The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - Follow up to medical miracle by Elizabeth
    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: Follow up to medical miracle, published by Elizabeth on November 4, 2022 on LessWrong.
    The response to my medical miracle post has been really gratifying. Before I published I was quite afraid to talk about my emotional response to the problem, and worried that people would strong arm in the comments. The former didn’t happen and the latter did but was overwhelmed by the number of people writing to share their stories, or how the post helped them, or just to tell me I was a good writer. Some of my friends hadn’t heard about the magic pills or realized what a big deal it was, so I got some very nice messages about how happy they were for me.
    However, it also became clear I missed a few things in the original post.
    Conditions to make luck-based medicine work
    In trying to convey the concept of luck-based medicine at all, I lost sight of traits I have that made my slot machine pulls relatively safe. Here is a non-exhaustive list of traits I’ve since recognized are prerequisites for luck based medicine:
    I can reliably identify which things carry noticeable risks and need to be assessed more carefully. I feel like I’m YOLOing supplements, but that’s because it’s a free action to me to avoid combining respiratory depressants, and I know to monitor CYP3A4 enzyme effects. A comment on LessWrong that casually suggested throwing activated charcoal into the toolkit reminded me that not everyone does this as a free action, and the failure modes of not doing so are very bad (activated charcoal is typically given to treat poison consumption. Evidence about its efficacy is surprisingly equivocal, but to the extent it works, it’s not capable of distinguishing poison, nutrients, and medications).
    This suggests to me that an easy lever might be a guide to obvious failure modes of supplements and medications, to lower the barrier to supplement roulette. I am not likely to have the time to do a thorough job of this myself, but if you would like to collaborate please e-mail me ([email protected]).
    A functioning liver. A lot of substances that would otherwise be terribly dangerous are rendered harmless by the human liver. It is a marvel. But if your liver is impaired by alcohol abuse or medical issues, this stops being true. And even a healthy liver will get overwhelmed if you pile the load high enough, so you need to incorporate liver capacity into your plans.
    A sufficiently friendly epistemic environment. If it becomes common and known that everyone will take anything once, the bar for what gets released will become very low. I’m not convinced this can get much worse than it already has, but it is nonetheless the major reason I don’t buy the random health crap facebook advertises to me. The expected value of whatever it is probably is high enough to justify the purchase price, but I don’t want to further corrupt the system.
    Ability to weather small bumps. I’m self-employed and have already arranged my work to trade money for flexibility so this is not a big concern for me, but a few days off your game can be a big deal if your life is inflexible enough. Somehow I feel obliged to say this even though I’ve lost work due to side effects exactly once from a supplement (not even one I picked out; a doctor prescribed it) and at least three times from prescription medications.
    A system for recognizing when things are helping and hurting, and phasing treatments out if they don’t justify the mental load. It’s good to get in the habit of asking what benefits you should see when, and pinning your doctor down on when they will give a medication up as useless.
    Although again, I’ve had a bigger problem with insidious side effects from doctor-initiated prescription meds than I ever have with self-chosen supplements.
    Probably there are other things I do without realizing how critical they ar...
    8 min
  • EA - How CEA approaches applications to our programs by Amy Labenz
    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 CEA approaches applications to our programs, published by Amy Labenz on November 4, 2022 on The Effective Altruism Forum.
    Our programs exist to have a positive impact on the world, rather than to serve the effective altruism community as an end goal. This unfortunately means EAs will sometimes be disappointed because of decisions we’ve made — though if this results in the world being a worse place overall, then we’ve clearly made a mistake. This is one of the hard parts about how EA is both a community and a professional space.
    Naturally, people want to know things like
    Why didn’t I get admitted to a conference, when EA is really important to me and I’m taking actions inspired by EA?
    Why didn’t my friend get admitted to a conference, when they seem like a good applicant?
    We can understand why people would often like feedback on what they could have done differently or what they can try next time to get a better result. Or they just want to know what happened. When we have a specific idea about what would improve someone’s chances (like “you didn’t give much detail on your application, could you add more information?”) we’ll often give it.
    But we get thousands of applications and we don’t think it’s the best use of our staff’s time to give specific feedback about all of them. Often we don’t have constructive feedback to give.
    Many of the things that go into a decision are not easy to pin down — how well we think you understand EA, how we think you’ll add to the social environment, how much we think you’ll benefit from the event given the program we’ve prepared, etc. These things are subjective, and in a lot of cases, reasonable people could disagree about what call to make. There are also cases where we’ll just make mistakes (by our own standards), sometimes in favor of an applicant and sometimes against them.
    How we communicate about our programs
    In responding to public discussion of our programs, sometimes we’ve gotten more in the weeds than we think was ideal. We’ve provided rebuttals or more information about some points but not others, which makes people understandably confused about how much information to expect from us and what the full picture is. It also uses a lot of our staff time. As the EA community grows, we need to adjust how we handle communications with the community.
    What you should expect from us going forward:
    When we think there are significant updates to our programs that the community should know about, or when there seems to be widespread confusion or misunderstanding about a particular topic, we’ll likely write a post (like this one).
    We’ll likely be less involved in the comments or extended public back-and-forth.
    We’ll read much of the public feedback about our programs, and will definitely read feedback you send directly to us (unless it’s extraordinarily long). We won’t respond in-depth to much of the feedback, though.
    Our programs are a work in progress, and we’ll take feedback into account as we try to improve them.
    What we hope you’ll do:
    Feel free to express your criticisms, observations, and advice about our programs. This could be publicly if you think that’s best, or by writing to us directly. Our contact form can be filled out anonymously. Or you can reach specific programs at [email protected], [email protected], or [email protected], or community health’s form.
    In general, we think it’s a good idea to fact-check public criticisms before publishing. If you send us something to fact-check, we’ll try to do so.
    If you think we’ve missed important considerations or information on a decision, like about someone’s application, you can send us additional information.
    On events specifically:
    We try to run a range of events that serve the breadth of the community.
    We recognize that there are people who ar...
    5 min
  • EA - Metaforecast late 2022 update: GraphQL API, Charts, better infrastructure behind the scenes. by NunoSempere
    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: Metaforecast late 2022 update: GraphQL API, Charts, better infrastructure behind the scenes., published by NunoSempere on November 4, 2022 on The Effective Altruism Forum.
    tl;dr: Metaforecast is a search engine and an associated repository for forecasting questions. Since our last update, we have added a GraphQL API, charts, and dashboards. We have also reworked our infrastructure to make it more stable.
    New API
    Our most significant new addition is our GraphQL API. It allows other people to build on top of our efforts. It can be accessed on metaforecast.org/api/graphql, and looks similar to the EA Forum's own graphql api.
    To get the first 1000 questions, you could use a query like:
    title
    url
    description
    options { name
    probability
    } qualityIndicators { numForecasts
    stars
    } timestamp
    } } pageInfo { endCursor
    startCursor
    You can find more examples, like code to download all questions, in our /scripts folder, to which we welcome contributions.
    Charts and question pages.
    Charts display a question's history. They look as follows:
    Charts can be accessed by clicking the expand button on the front page although they are fairly slow to load at the moment.
    Clicking on the expand button brings the user to a question page, which contains a chart, the full question description, and a range of quality indicators:
    We are also providing an endpoint at metaforecast.org/questions/embed/[id] to allow other pages to embed our charts. For instance, to embed a question whose id is betfair-1.178163916, the endpoint would be here. One would use it in the following code:
    You can find the necessary question id by clicking a toggle under "advanced options" on the frontpage, or simply by noticing the id in our URL when expanding the question.
    With time, we aim to improve these pages, make them more interactive, etc. We also think it would be a good idea to embed Metaforecast questions and dashboards into the EA Forum, and we are trying to collaborate with the Manifold team, who have done this before, to make that happen.
    Dashboards
    Dashboards are collections of questions. For instance, here is a dashboard on global markets and inflation, as embedded in Global Guessing.
    Like questions, you can either view dashboards directly, or embed them. You can also create them, at.
    Better infrastructure
    We have also revamped our infrastructure. We moved to from JavaScript to Typescript, from MongoDB to Postgres, and simplified our backend.
    We are open to collaborations
    We are very much open to collaborations. If you want to integrate Metaforecast into your project and need help do not hesitate to reach out, e.g., on our Github.
    Metaforecast is also open source, and we welcome contributions. You can see some to-dos here. Developing is going more slowly now because it's mostly driven by Nuño working in his spare time, so contributions would be counterfactual.
    Acknowledgements
    Metaforecast is hosted by the Quantified Uncertainty Research Institute, and has received funding from Astral Codex Ten. It has received significant contributions from Vyacheslav Matyuhin, who was responsible for the upgrade to Typescript and GraphQL. Thanks to Clay Graubard of Global Guessing for their comments and dashboards, to Insight Prediction for help smoothing out their API, to Nathan Young for general comments, to others for their comments and suggestions.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    4 min
  • AF - For ELK, truth is mostly a distraction by Cristian Trout
    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: For ELK, truth is mostly a distraction, published by Cristian Trout on November 4, 2022 on The AI Alignment Forum.
    Epistemic Status: Pretty confident in the central conclusions, and very confident in the supporting claims from meta-logic. Any low confidence conclusions are presented as such. NB: I give an intentionally revisionary reading of what ELK is (or should be) about. Accordingly, I assume familiarity with the ELK report. Summary here.
    Executive Summary
    ELK collapses into either the automation of science or the automation of mechanistic interpretability. I promote the latter.
    Abstract
    After reframing ELK from the perspective of a logician, I highlight the problem of cheap model-theoretic truth: by default reporters will simply learn (or search for) interpretations of the predictor’s net that make the teacher’s answers “true” in the model-theoretic sense, whether or not they are True (correspond with reality)! This will be a problem, even if we manage to avoid human simulators and are guaranteed an honest translator.
    The problem boils down to finding a way to force the base optimizer (e.g. gradient descent) to pay attention to the structure of the predictor’s net, instead of simply treating it like putty. I argue that trying to get the base optimizer to care about the True state of affairs in the vault is not a solution to this problem, but instead the expression of a completely different problem – something like automating science. Arguably, this is not the problem we should be focused on, especially if we’re just trying to solve intent alignment. Instead I tentatively propose the following solution: train the reporter on mechanistic interpretability experts, in the hope that it internalizes and generalizes their techniques. I expand this proposal by suggesting we interpret in parallel with training, availing ourselves of the history of a predictor’s net in order to identify and track the birth of each term in its ontology.
    The over-arching hope here is that if we manage to fully interpret the predictor at an earlier stage in its development, we can then maintain that transparency as it develops into something much more complex.
    To finish off, I zoom out, outlining three different levels of ELK-like projects, each building on the last, each more difficult than the last.
    Terminology
    I need to address some conflicts in namespace. I use the noun "model" in three different senses: as standing for logical models, ML models, and finally in the colloquial English sense, as synonymous with “a smaller, compressed representation of something.” Context should be enough to keep matters clear and I try to flag when I switch meanings, but I apologize in advance for any confusion. It’s a similar story with “truth” – many different senses in play. Sorry for any confusion; please watch for cues.
    I should also note that, because I’m not very familiar with Bayes nets, I’ll instead be (sloppily) talking as if the predictor is a Neural Net (NN) that somehow “thinks” in neat propositions. Basically, I know NNs work on fuzzy logic, and I’m then idealizing them as working on classical bivalent logic, for simplicity. I don’t think any of my arguments turn on this idealization, so it should be permissible.
    Finally, some shorthand:
    Translator ≝ ML model that operates roughly as follows:
    Take a question from humans as input
    Generate candidate answers using NL processing or something
    Using only a mapping which takes terms of these candidate answers to referents in the predictor’s net, generate a probability distribution over the candidate answers. This probability distribution is understood as a distribution of truth values in a fuzzy logic. (Again though, for simplicity I mostly pretend we’re working with bivalent logic).
    Output the answer(s) with the highest truth-value.
    Honest translat...
    35 min
  • EA - Supporting online connections: what I learned after trying to find impactful opportunities by Clifford
    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: Supporting online connections: what I learned after trying to find impactful opportunities, published by Clifford on November 4, 2022 on The Effective Altruism Forum.
    I’ve been exploring ideas for how the Forum could help people make a valuable connection over the course of 4+ months. In this doc, I share the different areas I’ve been thinking about, how promising I think they are and what I’ve learnt about them.
    Overall, I don't think any of the ideas meet the bar for focusing on connections via an online platform. My bar for an idea was that it could facilitate roughly an EAG’s worth of quality-adjusted connections each year (~10,000). I’m now more excited about meeting this bar by targeting high-value outcomes (e.g. collaborations, jobs), even if that means a lower number of total connections.
    I cover:
    Overall take
    Why did we explore connections?
    What ideas have we explored?
    FAQ: Why don’t you build a year-round Swapcard?
    Other ideas we didn’t explore
    Overall take
    I’m more excited about targeting high-value outcomes that come from connections (e.g. collaborations, jobs) rather than trying to broadly facilitate conversations online.
    I wanted to find a product that could plausibly match a single EAG (Effective Altruism Global) in terms of the number of connections made in a year.
    A single EAG costs roughly 2 FTEs a year to run (excluding other costs)
    If Sarah (my colleague) and I are choosing what to work on, you could imagine that the alternative we’re trying to beat or match is running another EAG.
    Note: this is a relatively ambitious target. I think it would be more achievable to build something that’s still cost-effective but which has less potential for scale. For example, some of the ideas I explore below would be much more cost-effective than an EAG, because it's much cheaper to run online services than in-person conferences but would generate far fewer connections. [Related Forum post.]
    I’ve found it hard to compete with EAG on number of “connections”, where a connection means someone you feel you can ask a favour of (likely after a half-hour conversation).
    At an EAG, each attendee makes an average of 8 connections via lots of 1:1 chats in an environment (in-person) optimised for good conversations.
    To match this, we’d need 10% of our monthly active readership to make 4 connections a year, which isn’t crazy but feels like a stretch as only 8% of that number have ever sent a message on the forum.
    I think it might be easier to compete with EAG on the downstream effects of connections, e.g. a person ended up collaborating on a paper, a person ended up getting hired.
    My best guess is that EAGs get outcomes like this for something like 1-5% of connections. I think it’s much easier for the Forum to get 100-500 additional concrete outcomes (like collaborations or jobs) a year, than 10k connections.
    The rest of this contains details on what we explored, why and our assessment of each idea.
    Why did we explore connections?
    We thought we should work on connections because:
    Connections are valuable
    They are cited as an important reason that people get involved in EA in the Open Phil EA/Longtermist Survey
    35% of people said personal contact with EAs was important for them getting involved
    38% said personal contacts had the largest influence on their personal ability to have a positive impact
    We had some evidence that people are making valuable connections through the Forum, even though the Forum wasn’t designed well for this
    The 2020 EA Survey revealed that a surprising number of people found connections through the EA Forum
    5.9% of people said they found a valuable connection through the Forum, where 11.3% said this of EAG (an event optimised for connections)
    Having followed up with a subset of the forum users who said this, I think it’s more like that ~2% of respondent...
    24 min
  • AF - Toy Models and Tegum Products by Adam Jermyn
    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: Toy Models and Tegum Products, published by Adam Jermyn on November 4, 2022 on The AI Alignment Forum.
    (Thanks to Adam Scherlis, Kshitij Sachan, Buck Shlegeris, Chris Olah, and Nicholas Schiefer for conversations that informed this post.)
    Anthropic recently published a paper on toy models of superposition [Elhage+22]. One of the most striking results is that, when features are sparse, feature embeddings can become divided into disjoint subspaces with just a few vectors per subspace. This type of decomposition is known as a tegum product.
    This post aims to give some intuition for why tegum products are a natural feature of the minima of certain loss functions.
    Setup
    Task
    Suppose we’ve got d embedding dimensions and n>d features. We want to embed features into dimensions in a way that minimizes overlap between their embedding vectors. In the simplest case this could be because we’re building an autoencoder and want to compress the features into a low-dimensional space.
    One approach is to encode all of the features in all of the dimensions. With this approach there is some interference between every pair of features (i.e. no pair is embedded in a fully orthogonal way), but we have a lot of degrees of freedom that we can use to minimize this interference.
    Another approach is to split the d dimensions into k orthogonal subspaces of d/k dimensions. This has the advantage of making most pairs of vectors exactly orthogonal, but at the cost that some vectors are packed more closely together. In the limit where k=1 this reduces to the first approach.
    Our aim is to figure out the k that minimizes the loss on this task.
    Loss
    Suppose our loss has the following properties:
    L=∑i≠jℓ(cosθij). That is, the loss decomposes into a sum of terms involving the cosine similarities of feature vectors, and all features are equally important.
    ℓ(0)=0. The loss vanishes for orthogonal vectors.
    dℓ/dcosθ>0. The loss is greater the more the vectors overlap.
    Using these properties, we find that the loss is roughly
    where ϵ is the typical cosine similarity between vectors in a subspace.
    Loss-Minimizing Subspaces
    The Johnson-Lindenstrauss lemma says that we can pack m nearly-orthogonal vectors into D dimensions, with mutual angles satisfying
    where
    and ϵ0 is a constant. Setting m=n/k and D=d/k gives
    Assuming we pick our vectors optimally to saturate the Johnson-Lindenstrauss bound, we can substitute this for ϵ in the loss and differentiate with respect to k to find
    There are three possible cases: either the minimum occurs at k=d (the greatest value it can take), or at k=1 (the smallest value it can take) or at some point in between where dL/dk vanishes.
    The derivative vanishes if
    which gives
    where
    When α≥2 there is no place where the derivative vanishes, and the optimum is k=1. Otherwise there is an optimum at
    so long as this is less than d. If it reaches d, the optimum sticks to k=d.
    Interpretation
    We can think of α as the sensitivity of the loss to interference. Specifically, which moment of the interference distribution do we care about?
    When α is large, we care more about decreasing higher moments, and in the limit of infinite α what matters is just the maximum interference between vectors. Hence when α is large we want to have fewer subspaces, each with more vectors but smaller cosine similarities.
    By contrast, when α is small, we care more about decreasing smaller moments, and in the limit as α0 what matters is the fraction of vectors that interfere at all. Hence when α is small we want to have more subspaces, each with fewer vectors but larger cosine similarities.
    So tegum products are preferred when we can tolerate larger “peak” interference and want fewer instances of interference, whereas a single large subspace is preferred when we can tolerate lots of instances of interference and want to mi...
    7 min
  • AF - A newcomer’s guide to the technical AI safety field by Chin Ze Shen
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: A newcomer’s guide to the technical AI safety field, published by Chin Ze Shen on November 4, 2022 on The AI Alignment Forum.
    This post was written during Refine. Thanks to Jonathan Low, Linda Linsefors, Koen Holtman, Aaron Scher, and Nicholas Kees Dupuis for helpful discussion and feedback.
    Disclaimer: This post reflects my current understanding of the field and may not be an accurate representation of it. Feel free to comment if you feel that there are misrepresentations.
    Motivations
    I remember being fairly confused when I first started reading AI safety related posts, especially when it pertains to specific ideas or proposals, as there may be implicit assumptions behind those posts that relies on some background understanding about the research agenda. I have since had the opportunity to clear up many of those confusions by talking to many people especially while I was participating in Refine. Looking back, there were many background assumptions about the field I wish I had known earlier, so here’s the post I never had.
    This post does not intend to cover topics like why AI safety is important, how to get into the field, or an overview of research agendas, as there are plenty of materials covering these topics already.
    Some terminology
    Artificial intelligence (AI) refers to intelligences that are created artificially, where intelligence measures an agent's ability to achieve goals in a wide range of environments as per Shane Legg and Marcus Hutter’s definition. Artificial general intelligence (AGI) refers to a machine capable of behaving intelligently over many domains, unlike narrow AIs which only perform one task. Artificial superintelligence (ASI), or “superintelligent AI”, refers to an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills. On the other hand, transformative AI (TAI) is an AI that precipitates a transition at least comparable to the agricultural or industrial revolution, where the concept refers to its effects instead of its capabilities.
    In practice however, the terms AGI, ASI, and TAI are sometimes used interchangeably to loosely mean “an AI that is much more powerful than humans”, where “power” (related to “optimization”) is often used to loosely mean “intelligence”.
    AI safety is about making the development of AI go safely. It is often used to refer to AGI safety or AI alignment (or just “alignment” because “AI alignment” is too long), which roughly refers to aligning a hypothetical future AI to what humans want in a way that is not catastrophic to humans in the long term. There is of course the question of “which humans should we align the AI to” that is often raised, though the question of “how do we even properly align an intelligent system to anything at all” would be much a more central problem to most researchers in the field.
    In other communities however, AI safety is also used in the context of ensuring the safety of autonomous control systems such as self-driving cars and unmanned aircrafts, which is typically outside of the scope of AI alignment. In this post however, I will mostly use the term AI safety to mean AI alignment as how it is often used in introductory materials like this one, although it may be generally better to have clearer distinctions of these terms.
    In Steve Byrnes’ diagram from this post, the red box serves as a good representation of the AI alignment field, and separates “alignment” from “existential risk (x-risk) mitigation” in a nice way:
    The diagram below from a talk by Paul Christiano also describes alignment (“make AI aligned”) as a specific subset of “making AI go well”.
    Brief history of AI and AI safety
    AI development can roughly be divided into the following era:
    1952 - 1956: The birth of AI by a handful of ...
    18 min
  • EA - How bad could a war get? by Stephen Clare
    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 bad could a war get?, published by Stephen Clare on November 4, 2022 on The Effective Altruism Forum.
    Acknowledgements: Thanks to Joe Benton for research advice and Ben Harack and Max Daniel for feedback on earlier drafts.
    Author contributions: Stephen and Rani both did research for this post; Stephen wrote it and Rani gave comments and edits.
    Previously in this series: "Modelling great power conflict as an existential risk factor" and "How likely is World War III?"
    Introduction & Context
    In “How Likely is World War III?”, Stephen suggested the chance of an extinction-level war occurring sometime this century is just under 1%. This was a simple, rough estimate, made in the following steps:
    Assume that wars, i.e. conflicts that cause at least 1000 battle deaths, continue to break out at their historical average rate of one about every two years.
    Assume that the distribution of battle deaths in wars follows a power law.
    Use parameters for the power law distribution estimated by Bear Braumoeller in Only the Dead to calculate the chance that any given war escalates to 8 billion battle deaths
    Work out the likelihood of such a war given the expected number of wars between now and 2100.
    Not everybody was convinced. Arden Koehler of 80,000 Hours, for example, slammed it as “[overstating] the risk because it doesn’t consider that wars would be unlikely to continue once 90% or more of the population has been killed.” While our friendship may never recover, I (Stephen) have to admit that some skepticism is justified. An extinction-level war would be 30-to-100 times larger than World War II, the most severe war humanity has experienced so far. Is it reasonable to just assume number go up? Would the same escalatory dynamics that shape smaller wars apply at this scale?
    Forecasting the likelihood of enormous wars is difficult. Stephen’s extrapolatory approach creates estimates that are sensitive to the data included and the kind of distribution fit, particularly in the tails. But such efforts are important despite their defects. Estimates of the likelihood of major conflict are an important consideration for cause prioritization. And out-of-sample conflicts may account for most of the x-risk accounted for by global conflict. So in this post we interrogate two of the assumptions made in “How Likely is World War III?”:
    Does the distribution of battle deaths follow a power law?
    What do we know about the extreme tails of this distribution?
    Our findings are:
    That battle deaths per war are plausibly distributed according to a power law, but few analyses have compared the power law fit to the fit of other distributions. Plus, it’s hard to say what the tails of the distribution look like beyond the wars we’ve experienced so far.
    To become more confident in the power law fit, and learn more about the tails, we have to consider theory: what drives war, and how might these factors change as wars get bigger?
    Perhaps some factors limit the size of war, such as increasing logistical complexity. One candidate for such a factor is technology. But while it seems plausible that in the past, humanity’s war-making capacity was not sufficient to threaten extinction, this is no longer the case.
    This suggests that wars could get very, very bad: we shouldn’t rule out the possibility that war could cause human extinction.
    Battle deaths and power laws
    Fitting power laws
    One way to gauge the probability of out-of-sample events is to find a probability distribution, a mathematical function which gives estimates for how likely different events are, which describes the available data. If we can find a well-fitting distribution, then we can use it to predict the likelihood of events larger than anything we’ve observed, but within the range of the function describing the distribution.
    Several researchers have...
    19 min
  • EA - Major update to EA Giving Tuesday by Giving What We Can
    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: Major update to EA Giving Tuesday, published by Giving What We Can on November 4, 2022 on The Effective Altruism Forum.
    Announcing changes to EA Giving Tuesday and Meta's Giving Season match
    On Nov 1st 2022, Meta announced a significant change to their annual Giving Tuesday donation matching scheme, which affects EA Giving Tuesday.
    Here are the high level details of this year’s match:
    “To help nonprofits jumpstart their Giving Season fundraising, Meta will match your donors’ recurring donation 100% up to $100 in the next month (up to $100,000 per organization and up to $7 million in total across all organizations). All new recurring donors who start a recurring donation within November 15 - December 31, 2022 are eligible. Read the terms and conditions.”
    The match now requires participants to set up a recurring donation, in order to get up to $100 in matched funds. The matched funds are provided once the second transaction goes through i.e. you need to donate for two months to receive the match. We are unsure but it seems likely that a donor could set up recurring donations to multiple organizations in order to get multiple matches.
    We believe this opportunity is only available in the US due to the functionality appearing to be US only.
    What does this mean for EA Giving Tuesday?
    In the past, the value proposition of EA Giving Tuesday was to organise around the 100% match in the morning of Giving Tuesday. With the lower match amount per donation and the requirement for it to be recurring, we think that the matching is much less likely to be competitive and therefore the previous level of coordination does not make sense to continue.
    The EA Giving Tuesday team has also decided that it makes the most sense for people to donate directly to the charities via Facebook given the new requirement about recurring donations. These changes mean that EA Giving Tuesday will not support any organisations that require re-granting or restrictions due to ongoing administrative requirements of recurring donations. 501c3’s registered with Facebook fundraising tools will be able to participate in Meta’s Giving Season match.
    We encourage you to look for effective charities on Facebook for the match and will be listing effective charities who are interested in participating on our website.
    EA Giving Tuesday will share the details of any matching opportunities we think are worthwhile and conduct an impact analysis at the end of the season.
    How can I get my donations matched this year?
    This year there are two match opportunities we are sharing with you:
    Meta’s Giving Season match: Details and instructions (Starts Nov 15)
    Every.org’s Fall Giving Challenge: Details and instructions, provided by Will Kiely on the EA Forum (Already started)
    Once you’ve received confirmation of a match, please let us know the details via this impact evaluation form so we can quantify the value of these opportunities for future years.
    We’re disappointed that the match has changed significantly from the previous year, but we hope you find value in the matching opportunities from both Meta and Every.org.
    We will continue to search for new matching opportunities that have the ability to shift donations towards highly effective charities throughout the season.
    You can read more about EA Giving Tuesday at EAGivingTuesday.org
    Grace and the EA Giving Tuesday Team 2022
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    4 min
  • LW - Adversarial Policies Beat Professional-Level Go AIs by sanxiyn
    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: Adversarial Policies Beat Professional-Level Go AIs, published by sanxiyn on November 3, 2022 on LessWrong.
    An interesting adversarial attack at KataGo, a professional level Go AI. Apparently funded by Fund for Alignment Research (FAR). Seems to be a good use of fund.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min