The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - Top lesson from GPT: we will probably destroy humanity "for the lulz" as soon as we are able. by shminux
    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: Top lesson from GPT: we will probably destroy humanity "for the lulz" as soon as we are able., published by shminux on April 16, 2023 on LessWrong.
    Forget complicated "sharp left turn" schemes, nefarious nanobots, lists of lethalities, out-of-distribution actions, failed AI boxing. As Zvi pointed out in multiple posts, like this one, if humans get unrestricted access to a powerful enough tool, it is all over. People will intentionally twist even the most aligned Tool AI into an Agent of Death, long before it becomes superintelligent and is able to resist. You can find examples of it online.
    In that sense, Eliezer was wrong in the worst possible way: we have a lot less time to get our act together, because capabilities advance faster than intelligence and humans are very inventive at finding ways to misuse these capabilities. We will push these capabilities in the "gain of function" direction mercilessly and without regard for safety or anything else. We are worse than toddlers playing with matches. True, like with toddlers, our curiosity far outstrips our sense of self-preservation, probably because our brains are not wired to be afraid of something that is not like a snake or a spider or a steep cliff. But it is worse than that. People will try to do the worst thing imaginable because they do not "alieve" potential harm, even if they can track it logically, unlike a toddler.
    I guess the silver lining is that we have a bit of time to iterate. The AI tools are not yet at the level of causing widespread destruction, and probably will not be for some time. It does not mean that if and when some superintelligence emerges we will be well prepared, but if humanity survives until then without self-annihilation, we might have a better chance, compared to the "one shot at getting it right" before we are all wiped out, as Eliezer emphasized. It might not be an "alignment manual from the surviving future", but at least some wisdom and discipline of avoiding the early pitfalls, and if we die, we might die with "more dignity". The odds are not great, but maybe they are there.
    Edit: quanticle pointed out that Bostrom predicted it in the paper The Vulnerable World Hypothesis:
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • LW - grey goo is unlikely by bhauth
    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: grey goo is unlikely, published by bhauth on April 17, 2023 on LessWrong.
    The nanomachinery builds diamondoid bacteria, that replicate with solar power and atmospheric CHON, maybe aggregate into some miniature rockets or jets so they can ride the jetstream to spread across the Earth's atmosphere, get into human bloodstreams and hide, strike on a timer.
    Eliezer Yudkowsky
    To control these atoms you need some sort of molecular chaperone that can also serve as a catalyst. You need a fairly large group of other atoms arranged in a complex, articulated, three-dimensional way to activate the substrate and bring in the reactant, and massage the two until they react in just the desired way. You need something very much like an enzyme.
    Richard Smalley
    My understanding is that anyone who can grasp what "orthos wildly attacking the heterodox without reading their stuff and making up positions to attack" looks like, considers that this is what Smalley did with Drexler - made up an unworkable approach and argued against it.
    Eliezer Yudkowsky
    In this post, I use "nanobots" to mean "self-replicating microscopic machines with some fundamental mechanistic differences from all biological life that make them superior". Various specific differences from biological cells have been proposed. I've organized this post by those proposed differences.
    1. localized melting
    Most 3d printers melt material to extrude it through a nozzle. Large heat differences can't be maintained on a small scale.
    2. rare materials
    If a nanobot consists largely of something rare, getting more of that material to replicate is difficult outside controlled environments.
    Growth of algae and bacteria is often limited by availability of iron, which is more common than most elements. Iron is the active catalytic site of many enzymes, and is needed by all known life. The growth of something made mostly of iron would be far more limited, and other metals have more limited availability than that.
    3. metal surfaces
    Melting material isn't feasible per (1), so material must be built up by adding to the surface. Since that's the case, the inside of structures must be chemically the same as what was the exterior.
    Metal objects have a protective oxide layer. In an air or water environment, there's no way to add individual (eg) aluminum atoms to a metal surface and end up with metallic aluminum inside; the whole thing will typically be aluminum oxide or hydroxide.
    Corrosion is also a proportionately bigger problem for smaller objects. A micrometer-scale metal structure will rapidly corrode, perhaps doing some Ostwald ripening.
    4. electric motors
    Normal "electric motors" are all electromagnetic motors, typically using ferromagnetic cores for windings. Bigger is better for those, up to at least the point where you can saturate cores.
    On a very small scale, it's better to use electrostatic motors, and you can make MEMS electrostatic motors with lithography. (Not just theoretically; people actually do that.) But, per (2) & (3), bulk metals are a problem for a self-replicating system. If you need to have compounds floating around, electrical insulation is also difficult. You also need some way to switch current, and while small semiconductor switches are possible, per (3) building them is difficult.
    Instead of electrostatic charge of metal objects, it's better to use ions. Ions could bind to some molecule, and electrostatic forces could cause that to rotate relative to another molecule. Hmm, this is starting to sound rather familiar.
    5. inorganic catalysts
    Lab chemistry and drug synthesis often use metal catalysts in solution, perhaps with a small ligand. Palladium acetate is used for making drugs, but it's very toxic to humans, because it...catalyzes reactions.
    Life requires control of what happens, which means selective catalysis of r...
    16 min
  • LW - My experience getting funding for my biological research by Metacelsus
    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: My experience getting funding for my biological research, published by Metacelsus on April 16, 2023 on LessWrong.
    Linked from my Substack
    Research is expensive, especially these days. Even though grad students and postdocs are underpaid, it still costs a lot to employ them.
    And every year, lab suppliers increase their prices. Overall, it continually gets more expensive to do the same amount of work:
    Therefore, in order to carry out my meiosis project, I needed a source of funding. I am just a grad student and usually it is a professor’s job to get grants, but for various reasons I decided to take matters into my own hands.
    After my experience with the F31 fellowship and all its paperwork, I was rather turned off by the prospect of applying for an NIH grant (R01 or R21). In the best case, I would have to write a huge application, then wait for months before hearing if it was funded, then more months before actually being able to spend anything. And of course, funding chances are always low with the NIH.
    So, I decided to seek out private philanthropic funding. This, as it turned out, was not as simple as I had expected.
    Timeline of events
    September 2022
    I develop my project idea and figure out roughly how much funding I need ($300,000 per year over two years, to cover salaries and lab supplies).
    I reach out to some potential funders. I quickly focus on the FTX Future Fund, since they are known to have a giant pile of cash they want to get rid of.
    October 2022
    I write a formal research proposal and submit it to some FTX Future Fund regrantors.
    I also start meeting with some university admins who manage research finances. They agree that we can set the overhead rate at 15%.
    October 29: I meet with the regrantors who tell me they are excited about the project.
    November 2022
    November 3: The FTX Future Fund regrantors announce that they’ve approved the grant, and can I please send them my bank account info for them to wire $300,000?
    But before we can accept the funds, my university needs to do due diligence to make sure they’re not criminals.
    The Jeffrey Epstein scandal created some strong organizational scar tissue. Perhaps this is a good thing, because . . .
    November 8:
    November 14: It’s confirmed that I won’t be seeing a cent of the $300,000 from the FTX Future Fund. Well, at least I am not at risk of clawbacks.
    Late November: I meet with various other philanthropists, and receive multiple suggestions to apply for a Survival and Flourishing Fund (SFF) Speculation Grant. I start working on the application.
    November 17: The Repro Grants are announced. This seems like a perfect fit for my research.
    November 20: I submit my initial Repro Grant application. Overall I spent about 6 hours writing it (they claimed it would take 30 minutes, but it’s certainly better than the 100 hours I spent on the F31).
    December 2022
    December 7: I hear that my initial Repro Grant application was not funded. However, there is still time left in the application period, so I spend another ~6 hours writing a different proposal (focusing on ovarian organoids instead of meiosis).
    December 22: I submit my application for a SFF Speculation Grant. Around this time, I am also discussing the project with two private donors who may be interested in funding it. One of these backs out (due, I think, to losing money in the crypto downturn), but the other one is very excited about it.
    December 29: SFF agrees to provide $401,000 for the project, with the private donor also providing $300,000. This is at the high end of my requested budget, so I am especially happy about it.
    January 2023
    January 3: The Repro Grant was funded! Although they only gave me $65,000 out of the $100,000 I requested, it’s still excellent news.
    Early January: University admins agree to accept the money from SFF and the Repro Grant. Howev...
    10 min
  • EA - ZzappMalaria: Twice as cost-effective as bed nets in urban areas by Arnon Houri Yafin
    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: ZzappMalaria: Twice as cost-effective as bed nets in urban areas, published by Arnon Houri Yafin on April 16, 2023 on The Effective Altruism Forum.
    I'm Arnon Houri Yafin, CEO of ZzappMalaria. I'm writing about our larviciding pilot published in Malaria Journal (Vigodny et al. 2023), and this post is my interpretation of the results.
    ZzappMalaria: Twice as cost-effective as bed nets in urban areas
    TL;DR: Zzapp Malaria’s digital technology for planning and managing large-scale anti-malaria field operations attained results that are twice as cost-effective as bed nets in reducing malaria in urban and semi-urban settings.
    Call to action
    Use our solution
    Fund us
    How to save more than 140,000 people annually for a cost per person that is lower than bed nets?
    In 2021, 627,000 people died from malaria, more than 80% of whom were children. The number of people who live in areas at high risk for malaria is approximately 1.1 billion - which comprises the population of sub saharan Africa (excluding South Africa) and high burden malaria countries such as Odisha in India. According to estimates, more than 75% of these people live in urban and peri urban areas for which our solution costs less than US$0.7 per person per year. The cost to protect all these people is therefore:
    0.7 1.1B US$0.7 = US$0.54B.
    The malaria incidence in urban areas is lower than in villages so it is assumed that the urban and peri urban population accounted for 45% of the malaria death:
    0.45 627,000 = 282,100 people.
    Assuming that our solution can reduce malaria by 52.5% (based on a peer-reviewed article we recently published in Malaria Journal; elaboration below), our intervention can save:
    282,100 0.525 = 148,100 people.
    As would be elaborated below, we believe the actual numbers of both cost and of effectiveness will dramatically improve over time.
    What is Zzapp’s solution and how does its technology work?
    TL;DR: We believe in actively targeting disease-bearing mosquitoes, and use digitization to manage large-scale field operations focusing on treatment of mosquitoes breeding sites.
    ZzappMalaria harnesses entomological knowledge and data analyzed from satellite imagery and collected by field workers to optimize large-scale larviciding operations. In such operations, the stagnant water bodies in which Anopheles mosquitoes breed are treated with an environmentally benign bacteria that besides mosquitoes and blackflies does not harm any other animals (not even other insects) and is approved for use in sources of drinking water. Although larviciding has led to malaria elimination in many countries in the 20th century, it is not easy to implement since it requires planning, management and monitoring of large teams working kilometers away from each other. To complicate things, effective larviciding requires that a large proportion of water bodies are detected and treated on a regular basis (sometimes weekly).
    We developed a system comprising a planning tool, mobile app and dashboard designed to overcome these challenges. The system extracts from satellite images the location of houses and demarcates the general area for the intervention. It then recommends where to scan for water bodies, and helps in implementation through a designated GPS-based mobile app that allocates treatment areas to workers, monitors their location in the field to ensure they cover the entire area, and keeps track of schedules for water body treatment. All information is uploaded to a dashboard, allowing managers to monitor the operation in real time. Specifically designed for sub-Saharan Africa, our house-detection algorithm was developed to detect both modern houses and traditionally built huts. Our app was built to work on inexpensive smartphones and in areas with weak internet infrastructure. Its interface is simple and intuitiv...
    11 min
  • EA - Impactful (Side-)Projects and Organizations to Start by AlexandraB
    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: Impactful (Side-)Projects and Organizations to Start, published by AlexandraB on April 16, 2023 on The Effective Altruism Forum.
    Are you an (aspiring) changemaker looking for the next thing to do? Consider founding something!
    In Nov and Dec of 2022 I have scoured the depths of the internet to find as many lists of impactful project/organization/charity ideas as I could to gain inspiration for what to start myself. I hope that sharing this curated version of my 'list of lists' can help more people to get the creative juices flowing and to take the next step.
    There are definitely resources that I have overlooked or newer ones which did not exist yet during my search, so I'd invite everyone to share ideas and links to additional resources in the comments.
    Longtermist & Neartermist - mixed ideas
    Possible gaps in the EA community (2021)
    Things CEA is not doing (2021) by CEA
    Big List of Cause Candidates (2020)
    Cause candidates tag on the EA forum
    Annotated List of Project Ideas & Volunteering Resources (2020) which includes smaller and larger project ideas and links to many lists.
    69 things that might be pretty effective to fund (2018) lists many (potentially) high-impact projects.
    List of possible EA meta-charities (2019)
    EA Summit Project Ideas
    Projects I’d like to see (2017)
    Concrete project lists (2017), both the post and the comments might be valuable.
    EAV Ideas in Need of Implementation (2015)
    Longtermist ideas
    Nonlinear's idea list (2023) contains hundreds of ideas. It is not published yet but you can request access to the document.
    The Future Fund's Project Ideas Competition (2022) - The comments under this post.
    Ideas that were on the Future Fund's website (2022).
    Longtermist Entrepreneurship Fellowship ideas database (2020)
    Neartermist ideas
    Charity Entrepreneurship's research reports (2020-2023)
    Animal Welfare charity recommendations by Animal Charity Evaluators:
    Charities We’re Excited About (2018)
    Interventions we’d like to see (2017)
    Charities we’d like to see (2016)
    Charities we’d like to see by GiveWell (2015)
    Bonus resources
    Charity Entrepreneurship's handbook (2021)
    So you want to be a charity entrepreneur. Read these first. (2022)
    Help for coming up with ideas:
    Prompts to find problems (2020)
    Two EA project idea exchange platforms:
    EA Work Club Projects
    Airtable Project Platform
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • EA - Local EA groups: consider becoming more than a satellite group by Ada-Maaria Hyvärinen
    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: Local EA groups: consider becoming more than a satellite group, published by Ada-Maaria Hyvärinen on April 16, 2023 on The Effective Altruism Forum.
    I recently wrote about the history of Effective Altruism Finland, the local EA group I am part of and used to work for. I suspect most readers can guess the main activities we carry out even without clicking on the post. (Spoiler: we have an intro to EA program, a career advising service, an effective donations website, support for university groups and regular EA-themed discussion events).
    Why do local EA groups often seem so similar to each other? I believe this is because many groups inadvertently become satellite groups. In my post, I try to explain what I mean by this, and argue that local groups should consider developing more distinctive features, even though there may be reasons to deliberately remain a satellite group.
    What is a satellite group?
    Some characteristics that would define an archetypal satellite group are:
    They mainly follow materials produced by other groups, such as introductory fellowship curricula
    They try to undertake activities that are inspired by other EA groups and follow "official" EA advice
    The members' goals are to eventually join another "real" EA organization or move to an EA hub "where the real EAs are" (as opposed to aiming to do things as a group locally or as a collaboration between the members of the group they are currently affiliated with)
    They might feel the “real EA knowledge” is possessed by other people elsewhere, and that people in their surroundings are not fit to meaningfully contribute to the EA project yet
    The theory of impact of a satellite group comes from broadcasting EA material to new people and connecting their audience to other EA groups and organizations. The actual impactful stuff happens only when a group member starts doing impactful things somewhere else than in the context of the group.
    If you are a new group, starting out as a satellite group makes total sense. For some groups, like university groups, staying a satellite group indefinitely can also be also a good idea. However, for some groups, focusing only on satellite type of activities has also a cost of missed opportunities.
    Satellite activities are safe and beginner-friendly
    There are a bunch of reasons why groups might get stuck in the satellite phase, even if they intend to move towards more original activities. Satellite activities are often locally optimal: if you need to decide what you personally should do in the next couple of weeks or months, in terms of expected impact it often seems like you should focus on satellite activities. It is easy to make the case for organizing an 80 000 hours themed career planning workshop or a giving game.
    Satellite activities are also relatively well-regarded within the EA community. If you post on the EA forum about having organized such a career workshop or giving came, other readers will think that your group is doing the reasonable thing.
    This positive attitude also somewhat translates to funding. If you can show that your group has inspired many people to work for EA organizations, EA funders are more likely to provide funding for your group in the future. It is harder to prove that your members are doing impactful things if they are working on a original local project that the funders don't know much about.
    Furthermore, doing satellite activities can protect your group from doing obviously bad or useless things. If you can reliably organize an introductory fellowship to EA, it is guaranteed that a certain number of people will understand EA better than before. You don’t need to spend a lot of time calculating your expected impact or thinking about potential downsides: if you can execute the program well enough, you can be assured the thing you do probably...
    8 min
  • AF - Possibilizing vs. actualizing by Tsvi Benson-Tilsen
    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: Possibilizing vs. actualizing, published by Tsvi Benson-Tilsen on April 16, 2023 on The AI Alignment Forum.
    [Metadata: crossposted from. First completed December 31, 2022. This essay is more like research notes than exposition, so context may be missing, the use of terms may change across essays, and the text might be revised later; only the versions at tsvibt.blogspot.com are definitely up to date.]
    Some behavior seems like it's just making things possible, without actually doing much of anything, while other behavior seems to actually do something. Is there a principled, or a useful, distinction between possibilizing and actualizing? Is it possible to possibilize a large effect on the world without actualizing large effects on the world?
    It's not clear whether this is a real distinction, but to me it's a very intuitive and intuitively salient idea (because possibilizing seems safer than actualizing), so I'd like to have a better analysis or dissolution of it.
    Terms
    Possibilizing is making something possible for an agent to do. That is, it's setting the stage, preparing, unlocking, satisfying the preconditions, gathering the needed resources and tools, gaining the necessary understanding and skill and components and information, getting agents on board, making the plans, opening the way. "Possible" is cognate with "potent", "power", and "hospital". Other possible words: enabling, feasibilizing, empowering.
    Actualizing is (an agent) actually doing something, making something happen, achieving a goal, affecting something, having impact, delivering a payload. Other possible words: realizing, implementing, exerting.
    Examples
    Taking a step forward along a path possibilizes the next step, and contributes to possibilizing further steps; and it actualizes itself, a forward movement of the walker. In this example, it does seem like there's much distinction between possibilizing and actualizing: what's being possibilized (subsequent steps) is the same sort of thing as the possibilizing (the current step), and they interact in a straightforward way. Intuitively, this is mere progress, not possibilizing, so the above definition of possibilizing is incomplete.
    Speed cubers first look at the Rubik's cube, then put it down, and then pick it up and turn its faces. Before they pick it up the second time, in their head they're possibilizing solving the cube by figuring out the sequence of moves. Before that, they had to understand how to solve the cube and train that skill, which was also possibilizing solving the cube. The order can't be switched: you can't turn the faces so the cube is in a solved state, and then, after that, figure out for the first time what sequence of turns would solve the cube. But you can "think with the cube", trying out moves to see what happens, sometimes backtracking, which blurs the distinction.
    To build a complicated house, you look at the site where you'll build it, then you leave and elsewhere you think and draw pictures, and then later you come back and build the house.
    A lot of engineering tasks are solved straightforwardly. But engineering tasks that push the envelope require the engineers to recurse up to more abstract thinking (e.g. "first-principles thinking"), removing all but the most necessary assumptions/constraints, finding an abstract design that satisfies the abstract constraints, and then constructing a new concrete, detailed design in a way that's beholden to the abstract design.
    Coming up with a word possibilizes the expression of some thoughts.
    Coming up with a concept possibilizes making good predictions and finding good designs. If you don't know about chromatic dispersion, you'll be confused about why your lenses produce images with weird color-dependent blurring, and you'll have trouble knowing, without physically testing, how muc...
    10 min
  • LW - SmartyHeaderCode: anomalous tokens for GPT3.5 and GPT-4 by AdamYedidia
    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: SmartyHeaderCode: anomalous tokens for GPT3.5 and GPT-4, published by AdamYedidia on April 15, 2023 on LessWrong.
    TL;DR: There are anomalous tokens for GPT3.5 and GPT4 which are difficult or impossible for the model to repeat; try playing around with SmartyHeaderCode, APolynomial, or davidjl. There are also plenty which can be repeated but are difficult for the model to spell out, like edTextBox or legalArgumentException.
    A couple months ago, Jessica Rumbelow and mwatkins posted about anomalous tokens that cause GPT-2 and GPT-3 to fail. Those anomalous tokens don't cause the same failures on newer models, such as GPT-3.5 Default or GPT-4 on the ChatGPT website, or gpt-3.5-turbo over the API, because the newer models use a different tokenizer. For a very brief explanation of what a tokenizer is doing, the tokenizer has a large vocabulary of tokens, and it encodes ordinary text as a sequence of symbols from that vocabulary.
    For example, the string Hello world! gets encoded by the GPT-2 tokenizer as the sequence [15496, 995, 0], meaning that it's a sequence of three tokens, the first of which is the 15,946th token of the vocabulary, or Hello, the second of which is the 995th token of the vocabulary, or world, and the third of which is the 0th token of the vocabulary, or !. In general, a long string being represented by a single token implies that that string appears a lot in the training set (or whatever corpus was used to build the tokenizer), because otherwise it wouldn't have been "worth it" to give that string its own token.
    Because of the change in tokenizers, almost all of the tokens which produce anomalous behavior in GPT-2 and GPT-3 don't produce anomalous behavior in the later models, because rather than being a single weird token, they're broken up into many, more normal tokens. For example, SolidGoldMagikarp was encoded as a the single token SolidGoldMagikarp by the old tokenizer, but is encoded as five tokens by the new tokenizer: [' Solid', 'Gold', 'Mag', 'ik', 'arp']. Each of those five tokens is normal and common, so GPT-4 handles them just fine.
    Also, it's conveniently the case that tokenizers are released to the public and generally seem to be ordered, with earlier tokens in the vocabulary being shorter, more common, and more ordinary, and later tokens in the vocabulary being longer, less common, and weirder.
    The old tokenizer, r50k_base, used a vocabulary of about 50,000 tokens and was used by GPT-2 and GPT-3 (and possibly GPT-3.5 Legacy?). The new tokenizer, used by GPT-3.5 Default and GPT-4, is called cl100k_base and has a vocabulary of about 100,000 tokens. Unfortunately, we can't straightforwardly repeat the experiment that Jessica Rumbelow and mwatkins ran, of running k-means clustering on the model's embedding matrix, because (to my knowledge) we don't have access to the embedding matrix of the newer models. Instead, however, we can just look at the later tokens in the cl100k_base vocabulary and try messing around with each of them; the later tokens, being longer, rarer, and weirder, are easier to use to create prompts that are far from the model's training distribution.
    To give a sense for what a completely random sample of late-vocabulary cl100k_base tokens look like, here's tokens 98,000 through 98,020:
    ['.Cdecl', 'InstantiationException', ' collage', ' IOC', ' bais', ' onFinish', '-stars', 'setSize', 'mogul', ' disillusion', ' chevy', '(Schedulers', '(IR', '_locs', ' cannons', ' cancelling', '/bus', ' bufio', ' Yours', ' Pikachu', ' terme']
    I searched through tokens 98,000 through 99,999 in the cl100k_base vocabulary. I focused on just the tokens that contained only Roman-alphabet characters and spaces, to avoid confusing it for uninteresting reasons (like asking it to repeat the string ("");, which contains enough punctuation that it might f...
    12 min
  • EA - Giving Guide for Student Organisations – An ineffective outreach project by Karla Still
    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: Giving Guide for Student Organisations – An ineffective outreach project, published by Karla Still on April 14, 2023 on The Effective Altruism Forum.
    Executive summary
    EA Helsinki created a giving guide booklet for student organisations in Finnish.
    The main aim of the Giving Guide for Student Organisations was to raise awareness of EA among students by providing them with information on how to give effectively. A sub-goal was to redirect donations from student organisations, although we do not expect the expected value of redirected donations to be worth the effort of producing and marketing the guide.
    We distributed printed versions of the guide to 121 organisations at 5 universities, most by approaching them at student fairs we attended in the fall of 2022. A digital version of the guide was also sent to the organisations via email a few days after the printed booklet was distributed.
    The direct results or impact from the giving guide have been almost non-existent. I believe most of the impact comes later from having exposed people to the idea of effective giving, making them more receptive to EA later on.
    My original reason for writing this post was to offer other EA groups an opportunity to replicate the project, but now I am not so sure that it is worth it. Instead, I would love to get feedback on the impact assessment, the project, and hear if others have tried something similar. I hope the spelled-out project implementation and analysis can help other community builders who might think about starting similar projects. I also wrote this as a report for future community builders in Finland.
    Tip: Skip to the measuring the impact-section if you aren’t interested in the details about the content and implementation of the project.
    Content of the giving guide
    Below is the structure of the guide. (In parentheses the length of the text in A5 pages.) You can also find an auto translation on the content excluding the donation recommendations or download the Finnish guide on our website:
    Title: A giving guide for student organisations. How to achieve more impact with the same amount of money.
    Summary (1)
    Most help per euro (1)
    Donating in practice
    Inspirations from other associations (1)
    The association act is flexible (⅓)
    How to choose a target
    Comparing cause areas (1)
    Using research by evaluation organisations (⅓)
    Effective donation targets (summary ⅓)
    Global health and development (4)
    Givewell
    Maximum impact fund
    Preventing Malaria
    Vitamin A deficiency
    Vaccination coverage
    Deworming
    Direct cash transfers
    Climate change (2)
    The Founders Pledge Climate Change Fund
    Clean Air Task Force
    Carbon180
    Animal welfare (3.5)
    Animal Charity Evaluators
    Animal Charity Evaluatorsin Recommended Charity Fund
    The Humane League
    Faunalytics
    Wild Animal Initiative
    Securing the future (2)
    Johns Hopkins Center of Health Security
    ALLFED
    Nuclear Threat Initiative Biosecurity Program
    Background (½)
    Effective Altruism (½)
    Links (1)
    Put impact forward (essentially a call to action box) (1)
    References (2)
    Back cover (1)
    Language and approach of the guide
    We aimed for the giving guide to be
    Encouraging. The tone is hopeful and inviting, focusing on the opportunity of changing the world for the better together.
    Actionable. The guide includes simple legal advice on donating, inspiration from other student organisations on how to incorporate donations into the organisation’s activities, and recommended charities with links.
    Trustworthy. We explain why effective giving matters, how we have chosen the recommended cause areas and charities, provide references for further reading, and include a one-page about EA Helsinki.
    Easy to read.
    Project background
    The student organisation landscape and donating
    In Finland, it is common for students to be members of multiple student organisations and associations. E...
    16 min
  • EA - [linkpost] "What Are Reasonable AI Fears?" by Robin Hanson, 2023-04-23 by Arjun Panickssery
    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: [linkpost] "What Are Reasonable AI Fears?" by Robin Hanson, 2023-04-23, published by Arjun Panickssery on April 14, 2023 on The Effective Altruism Forum.
    Selected quotes (all emphasis mine):
    Why are we so willing to “other” AIs? Part of it is probably prejudice: some recoil from the very idea of a metal mind. We have, after all, long speculated about possible future conflicts with robots. But part of it is simply fear of change, inflamed by our ignorance of what future AIs might be like. Our fears expand to fill the vacuum left by our lack of knowledge and understanding.
    The result is that AI doomers entertain many different fears, and addressing them requires discussing a great many different scenarios. Many of these fears, however, are either unfounded or overblown. I will start with the fears I take to be the most reasonable, and end with the most overwrought horror stories, wherein AI threatens to destroy humanity.
    As an economics professor, I naturally build my analyses on economics, treating AIs as comparable to both laborers and machines, depending on context. You might think this is mistaken since AIs are unprecedentedly different, but economics is rather robust. Even though it offers great insights into familiar human behaviors, most economic theory is actually based on the abstract agents of game theory, who always make exactly the best possible move. Most AI fears seem understandable in economic terms; we fear losing to them at familiar games of economic and political power.
    He separates a few concerns:
    "Doomers worry about AIs developing “misaligned” values. But in this scenario, the “values” implicit in AI actions are roughly chosen by the organisations who make them and by the customers who use them. Such value choices are constantly revealed in typical AI behaviors, and tested by trying them in unusual situations."
    "Some fear that, in this scenario, many disliked conditions of our world—environmental destruction, income inequality, and othering of humans—might continue and even increase. Militaries and police might integrate AIs into their surveillance and weapons. It is true that AI may not solve these problems, and may even empower those who exacerbate them. On the other hand, AI may also empower those seeking solutions. AI just doesn’t seem to be the fundamental problem here."
    "A related fear is that allowing technical and social change to continue indefinitely might eventually take civilization to places that we don’t want to be. Looking backward, we have benefited from change overall so far, but maybe we just got lucky. If we like where we are and can’t be very confident of where we may go, maybe we shouldn’t take the risk and just stop changing. Or at least create central powers sufficient to control change worldwide, and only allow changes that are widely approved. This may be a proposal worth considering, but AI isn’t the fundamental problem here either."
    "Some doomers are especially concerned about AI making more persuasive ads and propaganda. However, individual cognitive abilities have long been far outmatched by the teams who work to persuade us—advertisers and video-game designers have been able to reliably hack our psychology for decades. What saves us, if anything does, is that we listen to many competing persuaders, and we trust other teams to advise us on who to believe and what to do. We can continue this approach with AIs."
    "If we assume that these groups have similar propensities to save and suffer similar rates of theft, then as AIs gradually become more capable and valuable, we should expect the wealth of [AIs and their owners] to increase relative to the wealth of [everyone else]. . . . As almost everyone today is in group C, one fear is of a relatively sudden transition to an AI-dominated economy. While perhaps not the most like...
    7 min