The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • EA - Learning as much Deep Learning math as I could in 24 hours by Phosphorous
    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: Learning as much Deep Learning math as I could in 24 hours, published by Phosphorous on January 8, 2023 on The Effective Altruism Forum.
    TL:DR I designed an experiment where I committed to spend two 12 hour days trying to learn as much deep-learning math as possible, basically from scratch.
    Table of Contents
    Origins and Motivations
    Results
    Takeaways
    Experiment set-up
    The Curriculum
    Documentation on hours
    Origins and Motivations
    For a long time, I’ve felt intimidated by the technical aspects of alignment research. I had never taken classes on linear algebra or multivariable calculus or deep learning, and when I cracked open many AI papers, I was terrified by symbols and words I didn’t understand.
    7 months ago I wrote up a short doc about how I was going to remedy my lack of technical knowledge: I collected some textbooks and some online courses, and I decided to hire a tutor to meet a few hours a week. I had the first two weeks of meetings, it was awesome, then regular meetings got disrupted by travel, and I never came back to it.
    When I thought about my accumulating debt of technical knowledge, my cached answer was “Oh, that might take six months to get up to speed. I don’t have the time.”
    Then, watching my productivity on other projects over the intervening months, I noticed two things:
    There appeared to be massive variance in my productivity. Sometimes, in a single day, I would get more done than I had accomplished in previous weeks.
    I seemed to both enjoy and get more done by “sprinting” through certain projects, eg. by spending 10 hours on it in a single day, rather than spreading that same work out over 2 hours a week for 5 weeks. It was, for some reason, way more motivating and seemingly more efficient to sprint.
    Also, when I asked myself what I thought the main bottlenecks were for addressing my technical debt problem, I identified two categories:
    Time (I felt busy all the time, and was afraid of committing too much to one project)
    A combination of lacking Motivation, Accountability and Fun
    Then, as my mind wandered, I started to put 2 and 2 together: Perhaps these new things I had noticed about my productivity, could be used to address the bottlenecks in my technical debt? I decided to embark on an experiment: how much technical background on deep learning could I learn in a single weekend? My understanding of the benefits of this experiment were as follows:
    Committing “a weekend” felt like a much smaller time cost than committing “a few months”, even if they were the same number of hours.
    No Distraction: I could design my environment to minimize distractions for two days, something it would be intractable to do to the same degree for several months.
    “Trying to learn as much as possible” felt like a challenge. I was, to be honest, pretty scared. I didn’t know what I was doing, it felt extreme, but that also made it exciting and fun.
    I had some historical data that I might be good at this kind of sprinting, and framing this as an experiment to see what I could learn about my productivity added another layer of discovery-driven motivation and fun. What if I learned more about how to be productive and get hard things done via this experiment?
    As far as I knew, nobody else among my peers had done this - but I suspected that more people than me had the same problems, and that if I conducted this experiment, I might learn things that would be helpful to others, which added yet another layer of discovery-driven motivation and fun.
    Accountability: Once I told somebody about this, it was hard to back out. It’s way easier for them to monitor me for a weekend than for a few months.
    Results
    I’d consider the experiment a success: I finished the whole curriculum in ~18 hours, and I got a lot of neat take-aways I’ll go over below.
    At the end of day 1, after 12 hours of cram...
    12 min
  • LW - Can Ads be GDPR Compliant? by jefftk
    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: Can Ads be GDPR Compliant?, published by jefftk on January 8, 2023 on LessWrong.
    I think the online ads ecosystem is most likely illegal in Europe, and as more decisions come out it will become clear that it can't be reworked to be within the bounds of the GDPR. This is a strong claim, but before I get into backing it up here's some background on me:
    I'm not a lawyer or an expert in privacy regulation; this is something I follow because I'm interested in it.
    I worked in ads until, but I'm speaking only for myself. I don't expect to go back into the industry.
    June 2022
    So, how are sites not compliant?
    When you visit a site in Europe, or an international site as a
    European, you'll typically see a prompt like this:
    In this screenshot El País is asking for permission to use cookies and use your data to personalize ads.
    Why are they asking you? A combination of two regulations:
    The ePrivacy
    Directive (2002) which requires the site to get your consent before using cookies or other storage on your device unless they're
    strictly necessary to provide a service you requested.
    The GDPR (2016) which tightly limits what companies can do with your data without your consent.
    The idea is, if you click "accept" then they can say they had your consent for all the advertising stuff they do. But I think it's very unlikely this is compliant with the GDPR.
    For example, in a recent case France's data privacy regulator CNIL recently fined
    Microsoft €60M (full text) for a similar popup on Bing. I'm going to come back to this decision later because it has other implications, but in paragraph 65 the CNIL ruled that their cookie banner was not collecting valid consent because it took more clicks to refuse cookies than to accept them.
    The principle here is that for consent to be valid under the GDPR it needs to be just as easy to give consent as it is to refuse it.
    This is not widely respected today, since for most companies it's going to be much more profitable to put up a not-really-legal banner that heavily pushes users towards saying yes and hope they don't get in trouble, but as the data protection agencies continue their enforcement I think this will become less practical.
    Another approach you see on a few sites is the one that Der Spiegel takes:
    They offer a choice between accepting their standard ad stuff or paying to subscribe to the site (more details). I'm glad they're giving users the choice here and I think this should be legal, but I'm pretty sure it isn't right now. The problem is that the user's consent isn't "freely given" in terms of the GDPR's Article
    4(11) if they would otherwise have to pay for access.
    The third option is to have a cookie banner that is as easy to reject as it is to accept:
    When I click "deny" and visit their site, they show a popup saying
    "Lower quality ads may be displayed." This includes (definitely low quality...) ads from Outbrain, with many network requests to outbrain.com and
    outbrainimg.com:
    The problem is, per the Schrems
    II ruling these are also not GDPR-compliant. Because US companies are required to share information with the US government and IP addresses are personal information, the GDPR requires sites to get consent from users before sending any of their information to American companies or their subsidiaries. European courts have applied this ruling to fine sites for using Google
    Analytics, Google
    Fonts, and the Akamai
    CDN. Since Outbrain is an American company, based in NYC, this is not compliant.
    Schrems II compliance rules out all commercially available adtech options I know about, and the only fully GDPR-compliant sites I've seen are ones where clicking "reject" means you don't get any ads at all.
    As a somewhat speculative aside, I think there's another problem with these consent popups: when you visit the site they read your cookies. Pe...
    11 min
  • LW - The Fountain of Health: a First Principles Guide to Rejuvenation by PhilJackson
    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: The Fountain of Health: a First Principles Guide to Rejuvenation, published by PhilJackson on January 7, 2023 on LessWrong.
    I’d like to live in a world where age really is just a number. For the last ~3 years or so, I’ve been nerding out with increasing fervour about longevity biotech, having originally been inspired by one of Aubrey de Grey's many interviews. I want to explain how and why I've become optimistic about the short-term prospects of indefinite healthy life extension, in terms of how I've come to think about ageing, disease, medicine and rejuvenation.
    A revolution is underway in how people think about health and medicine. Worst case scenario, it will unlock a whole new generation of therapies that can actually cure chronic age-related diseases, instead of just moderately slowing their progression at the expense of multiple poorly understood side effects. Best case scenario, it will enable us to live indefinitely in bodies that are functionally and cosmetically indistinguishable to those of 25 year olds. As with AGI, nobody can predict with certainty how long this will take, but optimism has grown tremendously over the past five years, and it is no longer unusual to encounter experts who believe that the ageing process is malleable.
    This timeframe depends massively on how rapidly we can divert funding and talent toward longevity biotechnology, because we are not going to achieve this outcome by doing more of what mainstream biomedicine has been doing. You could liken our position to where SpaceX are right now on the road to Mars: there's a long way to go still, but no Nobel prizes are required, no major scientific breakthroughs, just a lot of engineering. In this post, I will explain what ageing is, why we suck so much at treating chronic illnesses, and how this problem will be solved by a new kind of medicine, one that directly modifies the structure of your body so as to make you healthier.
    What is ageing?
    In biology it is said that function follows form, and in fact this is true of all machines. Machines do the things they do because of the way they are set up, because of their shape, because of their structure.
    Damage can be defined as any structural change that causes a loss of function.
    Any machine will do damage to itself in the course of normal operation.
    Ageing, then, is the process by which a machine loses function as a result of damage sustained from normal operation.
    The human body is a machine. Biological ageing is a progressive loss of function caused by damage that the body does to itself.
    Ageing is not a thing. There is no single process that is or causes ageing, it's just the body gradually breaking itself in many different ways.
    For this reason, I don't actually use the word "ageing" very much internally - it seems confusing and unnecessary to me. If you damage a machine then by definition it won't work as well as it did, and if you damage it too much then it'll stop working altogether. Doesn't matter if the damage is caused by something external or by the machine itself. The word "ageing" makes this sound more mysterious than it is, like there's some separate magical phenomenon that causes organisms to slowly decay and die. Ageing is a mysterious answer to a mysterious question, and you could argue that it belongs in the same dustbin as phlogiston or elan vital.
    I can grudgingly accept that we need a word for the general phenomenon by which people slowly decay and die with the passing of years, and ageing is that word. The problem is that it inclines us to imagine some single thing that causes decline. I'd accept "progressive loss of function due to intrinsic damage" as a reasonable definition of ageing, but only because it's clear about cause and effect. If you go looking for ageing as a cause of functional decline then you will search ...
    1 hr 9 min
  • EA - EA Germany's Strategy for 2023 by Sarah Tegeler
    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: EA Germany's Strategy for 2023, published by Sarah Tegeler on January 8, 2023 on The Effective Altruism Forum.
    Based on interviews with stakeholders, feedback from German community members and other national community builders, the new co-directors of EA Germany (EAD) drafted this strategy for 2023.
    Summary
    Our Vision is a diverse and resilient community of ambitious people in and from Germany who are thinking carefully about the world’s biggest problems and taking impactful action to solve them.
    Our Mission is to serve as a central point of contact for the German EA community and to continuously improve ways to guide people to take effective action, directly or by supporting local groups.
    Our Values are sustainable growth, a welcoming and nurturing culture and high professional standards.
    Our Focus Areas:
    EAD aims to guide people in Germany directly and indirectly to more impactful actions:
    Directly, e.g. through communications and events such as an EAGxBerlin 2023, career 1-1s, fellowships or retreats.
    Indirectly by training community builders, e.g. through regular calls, 1-1s and German-specific resources.
    EAD will offer efficiency services to save time and costs for committed EAs acting as an employer of record for individual grantees and providing fiscal sponsorship for local groups.
    A methodology of impact estimation based on a multi-touchpoint attribution model will serve as a basis for designing and prioritising exploratory programs using a lean startup approach.
    Background
    EA Germany (EAD)
    In the 2020 EA Survey, 7.4% of participants were from Germany, the third largest population behind the US and the UK. Apart from the US, Germany has the largest population and GNI of the ten largest countries in the survey. Germany has about 50 volunteer community builders in 25 local / university groups. 458 people have taken the GWWC pledge, and more than 400 Germans visited EAGxBerlin in September 2022. In 2021, Effektiv Spenden raised 18.86 Mio. Euros for effective charities.
    The registered association EA Germany was founded in 2019 as Netzwerk für Effektiven Altruismus Deutschland e.V. (NEAD) by EAs in Germany and has a board of volunteers. In parallel, one person on a national CEA Community Building Grant (CBG) worked independently from the association from 2020-22. The German website effektiveraltruismus.de was run by the national regranting organisation Effektiv Spenden. In late 2021, NEAD started offering employer-of-record services to grantees and EA organisations as well as fiscal sponsorship for local groups and hired a part-time operations associate on a CBG.
    A new board was elected in May 2022 and decided to apply for three CBGs – two co-directors and one project manager, in addition to the operations associate. The co-directors started in September and November 2022, and the project manager will start in January 2023. Funding for two other roles was promised but is not finalised as of December 2022. The association was renamed Effektiver Altruismus Deutschland (EAD) e.V. in 2022, and will now also run the website effektiveraltruismus.de.
    Epistemic Status
    Sarah Tegeler and Patrick Gruban drafted this document in November 2022 after having started working together as co-directors in the same month. Both have volunteered as local community builders, but this is their first role in an EA organisation. Most of the work on this document was influenced by interviews with stakeholders, other national EA community builders and reviews of different national strategies. About 150-200 hours went into discussing and writing the strategy.
    While the authors are confident the strategy will help foster a healthy community, their overall epistemic status is uncertain about the organisation’s and its programs’ counterfactual impact. Thus, many areas are not listed under founda...
    26 min
  • LW - Nothing New: Productive Reframing by adamShimi
    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: Nothing New: Productive Reframing, published by adamShimi on January 7, 2023 on LessWrong.
    In a comment on my last post, someone deplored the tendency of LW bloggers to rediscover ideas of famous philosophers and pretend that they discovered it first. This made me think of an interesting question of epistemology: is there value in reformulating and/or rediscovering things?
    A naive answer would be no— after all, we already have the knowledge. But a look at the history of science brings up many examples showing otherwise..
    In physics, the most obvious case comes from the formulations of classical mechanics. Newton gave the core of the theory in his original formulation, but what gets applied in modern science an play a key part in the revolutions of quantum mechanics and quantum electrodynamics among others, are the other two formulations: Lagrangian and Hamiltonian. And what these two, particularly the Lagrangian, pushed and pioneered the least action principle in physics, but capturing every system through an action that was maximized or minimized for every trajectory. This principle is one of the core building blocks of modern physics, and came out of a pure compression and reformulation of Newtonian mechanics!
    Similarly, potential energy started as a mathematical trick, the potential function of Lagrange which captured all gravitational forces acting on a system of N-bodies, which were the coefficients of its partial derivatives. Once again, there was nothing new here; even more than the Lagrangian and the action, this was a computational trick that sped up annoying calculations. And yet. And yet it’s hard to envision modern physics, even classical mechanics at all, without potential.
    In modern evolutionary biology, many advances like exaptation can be brought back to Darwin’s seminal work. And yet they often reveal different aspects of the idea, clarifying and bringing new intuitions to bear that make our models of life and its constraints that much richer.
    Again in theoretical computer science, reframing of the same core ideas is at the center of so many productive inquiries:
    all the models of computation unified by the Church Turing thesis, all equivalent yet providing different angles on questions of computations
    the non-deterministic complexity classes like NP reframed as proving/certifying classes, leading to far better intuitions and to interactive proofs and IP=PSPACE
    the Curry-Howard isomorphism shows that proof-calculi like Natural Deduction and type systems for computation models like the lambda calculus are actually the same thing. But this makes the two stronger, rather than weakening them for being the same thing.
    All in all, this pattern of rediscovering the same thing from a different perspective, of reframing the known, has been particularly productive in the history of science. Compressions, new formulations, new framing have delivered to us more than just the original insight.
    Why?
    Because we’re not logically omniscient — we don’t instantaneously know all the consequences of what we discover. Similarly, we’re not bayesian-omniscient — we don’t update on all the evidence available in the data. In both cases we are limited by what we can compute, what we can figure out. Reframings then are like shortcuts in the vastness of logical consequences, which highlight particularly interesting aspects of the evidence we originally unearthed.
    And I believe that one big reason for the misguided and simplistic models of science that so many share lies in the non-obviousness of these reframings’ power. We don’t remember them very well, we tend instead to ascribe the clarity and the power to the original thinker, even when we never used or touched their original formulation. We take for granted the intuitions and instincts that have been patiently built and bake...
    6 min
  • EA - EA university groups are missing out on most of their potential by Johan de Kock
    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: EA university groups are missing out on most of their potential, published by Johan de Kock on January 7, 2023 on The Effective Altruism Forum.
    The inception of the Purpose and Life Planning Track
    Context
    Many people think that Effective Altruism university groups are incredibly valuable. “To solve pressing global problems — like existential risk, global poverty, and factory farming - we need more talented, ambitious, altruistic people to focus full-time on these issues. Hundreds of thousands of these people are clustered at the world's top universities” (CEA, n.d.).
    I agree. However, I believe that most EA uni groups, and even the EA community as a whole, are missing out on the majority of their potential to make the most out of this opportunity. This is because the current paradigm for community building emphasises finding talented and ambitious people that want to tackle the world's most pressing problem, and not to create them. This strategy has potentially serious limitations which is preventing us from creating as much counterfactual impact as possible.
    TL;DR - A summary of the main points
    If EA university groups want to contribute as well as they can to empowering individuals to tackle the world's most pressing problems, we should not be cherry picking those students who are already naturally inclined to learn more about EA ideas. By only focusing on students who are already interested in EA ideas, we are missing the major opportunity to engage many more ambitious people to work on the world's most pressing problems, if approached from a different angle.
    Most university students are very young adults. Many are ambitious and conscientious but they are simply not at a point in their life where they have deeply internalised the desire to make doing good a core part of their life; If they don’t decide to join your introduction fellowship, or even drop out, it does not mean that they are not a good fit for EA. The life of university students is changing very quickly and there are many conflicting interests.
    Before people want to learn more about EA ideas and how to apply them to their lives they must regard this as valuable. Furthermore, before EA ideas can be properly internalised, the proper foundation must be laid.
    I identify four root causes, particularly for younger adults, that prevent an individual from being naturally inclined to EA ideas.
    First, people don't understand the link between being happy and doing good. Many people think that pursuing hedonic (feeling-based) happiness is the best way to live a happy life. Eudaimonic happiness (purpose-based happiness) tends to be more effective at this, however. People don’t know this. Making people aware of this difference might change their perspective on life and what they want to prioritise.
    Second, people often want to find a purpose in life, but it is not clear what that is and how to build one. Purpose consists to a large extent out of using your strengths to make the world a better place.
    Third, people have not internalised the underlying reasons about why doing good matters. Before somebody can be intrinsically motivated for something they need to understand why it is important and what the underlying reasons are. I think that we can do a lot better as a community to help people internalise these reasons.
    Four, learning about EA can be intimidating. Many EA ideas go against our evolutionary tendencies, such as prioritising our loved ones. Unless people have built a certain level of psychological and emotional resilience it is likely that taking EA ideas seriously is going to be too demanding.
    If these four points are addressed effectively it is possible to make a lot more people interested in learning about EA, and applying the ideas to their life (after initially being uninterested in your EA Introducti...
    49 min
  • LW - AI security might be a helpful ally for AI alignment by Igor Ivanov
    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 security might be a helpful ally for AI alignment, published by Igor Ivanov on January 6, 2023 on LessWrong.
    I am a person new to the AI alignment, and recently I am looking for a job at the AI governance field, but it seems like it is hard to find a long-term financially stable full-time job.
    For me, it basically means that only a tiny amount of people consider AI alignment important enough to pay money for the decrease of P(doom).
    It is important to get more people involved in the AI alignment field, but there is another good way to increase the number of people who might help achieve similar goals. We might cooperate with other groups of people, whose agenda is beneficial for AI alignment. One such group is the AI security community.
    AI security
    AI security is a branch of software security, and its goal is to make sure that current AI systems can not be exploited by hackers, are free of backdoors, and generally do things they are intended to do and do not do any unexpected and potentially harmful things.
    For example, there is a grocery store not far from my home, which uses cameras with facial recognition algorithms to identify a buyer and automatically charge money from their credit card. Malicious actors might use some kind of makeup or a mask to trick the algorithm. So, to prevent it, the operator of this payment system might hire an AI security engineer to simulate an attack on the model, find its vulnerabilities and fix them.
    Why this company should care that its facial recognition system has no security issues? Because if someone decides to hack it and steal money from its customer, it might become a huge scandal and lead to lawsuits, so it is relatively easy to convince top management to pay money for AI security. For them, it is just another branch of software security.
    A similar thing with governments. I think it is manageable to convince governments to implement policies and guidelines for safe and robust AI systems because we already have similar policies for other important software.
    AI security has a lot in common with the task of aligning AGI. There are few of them:
    It benefits from interpretable models.
    It requires the education of major decision-makers about risks of unsafe AI systems.
    Government and enterprise policies might require thorough testing and sertification of AI models, it is not only will make systems safer, but also might slow down rate of AI progress which is good for alignment.
    AI security also requires systematical audit of AI models, some kind of alarm system in case of an attack, and for containment its consequences.
    AI security researchers and organizations might help with exposing AI-associated risks to the general public. They might develop new secure and robust AI systems, they might become allies in policymaking.
    I further think that as technology will progress, AI security's goal of making sure that important and powerful AIs are doing what they are intended to do and do no harm to people will converge with the AI alignment goal of making sure that AGI will do what it is intended to do and not cause no harm to people.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    4 min
  • EA - Misha Yagudin and Ozzie Gooen Discuss LLMs and Effective Altruism by Ozzie Gooen
    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: Misha Yagudin and Ozzie Gooen Discuss LLMs and Effective Altruism, published by Ozzie Gooen on January 6, 2023 on The Effective Altruism Forum.
    Misha and I recently recorded a short discussion about large language models and their uses for effective altruists.This was mostly a regular Zoom meeting, but we added some editing and text transcription. After we wrote up the transcript, both Misha and myself edited our respective sections.
    I think the final transcript is clearer and contains more information than the original discussion. I might even suggest using text-to-speech on the transcript rather than listening to the original audio. This back-and-forth might seem to ruin the point of presenting the video and audio, but I think it might be straightforwardly more pragmatic.
    Transcript
    Sections
    Opening
    Introduction
    How do we use LLMs already?
    Could EAs contributing to applied LLMs be harmful?
    Potential LLM Application: Management and Emotional Assistance
    Potential LLM Application: Communication, Broadly
    Aside: Human-AI-Human Communication
    Potential LLM Application: Decision Automation
    Potential LLM Application: EA Forum Improvements
    Potential LLM Application: Evaluations
    LLM user interfaces
    What should EAs do with LLMs?
    Opening
    Ozzie: Hello. I just did a recording with my friend Misha, an EA researcher at ARB Research. This was a pretty short meeting about large language models and their use by effective altruists. The two of us are pretty excited about the potential for large language models to be used by effective altruists for different kinds of infrastructure.
    This is an experiment with us presenting videos publicly. Normally, our videos are just Zoom meetings. If anything, the Zoom meetings would be unedited. I found that to be quite a pain. These Zoom meetings typically don't look that great on their own, and they don't sound too terrific. So we've been experimenting with some methods to try to make that a little bit better.
    I am really curious about what people are going to think about this and am looking forward to what you say. Let's get right into it.
    Introduction
    Ozzie: For those watching us, this is Misha and me just having a meeting about large language models and their use for effective altruism.
    Obviously, large language models have been a very big deal very recently, and now there's a big question about how we could best apply them to EA purposes and what EAs could do best about it. So this is going to be a very quick meeting. We only have about half an hour.
    Right now, we have about seven topics. The main topic, though, is just the LLM applications.
    How do we use LLMs already?
    Ozzie: So, how do we use LLMs already?
    Misha: I think I use them for roughly 10 minutes on average per day.
    Sometimes I just ask questions or ask queries like, "Hey, I have these ingredients. What cocktails can I make?" Sometimes I try to converse with them about stuff. Sometimes I just use it (e.g., text-davinci-003) as a source of knowledge. I think it's more suitable for areas where verifiable expertise is rare.
    Take non-critical medicine, like skincare. I had a chat with it and got some recommendations in this domain, and I think it turned out really well. I previously tried to search for recommendations and asked a few people, but it didn't work.
    I also use it as an amplification for journaling whenever I'm doing any emotional or self-coaching work. Writing is great. I personally find it much easier to write “as if” I'm writing a message to someone—having ChatGPT obviously helps with that.
    Having a conversation partner activates some sort of social infrastructure in my brain. Humans solve math problems better when they are framed socially. And yeah, doing it with language models is straightforward and really good. Further, sometimes models give you hints or insights that yo...
    23 min
  • EA - Do short timelines impact the tractability of 80k’s advice? by smk
    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: Do short timelines impact the tractability of 80k’s advice?, published by smk on January 5, 2023 on The Effective Altruism Forum.
    Epistemic status: uncertain, looking to clarify my thinking on this
    Hello,
    80k recommends going into a graduate programme in machine learning to work on the alignment problem. For someone starting out studies, finishing a PhD will take at least 6-10 years including undergraduate/Master studies.
    Some put AGI take-off at ~4 years away, while the median Metaculus prediction for weak AGI is 2027 and the first superintelligence ~10 months after the first AGI. (The first public 'general AI' system is predicted in 2038, which makes me a bit confused. I fail to see how there's an 11 year gap between weak and 'strong' AI, especially with superintelligence ~10 months after the first AGI. Am I missing something?).
    To what extent is it still worthwhile to pursue a PhD in ML to work on the alignment problem when timelines are this short?
    Best, smk
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min
  • LW - Why didn't we get the four-hour workday? by jasoncrawford
    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: Why didn't we get the four-hour workday?, published by jasoncrawford on January 6, 2023 on LessWrong.
    John Maynard Keynes famously predicted in 1930 that by now we would only be working fifteen hours a week. What is less well-known is that his was nowhere near the only such prediction, nor the first—a wide range of commentators, including Charles Steinmetz and Buckminster Fuller, made similar forecasts. (And even Keynes’s prediction is generally misquoted.)
    Why didn’t any of them come true? I recently discussed this with Jason Feifer on his podcast Build for Tomorrow. Here’s some elaboration with more quotes and charts.
    The predictions
    A 1934 book, The Economy of Abundance, summarizes many of the predictions (Chapter 2):
    The technocrats promised every family on the continent of North America $20,000 a year [about $400,000 today], and a sixteen-hour work week. This is perhaps the peak of promises based on an abundance economy. Charles P. Steinmetz saw a two-hour working day on the horizon—he was the scientist who made giant power possible—but he stipulated no family budget total beyond “necessities and comforts.” .
    Fred Henderson, in his Economic Consequences of Power Production, is more specific: “Without any further increase in our knowledge of power and of technical processes, or of our available materials, we could multiply production ten times over if the needs of the world were permitted to express themselves in effective demand. . It would not be a question of an eight-hour day or a six-day week, but more probably of a six-months working year—which is already the rule for university dons.”
    Buckminster Fuller is still more definite. Modern man, he calculates, is 630 times more able than was Adam. Eliminating wasteful forms of work, four million Americans laboring fifty-two seven-hour days in the year (364 working hours, an average of one per day) “could keep up with every survival need”—meaning basic necessities for the whole population.
    Walter N. Polakov announces that “fifty weeks, four days, six hours is enough”—a twenty-four hour week and two weeks’ vacation.
    Harold Rugg in The Great Technology estimates a possible minimum living standard between ten- and twenty-fold greater than the minimums of 1929, on a sixteen- to twenty-hour work-week. .
    One can continue to cite such evidence indefinitely. Fortunately, A. M. Newman has been collecting it for years and saves us the trouble by the following summary: “Among them [such estimates] a substantial agreement is found that by better use of the mechanical facilities at our disposal we could produce many times our present supply of goods at considerably less effort.” The five-hour day tends to be the maximum estimate in Mr. Newman’s collection.
    As for Keynes, his essay “Economic Possibilities for our Grandchildren” wasn’t even saying that a fifteen-hour work week would be necessary for production. He thought it would be necessary to satisfy our psychological need for work, implying that our physical needs could be satisfied with less (exactly how much less, he doesn’t estimate):
    For many ages to come . everybody will need to do some work if he is to be contented. We shall do more things for ourselves than is usual with the rich today, only too glad to have small duties and tasks and routines. But beyond this, we shall endeavour to spread the bread thin on the butter—to make what work there is still to be done to be as widely shared as possible. Three-hour shifts or a fifteen-hour week may put off the problem for a great while.
    Why is the 40-hour work week still standard?
    Here are my hypotheses (not mutually exclusive):
    The predictions got the elasticity wrong
    When labor gets more productive, workers can choose to work less for the same real wage, to make more money by working the same amount, or something in betwee...
    10 min