results, recommending the action with the highest expected value. But because low probabilities can always be offset by corresponding increases in the value of outcomes, traditional expected value theory is vulnerable to the charge of fanaticism, "risking arbitrarily great gains at arbitrarily long odds for the sake of enormous potential" (Beckstead and Thomas, 2021). Put differently, it seems to recommend spending all of our efforts on actions that, predictably, won't achieve our ends.Alternative decision theories have significant drawbacks of their own, giving up one plausible axiom or another. The simple alternative is expected value maximization but with very small probabilities rounded down to zero. This gives up the axiom of continuity, which suggests for a relation of propositions A B C, that there exists some probability that would make you indifferent between B and a probabilistic combination of A and C. This violation causes some weird outcomes where, say, believing the chance of something is 1 in 100,000,000,000 can mean an action gets no weight but believing it's 1.0000001 in 100,000,000,000 means that the option dominates your considerations if the expected value upon success is high enough, which is a kind of attenuated fanaticism. There are also other problems like setting the threshold for where you should round down.[2] Alternatively, you could go with a procedure like weighted-linear utility theory (WLU) (Bottomley and Williamson, 2023), but that gives up the principle of homotheticity, which involves indifference to mixing a given set of options with the worst possible outcome. Or you could go with a version of risk-weighted expected utility (REU) (Buchak, 2013) and give up the axiom of betweenness which suggests the order in which you are presented information shouldn't alter your conclusions.[3] It's very unclear to us, for example, that giving up continuity is preferable to giving up homotheticity, and neither REU or WLU really logically eliminate issues 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: Incorporating and visualizing uncertainty in cost effectiveness analyses: A walkthrough using GiveWell's estimates for StrongMinds, published by Jamie Elsey on November 7, 2023 on The Effective Altruism Forum. A common first step towards incorporating uncertainty into a cost effectiveness analysis (CEA) is to express not just a point estimate (i.e., a single number) for an input to the CEA, but to provide some indicator of uncertainty around that estimate. This might be termed an optimistic vs. pessimistic scenario, or as the lower and upper bounds of some confidence or uncertainty interval. A CEA is then performed by combining all of the optimistic inputs to create an optimistic final output, and all of the pessimistic inputs to create a final pessimistic output. I refer to this as an 'interval-based approach'. This can be contrasted with a fuller 'probabilistic approach', in which uncertainty is defined through the use of probabilistic distributions of values, which represent the range of possibilities we believe the different inputs can take. While many people know that a probabilistic approach circumvents shortcomings of an interval-based approach, they may not know where to even begin in terms of what different distributions are possible, or the kinds of values they denote. I hope to address this in the current post and the accompanying application. Concretely, I aim to: Show how performing a CEA just using an interval-based approach can lead to a substantial overestimation of the uncertainty implied by one's initial inputs, and how using a probabilistic approach can correct this while also enabling additional insights and assessments Introduce a new tool I have developed - called Distributr - that allows users to get more familiar and comfortable with a range of different distributions and the kinds of values they imply Use this tool to help generate a probabilistic approximation of the inputs GiveWell used in their assessment of Strongminds,[1] and perform a fuller probabilistic assessment based upon these inputs Show how this can be done without needing to code, using Distributr and a simple spreadsheet I ultimately hope to help the reader to feel more capable and confident in the possibility of incorporating uncertainty into their own cost effectiveness analyses. Propagating uncertainty and the value of moving beyond an interval-based approach Cost effectiveness analysis involves coming up with a model of how various different factors come together to determine both how effective some intervention is, and the costs of its delivery. For example, when we think about distributing bed nets for malaria prevention, we might consider how the cost of delivery can vary across different regions, how the effects of bed net delivery will depend on the likelihood that people use the bed nets for their intended purpose, and the probability that recipients will install the bed nets properly. These and other factors all come together to produce an estimate of the cost effectiveness of an intervention, which will depend on the values we ascribe to the various inputs. One way that a researcher might seek to express uncertainty in these inputs is by placing reasonable upper and lower bounds on their estimates for each of them. The researcher might then seek to propagate this uncertainty in the inputs into the anticipated uncertainty in the outputs by performing the same cost effectiveness calculations as they did on their point estimates on their upper bounds, and on their lower bounds, thereby producing corresponding upper and lower bounds on the final cost effectiveness. An example of an interval-based approach is GiveWell's assessment of Happier Lives Institute's CEA for StrongMinds. The purpose of this post is not to provide an independent evaluation of Strongminds, nor i...