The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • AF - On Developing a Mathematical Theory of Interpretability by Spencer Becker-Kahn
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: On Developing a Mathematical Theory of Interpretability, published by Spencer Becker-Kahn on February 9, 2023 on The AI Alignment Forum.
    If the trajectory of the deep learning paradigm continues, it seems plausible to me that in order for applications of low-level interpretability to AI not-kill-everyone-ism to be truly reliable, we will need a much better-developed and more general theoretical and mathematical framework for deep learning than currently exists. And this sort of work seems difficult. Doing mathematics carefully - in particular finding correct, rigorous statements and then finding correct proofs of those statements - is slow. So slow that the rate of change of cutting-edge engineering practices significantly worsens the difficulties involved in building theory at the right level of generality. And, in my opinion, much slower than the rate at which we can generate informal observations that might possibly be worthy of further mathematical investigation.
    Thus it can feel like the role that serious mathematics has to play in interpretability is primarily reactive, i.e. consists mostly of activities like 'adding' rigour after the fact or building narrow models to explain specific already-observed phenomena.
    My impression however, is that the best applied mathematics doesn’t tend to work like this. My impression is that although the use of mathematics in a given field may initially be reactive and disunited, one of the most lauded aspects of mathematics is a certain inevitability with which our abstractions take on a life of their own and reward us later with insight, generalization, and the provision of predictions. Moreover - remarkably - often those abstractions are found in relatively mysterious, intuitive ways: i.e. not as the result of us just directly asking "What kind of thing seems most useful for understanding this object and making predictions?" but, at least in part, as a result of aesthetic judgement and a sense of mathematical taste.
    One consequence of this (which is a downside and also probably partly due to the inherent limitations of human mathematics) is that mathematics does not tend to act as an objective tool that you can bring to bear on whatever question it is that you want to think about. Instead, the very practice of doing mathematics seeks out the questions that mathematics is best placed to answer. It cannot be used to say something useful about just anything; rather it finds out what it is that it can say something about.
    Even after taking into account these limitations and reservations, developing something that I'm clumsily thinking of as 'the mathematics of (the interpretability of) deep learning-based AI' might still be a fruitful endeavour. In case it is not clear, this is roughly speaking, because a) Many people are putting a lot of hope and resources into low-level interpretability; b) Its biggest hurdles will be making it 'work' at large scale, on large models, quickly and reliably; and c) - the sentiment I opened this article with - doing this latter thing might well require much more sophisticated general theory.
    In thinking about some of these themes, I started to mull over a couple of illustrative analogies or examples. The first - and more substantive example - is algebraic topology. This area of mathematics concerns itself with certain ways of assigning mathematical (specifically algebraic) information to shapes and spaces. Many of its foundational ideas have beautiful informal intuitions behind them, such as the notion that a shape my have enough space in it to contain a sphere, but not enough space to contain the ball that that sphere might have demarcated. Developing these informal notions into rigorous mathematics was a long and difficult process and learning this material - even now when it is presented in its ...
    11 min
  • EA - Animal Welfare - 6 Months in 6 Minutes by Zoe Williams
    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: Animal Welfare - 6 Months in 6 Minutes, published by Zoe Williams on February 8, 2023 on The Effective Altruism Forum.
    In August 2022, I started making summaries of the top EA and LW forum posts each week. This post collates together the key themes I’ve seen within top-rated animal welfare posts since then. (Note a lot of good work is happening outside what's posted on the forum too! This post doesn't try to cover that work.)
    If you’re interested in staying up to date on a more regular basis, consider subscribing to the Weekly EA & LW Forum Summaries, or to the Animal Advocacy Biweekly Digest. Forum announcements here and here.
    And for a great overview of the good the community is doing in this space, I highly recommend reading Big Wins for Farm Animals This Decade by Lewis Bollard.
    Key Takeaways
    A multi-proxy method has been suggested as a better option than neuron counts for estimating the moral weights of different species, with a first stab completed by Rethink Priorities based on empirical and philosophical research.
    There is an increasing focus on small animal welfare eg. fish, crustaceans and insects. This is particularly relevant for interventions which may cause substitution effects (consumers moving from one form of animal product to another).
    Wild animal welfare is becoming a more established cause area, with recent launches including WildAnimalSuffering.org and the NYU Wild Animal Welfare Program.
    Several major policy wins were achieved, including the first FDA approval of cultivated meat, and an EU announcement that it would put forward a proposal to end the systematic killing of male chicks.
    Themes
    Cross-Species Comparisons
    Moving beyond neuron counts
    Big strides have been made in modeling cross-species welfare comparisons.
    Rethink Priorities published the Moral Weight Project Sequence (led by Bob Fischer), which tackles philosophical and empirical questions related to the relative welfare capacities of 11 different farmed animals. This included looking at the evidence for 90 different hedonic and cognitive proxies in those animals, discussing why we shouldn’t just use neuron counts, and publishing a model of relative differences in the possible intensities of these animals’ pleasure and pains (relative to humans). You can see the results below - they suggest using these as the best-available placeholders until further research can be completed, and noting the translation from intensity of experience (welfare range) into ‘moral weight’ is dependent on several philosophical assumptions:
    Other work in this area has included:
    The launching of the NYU Mind, Ethics, and Policy Program, which will conduct and support foundational research about the nature and intrinsic value of nonhuman minds, including biological and artificial minds. There will be a special focus on invertebrates and AIs.
    The Shrimp Welfare Project (founded 2021) released research on using biological markers to measure the welfare of shrimp, and prioritize the practices causing the worst harms.
    MHR points out the issues with using neuron counts extend to the practical as well as the philosophical - the only publicly-available empirical reports of fish neuron counts sample exclusively from very small species (<1 g bodyweight), while many farmed fish are 1000x larger. This should add extra skepticism to neuron-count-based estimates of the moral weight of farmed fish.
    Rethink Priorities released research on welfare considerations for farmed black soldier flies - anticipated to become the most farmed insect species in the next decade.
    The danger of substitution effectsA team of entomologists and philosophers from Iran, the UK, and the USA found strong or substantial evidence for pain in adult insects of five orders, and ishankhire attempted to model cricket suffering per kg consumed and found t...
    9 min
  • AF - A note on terminology: AI Alignment != AI x-safety by David Scott Krueger
    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 note on terminology: AI Alignment != AI x-safety, published by David Scott Krueger on February 8, 2023 on The AI Alignment Forum.
    I think the terms "AI Alignment" and "AI existential safety" are often used interchangeably, leading the ideas to be conflated.In practice, I think "AI Alignment" is mostly used in one of the following three ways, and should be used exclusively for Intent Alignment (with some vagueness about whose intent, e.g. designer vs. user):1) AI Alignment = How to get AI systems to do what we want
    2) AI Alignment = How to get AI systems to try to do what we want
    3) AI Alignment = A rebranding of “AI (existential) safety”... A community of people trying to reduce the chance of AI leading to premature human extinction.
    The problem with (1) is that it is too broad, and invites the response: "Isn't that what most/all AI research is about?"The problem with (3) is that it suggests that (Intent) Alignment is the one-and-only way to increase AI existential safety.
    Some reasons not to conflate (2) and (3):
    The case that increasing (intent) alignment increases x-safety seems much weaker on the margin than in the limit; the main effect of a moderate increase in intent alignment might simply be a large increase in demand for AI.
    Even perfect intent alignment doesn't necessarily result in a safe outcome; e.g. if everyone woke up 1000000x smarter tomorrow, the world might end by noon.
    X-safety can be increased through non-technical means, e.g. governance/coordination.
    In my experience, this sloppy use of terminology is common in this community, and leads to incorrect reasoning (if not in those using it than certainly at least sometimes in those hearing/reading it).
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min
  • LW - EigenKarma: trust at scale by Henrik Karlsson
    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: EigenKarma: trust at scale, published by Henrik Karlsson on February 8, 2023 on LessWrong.
    Upvotes or likes have become a standard way to filter information online. The quality of this filter is determined by the users handing out the upvotes.
    For this reason, the archetypal pattern of online communities is one of gradual decay. People are more likely to join communities where users are more skilled than they are. As communities grow, the skill of the median user goes down. The capacity to filter for quality deteriorates. Simpler, more memetic content drives out more complex thinking. Malicious actors manipulate the rankings through fake votes and the like.
    This is a problem that will get increasingly pressing as powerful AI models start coming online. To ensure our capacity to make intellectual progress under those conditions, we should take measures to future-proof our public communication channels.
    One solution is redesigning the karma system in such a way that you can decide whose upvotes you see.
    In this post, I’m going to detail a prototype of this type of karma system, which has been built by volunteers in Alignment Ecosystem Development. EigenKarma allows each user to define a personal trust graph based on their upvote history.
    EigenKarma
    At first glance, EigenKarma behaves like normal karma. If you like something, you upvote it.
    The key difference is that in EigenKarma, every user has a personal trust graph. If you look at my profile, you will see the karma assigned to me by the people in your trust network. There is no global karma score.
    If we imagine this trust graph powering a feed, and I have gamed the algorithm and gotten a million upvotes, that doesn’t matter; my blog post won’t filter through to you anyway, since you do not put any weight on the judgment of the anonymous masses.
    If you upvote someone you don’t know, they are attached to your trust graph. This can be interpreted as a tiny signal that you trust them:
    That trust will also spread to the users they trust in turn. If they trust user X, for example, you too trust X—a little:
    This is how we intuitively reason about trust when thinking about our friends and the friends of our friends. Only EigenKarma being a database, it can remember and compile more data than you, so it can keep track of more than a Dunbar’s number of relationships. It scales trust. Karma propagates outward through the network from trusted node to trusted node.
    Once you’ve given out a few upvotes, you can look up people you have never interacted with, like K., and see if people you “trust” think highly of them. If several people you “trust” have upvoted K., the karma they have given to K. is compiled together. The more you “trust” someone, the more karma they will be able to confer:
    I have written about trust networks and scaling them before, and there’s been plenty of research suggesting that this type of “transitivity of trust” is a highly desired property of a trust metric. But until now, we haven’t seen a serious attempt to build such a system. It is interesting to see it put to use in the wild.
    Currently, you access EigenKarma through a Discord bot or the website. But the underlying trust graph is platform-independent. You can connect the API (which you can find here) to any platform and bring your trust graph with you.
    Now, what does a design like this allow us to do?
    EigenKarma is a primitive
    EigenKarma is a primitive. It can be inserted into other tools. Once you start to curate a personal trust graph, it can be used to improve the quality of filtering in many contexts.
    It can, as mentioned, be used to evaluate content.
    This lets you curate better personal feeds.
    It can also be used as a forum moderation tool.
    What should be shown? Work that is trusted by the core team, perhaps, or work trusted by the user accessing ...
    8 min
  • LW - GPT-175bee by Adam Scherlis
    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: GPT-175bee, published by Adam Scherlis on February 8, 2023 on LessWrong.
    Epistemic status: whimsical
    Bees: a new unit of measurement for ML model size
    Talking about modern ML models inevitably leads to a bunch of hard-to-intuit large numbers, especially when it comes to parameter count.
    To address this, we propose that we adopt a new, human-friendly unit to measure the number of learnable parameters in an architecture:
    1 beepower = 1 BP = 1 billion parameters
    Bees have about one billion synapses in their forebrain, so this gives a nice basis for comparisons between animal brains and artificial neural nets.
    Like horsepower and candlepower, the unit of beepower expresses the scale of a new and unfamiliar technology in terms that we are already familiar with. And it makes discussion of model specs flow better.
    "This model has twenty bees", you might say. Or "wow, look at all that beepower; did you train long enough to make good use of it?"
    Here's a helpful infographic to calibrate you on this new unit of measurement:
    Other animals
    We can even benchmark against more or less brainy animals.
    The smallest OpenAI API model, Ada, is probably 350 million parameters, or about a third of a bee, which is comparable to a cricket:
    The next size up, Babbage, is around 1.3 BP, or cockroach-sized.
    Curie has almost seven bees, which is... sort of in an awkward gap between insects and mammals.
    Davinci is a 175-bee model, which gets us up to hedgehog (or quail) scale:
    Gopher (280 BP) is partridge (or ferret) sized. More research into actual gophers is needed to know how many gophers worth of parameters Gopher has.
    Amusingly, PaLM, at 540 bees, has about as many parameters as a chinchilla has synapses:
    Tragically, we could not figure out how many palms worth of parameters Chinchilla (70 bees) has. We leave this as an exercise for the reader.
    There are about 170,000 neurons in the corpora pendiculata of a honeybee, or roughly 140,000 after adjusting for the tendency of optical fractionators to overcount, and some sources give about 7,000 synapses per neuron for the human brain, and it turns out humans and mice have comparable synapse-per-neuron counts so it doesn't scale that badly with brain size; skeptical readers are encouraged to shut up and multiply.
    Is one synapse equivalent to one parameter? Well, there are about five bits of recoverable information encoded in the strength of a synaptic connection, and neural-net parameters can be compressed to eight bits (or even 4 bits!) without too much loss of performance, so kinda-ish yeah.
    Wikipedia claims the corpora pendiculata ("mushroom bodies") of insects, which "are known to play a role in olfactory learning and memory", are analogous to the mammalian cerebral cortex or the avian hypopallium. Sure, why not.
    The thing about apples and oranges is that nobody can actually stop you from comparing them.
    A hundred-watt incandescent gets about a thousand candlepower per horsepower; LEDs can get 6,000 candles per horse or even more.
    Based on "forebrain" neuron numbers from this Wikipedia article, assuming 7,000 synapses per neuron, with optical fractionator counts discounted by a factor of about .8 based on the pairwise comparisons available there.
    Due to a lack of interest in studying chinchillas, there doesn't seem to have been a direct measurement of the synapse or neuron count for chinchillas. That being said, rabbits have around 500 billion synapses, and (domesticated) Chinchillas are around the same size and body weight as (smaller) rabbits and have the same cerebellum weight, so we feel justified in making this claim anyways. :)
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    4 min
  • EA - EigenKarma: trust at scale by Henrik Karlsson
    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: EigenKarma: trust at scale, published by Henrik Karlsson on February 8, 2023 on The Effective Altruism Forum.
    Upvotes or likes have become a standard way to filter information online. The quality of this filter is determined by the users handing out the upvotes.
    For this reason, the archetypal pattern of online communities is one of gradual decay. People are more likely to join communities where users are more skilled than they are. As communities grow, the skill of the median user goes down. The capacity to filter for quality deteriorates. Simpler, more memetic content drives out more complex thinking. Malicious actors manipulate the rankings through fake votes and the like.
    This is a problem that will get increasingly pressing as powerful AI models start coming online. To ensure our capacity to make intellectual progress under those conditions, we should take measures to future-proof our public communication channels.
    One solution is redesigning the karma system in such a way that you can decide whose upvotes you see.
    In this post, I’m going to detail a prototype of this type of karma system, which has been built by volunteers in Alignment Ecosystem Development. EigenKarma allows each user to define a personal trust graph based on their upvote history.
    EigenKarma
    At first glance, EigenKarma behaves like normal karma. If you like something, you upvote it.
    The key difference is that in EigenKarma, every user has a personal trust graph. If you look at my profile, you will see the karma assigned to me by the people in your trust network. There is no global karma score.
    If we imagine this trust graph powering a feed, and I have gamed the algorithm and gotten a million upvotes, that doesn’t matter; my blog post won’t filter through to you anyway, since you do not put any weight on the judgment of the anonymous masses.
    If you upvote someone you don’t know, they are attached to your trust graph. This can be interpreted as a tiny signal that you trust them:
    That trust will also spread to the users they trust in turn. If they trust user X, for example, you too trust X—a little:
    This is how we intuitively reason about trust when thinking about our friends and the friends of our friends. Only EigenKarma being a database, it can remember and compile more data than you, so it can keep track of more than a Dunbar’s number of relationships. It scales trust. Karma propagates outward through the network from trusted node to trusted node.
    Once you’ve given out a few upvotes, you can look up people you have never interacted with, like K., and see if people you “trust” think highly of them. If several people you “trust” have upvoted K., the karma they have given to K. is compiled together. The more you “trust” someone, the more karma they will be able to confer:
    I have written about trust networks and scaling them before, and there’s been plenty of research suggesting that this type of “transitivity of trust” is a highly desired property of a trust metric. But until now, we haven’t seen a serious attempt to build such a system. It is interesting to see it put to use in the wild.
    Currently, you access EigenKarma through a Discord bot or the website. But the underlying trust graph is platform-independent. You can connect the API (which you can find here) to any platform and bring your trust graph with you.
    Now, what does a design like this allow us to do?
    EigenKarma is a primitive
    EigenKarma is a primitive. It can be inserted into other tools. Once you start to curate a personal trust graph, it can be used to improve the quality of filtering in many contexts.
    It can, as mentioned, be used to evaluate content.
    This lets you curate better personal feeds.
    It can also be used as a forum moderation tool.
    What should be shown? Work that is trusted by the core team, perhaps, or work trusted by ...
    8 min
  • AF - Conditioning Predictive Models: Interactions with other approaches by Evan Hubinger
    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: Conditioning Predictive Models: Interactions with other approaches, published by Evan Hubinger on February 8, 2023 on The AI Alignment Forum.
    This is the fifth of seven posts in the Conditioning Predictive Models Sequence based on the paper “Conditioning Predictive Models: Risks and Strategies” by Evan Hubinger, Adam Jermyn, Johannes Treutlein, Rubi Hudson, and Kate Woolverton. Each post in the sequence corresponds to a different section of the paper. We will be releasing posts gradually over the course of the next week or so to give people time to read and digest them as they come out.
    5. Interactions with other approaches
    Imitation learning
    One very related approach is imitation learning: rather than try to predict humans, as we are proposing, one could simply train to imitate them instead. Such an approach would have many of the same safety benefits, since it would also be exclusively trying to produce outputs from safe humans.
    The basic problem with such an approach, however, is that there’s no reason to believe that a model trained via pure imitation learning would generalize beyond the capability level of the human(s) it is trained to imitate. While using predictive models to predict humans also cannot produce outputs that humans would never be able to generate, it can produce outputs that no humans that it has ever previously seen would be able to generate, since it might e.g. predict that such humans will exist under some conditional.
    Thus, we think that predictive modeling at least has the potential to be just as safe as imitation learning while being able to generalize to substantially more advanced capabilities—though, similarly to imitation learning, predicting humans still cannot elicit capabilities beyond those that any conceivable human would be capable of, as we discussed previously.
    Supervised fine-tuning
    For some conditionals we might have a very precise notion of what we want the model to observe (e.g. “exactly this image coming from this camera”). Ideally, this sort of a conditional should be straightforwardly implementable via prompting, just by fixing the relevant tokens in the model’s context window.[1] However, at least for current models, prompting has some basic structural limitations—for example, if you want to condition on something very long, context window length could start to become quite problematic. In that sort of a case, it might be quite helpful to instead turn to supervised fine-tuning, fine-tuning on the observation to condition on rather than including it in a prompt. Effectively, this sort of fine-tuning lets you give the model substantially more bits of evidence for it to condition on than is possible via just prompting.
    For the most part, we think this is likely to be basically fine, since it’s essentially continuous with pre-training: if we think that pre-training produces the sort of predictive model we want, then including some extra pre-training-style data and fine-tuning on it should do the same. The primary concern here, however, would be situations where the fine-tuning data is for some reason not very continuous with the pre-training data.
    One way that the fine-tuning data could be substantively different than pre-training is if it directly depends on the model itself—e.g. fine-tuning on the model’s own outputs. Not only is this substantially less continuous with pre-training, but it also specifically raises the risk of the model imitating AIs and/or producing self-fulfilling prophecies.
    Such fine-tuning could also be particularly problematic if the data is specifically selected according to some criterion other than actual representativeness of the world—that is, if there’s no clear “camera” that corresponds to how the data was collected. Probably the most notable way this could happen is via reinforcement learning ...
    18 min
  • EA - How will we know if we are doing good better: The case for more and better monitoring and evaluation by TomBill
    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 will we know if we are doing good better: The case for more and better monitoring and evaluation, published by TomBill on February 8, 2023 on The Effective Altruism Forum.
    Written by Sophie Gulliver and Thomas Billington
    TL;DR - EAs want to do the most good with the resources we have. But how do we know if we are actually doing the most good? How do we know if our projects are running efficiently and effectively? How do we know we are achieving impact? How do we check we are not causing harm? Monitoring and evaluation (M&E) theories and tools can help EA organisations answer these questions—but are currently not being applied to a sufficient degree. These theories and practices will help us achieve more impact and value in the short-term and long-term.
    Here, we outline a few ways to build M&E knowledge and skills within EA, including:
    Helpful resources
    The EA M&E Slack community group
    Signing up for our pro bono M&E support
    M&E as a career choice
    Introduction
    EA is “a project that aims to find the best ways to help others, and put them into practice”. Specifically, EA distinguishes itself from general do-gooding through its commitment to doing the most good possible with its resources.
    To these ends, the EA community is only as good at achieving its goal of “finding the best way to help others” as it is at knowing what the best way to help others is. But how can we know what the best ways to help others are? And how will we know if what we are doing is actually helping others in a cost-effective and impactful way?
    Monitoring and evaluation (M&E) experts, tools and concepts are some of the best ways to realise effective altruism’s philosophy of maximising impact and doing good better.
    M&E are two interrelated concepts that help us track and assess a project or organisation’s progress, impact, and value. In essence, if you really care about whether a project is making the world better, then you should also care about M&E.
    However, in our experience, the integration of M&E tools and expertise within the EA community has been variable and mostly restricted to the global health and development sector.
    This post argues that the broader EA movement should also be engaging with M&E more to ensure we are doing the most good possible.
    Note: This post aims at being a broad overview of M&E. Our objective is to increase knowledge and awareness of M&E in the EA space, as well as give some first steps to learn more. Don't worry if you feel like you can't apply it immediately. Future posts will delve into details, and see our section below for what you can do right now.
    What is M&E?
    Monitoring and evaluation are two distinct functions that work synergistically. They can be defined as:
    Monitoring: the systematic and routine collection of information to track progress and identify areas for improvement. Monitoring asks questions like:
    Are we on track?
    Are we reaching the right people?
    Are we using our money and time efficiently?
    What can we improve?
    For example, if you were running a project to reduce deaths through improved water quality, you might regularly monitor chlorine availability at water points, or run quarterly surveys asking community members how often they are chlorinating their water.
    Evaluation: the rigorous assessment of the value of a project or programme to inform decision-making. This assessment usually measures the performance of the project against criteria that define what a ‘valuable’ project looks like. These could be criteria like ‘impactful’ or ‘cost-effective’ or ‘sustainable’. Unlike monitoring, evaluations answer bigger questions to inform important decisions about a project’s future:
    How well is this project doing overall?
    How valuable is it?
    Is it worth it?
    Should we adapt, scale up or scale down?
    Given the deeper analysis required, evaluations are t...
    16 min
  • EA - Solidarity for those Rejected from EA Global by Wil Perkins
    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: Solidarity for those Rejected from EA Global, published by Wil Perkins on February 8, 2023 on The Effective Altruism Forum.
    This time of year, there are surely many folks out there upset they didn’t get into EA Global (EAG). As the year continues, there will be plenty more people who get rejected, and feel hurt. I am one of them.
    I'm writing this post to briefly share my experience, and give space for others to vent and commiserate about not being able to attend EAG.
    As a quick aside, I understand that not everyone can get into EAG due to the high standards the admittance committee has. (at least with the way it’s currently structured.) I also understand that with funding constraints many people had to be rejected, almost definitely more than in previous years. I’m sure that it’s not easy being on the EAG team and making these decisions.
    That being said, as many of us know, being rejected from EAG can be miserable. Last year, after reading and learning about Effective Altruism for many years, I upended my entire life to work towards having a larger impact. I started an EA group in my local city. I quit my lucrative job and joined an early stage, mission driven AI startup, working for free for months until we secured funding. All this in the hopes I would have more impact and be able to give more to the EA community, and the world.
    Unfortunately despite all of this effort, I was rejected from EAG. Surprisingly I got in last year, despite being much less involved and having less potential impact from my own perspective. It stings, and I’m frustrated. I don’t blame the people making the decision as I’m sure they had good reasons not to accept my application. But it still hurts. It feels like I devoted hundreds of hours of my life and tied my identity to a group, only to be told I wasn’t good enough.
    As I said, I want to invite others feeling the same way to comment. I don’t want to encourage destructive or vindictive dog piling on CEA or the EAG team, but I do think it’s important to share what a rejection from EAG means to people.
    I'd also like to encourage people who did get invited to EAG, or look down on this type of post as complaining, to try and have charity towards folks like me. A bit of empathy can go a long way.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • LW - Living Nomadically: My 80/20 Guide by KatWoods
    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: Living Nomadically: My 80/20 Guide, published by KatWoods on February 8, 2023 on LessWrong.
    I’ve been living nomadically for three years, and I’m often asked what my advice is for people trying it. Here’s the 80/20 of all my advice:
    Work out of restaurants and 5-star resorts. They’re gorgeous and if you pay for a meal, ~99.9% of restaurants and resorts are fine with you working there.
    Ask ~20 different AirBnBs for a 40% discount if you stay for a month. ~5-10% will say yes, then you’ll get to stay in way nicer places for cheaper.
    Travel with a folding bike. It will help you get around and really see the cities.
    Use NomadList.com to find the best places to stay. It allows you to search cities by cheapness, safety for women, weather, number of nomads, etc. Nomad List is the best and I wish everything were Nomad List.
    Social contact is the main reason people stop nomading. To fix that a) travel with at least one friend/partner and b) become excellent at location-independent friendships. The main way to do this is to set up recurring calls. Make it the default to talk to people. If you have to choose to hangout with people each time, you will just forget to. You should probably already be doing this. Full short article on how to do this well.
    Some of the best places to nomad, according to Nomad List / me
    Bali (particularly Canggu and Ubud). Tropical paradise, tons of nomads, insanely cheap and beautiful living arrangements.
    Thailand (particularly Chiang Mai). Same as Bali.
    Istanbul. The Mediterranean with mosques and cats. (So many cats!) Very safe for women, filled with history. First world infrastructure and third world prices. Better time zone for working remotely in Europe.
    Buenos Aires. European style city, very cheap and safe. Better time zone for working remotely from North America
    Medellín. Jungle in a city. Cheap and gorgeous. Better time zone for working remotely from North America
    Other articles you might like:
    How to maintain long-distance friendships without losing touch
    App and book recommendations for people who want to be happier and more productive
    EA Houses: the AirBnB of EA
    Cross-posted from my personal blog
    Reminder that you can listen to this post on your podcast player using the Nonlinear Library if it reaches above 35 upvotes
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min