The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - How do I start a programming career in the West? by Lao Mein
    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 do I start a programming career in the West?, published by Lao Mein on November 25, 2022 on LessWrong.
    I was expecting great things when I graduated from high school. I scored 35 on the ACT, was messing around with crypto, and fully expected to drop out of college to start a business. Things were looking up.
    Instead, due to a combination of depression and procrastination, I ended up spending the next 8 years drifting in and out of college before I finally got my undergraduate degree. I wish I was out partying during this time since that would at least give me some good memories, but I honestly just stayed inside and played video games all day. Not my proudest decade.
    But eventually, I pulled myself together and graduated. I've been employed as a bioinformatics engineer in China for the past 2 years, and have been steadily improving my life. I'm contributing to group discussions, pointing out major problems with research plans, and picking up programming and data management skills. Instead of putting in minimal effort, I'm spending most of my workday coding or researching, an incredible improvement over my productivity in college. I feel like a highly valued member of my team. I've published papers, helped design gene panels for cancer treatment, and sifted through papers for new research avenues for my team to pursue.
    And then I remembered that I was making what would be considered minimum wage in the US. I know I can do better. I also don't want the possibility of a multi-month lockdown starting at any time to hang over my head. I have a passion for AI and want to pursue a career in programming/data science in the West. Since many of you are involved in these fields, do you have any recommendations? I see 3 paths forward:
    Continue working for a few years. Pile up papers written, find character references, and acquire more job skills. My problem is that I don't think our prediction models are going to go beyond logarithmic regression, and I'm feeling increasingly stuck. I suppose this is my default path.
    Try to participate in a really impressive side project, and use that as a jump-off point to get into an AI alignment/capabilities company. The problem is that my personal passions might not match up with what such companies are looking for. I'm currently idea-boarding a cheap (<$20) laser-guidance package for quad-rotor-dropped grenades, but I think war is low-status in Western tech companies and this might actually hurt my chances. Maybe I should write an R package for automated primer design and put it on Github? It was a major pain in the ass to get the existing program in C working. I'm open to offers for collaboration.
    Just apply to something immediately. The problem is that I don't have anything to really impress my employers. I'm not the first author in any of my papers, and the one I'm proud to be associated with only lists me as an editor. I have a pretty good knowledge of cancer genomics, and have neat stories, like the time I helped improve a gene panel by pointing out that having an entirely age <30 control group would cause a lot of false positives from common benign age-related mutations. But they're just stories, not things that are legible to employers. Most of my code is <100 lines of using mutation data to make plots for ppts with ggplot. I did do most of the data analysis/graphics for a paper analyzing a subsection of TCGA data, but how do I even explain that to someone not working in the field? "I found that these genes correlated with X in patients with Y. I also made a plot of the same" sounds... not very impressive?
    I wanted to apply to this job offering from Conjecture, but then realized they were pretty far out of my league. On my resume, I'm just a guy with a bachelor's in biology.
    Transfer to a machine learning job in China. I was going...
    4 min
  • LW - Planes are still decades away from displacing most bird jobs by guzey
    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: Planes are still decades away from displacing most bird jobs, published by guzey on November 25, 2022 on LessWrong.
    Originally published here:/
    Note: Parts of this essay were written by GPT-3, so it might contain untrue facts.
    Introduction
    Many of my friends are extremely excited by planes, rockets, and helicopters. They keep showing me videos of planes flying at enormous speed, rockets taking off from the ground while creating fiery infernos around them, and of helicopters hovering midair seemingly denying the laws of gravity.
    I've been on a plane already, and it was nothing special. It was just a big metal tube with a bunch of people inside. It was loud and it smelled weird and I had to sit in a tiny seat for hours. So what is it that makes planes so special? Is it the fact that they're machine? Is it the fact that they're big? Is it the fact that they cost a lot of money?
    Here's the thing: all human-built artificial flight (AF) machines are incredibly specialized and are far away from being able to perform most of the tasks birds -- the only general flight (GF) machines we are aware of -- can perform.
    More than 200 years after hot air balloons became operational and more than 100 years after the first planes flew, it's clear that building a GF machine is much harder than anticipated and that we are nowhere close to reaching bird-level abilities.
    1. Planes vs eagles
    First, take a look at this video of an eagle catching a goat, throwing it off a cliff, and then feasting on it:
    I haven't ever seen a plane capable of catching a live animal and deliberately throwing it off a cliff. Not in 1922, not in 2022. Not even a tech demo. Such a feat vastly exceeds the abilities of any planes we have built, however fast they can fly.
    2. Planes vs cuckoos
    Second, let's watch this video of a cuckoo chick ejecting the eggs of its competitors out of a nest:
    You could say that this ability has nothing to do flight but, again, this misses the forest for the trees. Building a GF machine is not about Goodharting random "flight" benchmarks by flying high and fast, it's about real-world performance on tasks GF machines created by nature are capable of. And, however impressive planes are, as soon as we try to see how well they perform in the real-world, they can't even match a cuckoo chick.
    3. Planes vs a hummingbirds
    Third and final example. Take a look at the hummingbird's amazing ability to maintain stability in the harshest aerial conditions:
    Take any plane we have built and it stands no chance of survival placed in anything even close to these kinds of conditions, while a tiny-yet-mighty hummingbird doesn't break a sweat navigating essentially a tornado.
    Future of bird jobs: no plane danger
    Birds can flap their wings up to three times per second, whereas the fastest human-made aircraft only flaps its wings at 0.3 times per second. Birds can fly for long periods of time, whereas airplanes need to refuel regularly. Birds use orders of magnitude less energy to lift the same amount of mass in the air, compared to planes.
    Planes, rockets, and helicopters are (optimistically) decades away from being able to carry out most of the tasks birds are capable of. Therefore, for the foreseeable future, most bird jobs such as carrying messages (pigeons), carrying cargo (pigeons), hunting (hawks), and others, will remain safe from being displaced by human-built AF machines.
    Even if planes start to approach birds in some of their abilities, birds will be able to simply move towards performing other jobs. For example, planes can't navigate by themselves. So perhaps they will carry messages in simple conditions or to short distances, while pigeons will move towards specializing in complex message carrying or will learn to supervize plane routing, e.g. by piloting planes or by flying alongside and course...
    5 min
  • AF - Mechanistic anomaly detection and ELK by Paul Christiano
    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: Mechanistic anomaly detection and ELK, published by Paul Christiano on November 25, 2022 on The AI Alignment Forum.
    (Follow-up to Eliciting Latent Knowledge. Describing joint work with Mark Xu. This is an informal description of ARC’s current research approach; not a polished product intended to be understandable to many people.)
    Suppose that I have a diamond in a vault, a collection of cameras, and an ML system that is excellent at predicting what those cameras will see over the next hour.
    I’d like to distinguish cases where the model predicts that the diamond will “actually” remain in the vault, from cases where the model predicts that someone will tamper with the cameras so that the diamond merely appears to remain in the vault. (Or cases where someone puts a fake diamond in its place, or.)
    One approach to this problem is to identify (the diamond remains in the vault) as the “normal” reason for the diamond to appear on camera. Then on a new input where the diamond appears on camera, we can ask whether it is for the normal reason or for a different reason.
    In this post I’ll describe an approach to ELK based on this idea and how a the same approach could also help address deceptive alignment. Then I’ll discuss the empirical and theoretical research problems I’m most excited about in this space.
    ELK and explanation
    Explanations for regularities
    I’ll assume that we have a dataset of situations where the diamond appears to remain in the vault, and where that appearance is always because the diamond actually does remain in the vault. Moreover, I’ll assume that our model makes reasonable predictions on this dataset. In particular, it predicts that the diamond will often appear to remain in the vault.
    “The diamond appears to remain in the vault” corresponds to an extremely specific pattern of predictions:
    An image of a diamond is a complicated pattern of millions of pixels.
    Different cameras show consistent views of the diamond from different angles, suggesting that there is a diamond “out there in the world” being detected by the cameras.
    The position and physical characteristics of the diamond appear to be basically constant over time, suggesting that it’s “the same diamond.”
    In one sense the reason our model makes these predictions is because it was trained to match reality, and in reality the camera’s observations have these regularities. (You might call this the “teleological explanation.”)
    But we could also ignore the source of our model, and just look at it as a set of weights. The weights screen off the training process and so it should be possible to explain any given behavior of the model without reference to the training process.
    Then we ask: why does this particular computation, run on this distribution of inputs, produce this very specific pattern of predictions? We expect an explanation in terms of the weights of the model and the properties of the input distribution. (You might call this the “mechanistic explanation.”)
    Different predictors will give rise to this pattern in different ways. For example, a very simple predictor might have ontologically fundamental objects whose properties are assumed to be stable over time, one of which is a diamond. A more complex predictor might have a detailed model of physics, where object permanence is a complex consequence of photons reflecting from stable patterns of atoms, and the diamond is one such configuration of atoms.
    For a complex predictor like a physical simulation, we wouldn’t expect to be able to prove that the weights give rise to object permanence. That is, we don’t expect to be able to prove that on average if a diamond is present at t=0 it is likely to be present at t=1. But we do think that it should be possible to explain the pattern in a weaker sense.
    We don’t yet have an adequate notion of “explanati...
    33 min
  • EA - Could a single alien message destroy us? by Writer
    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: Could a single alien message destroy us?, published by Writer on November 25, 2022 on The Effective Altruism Forum.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min
  • EA - Clarifications on diminishing returns and risk aversion in giving by Robert Wiblin
    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: Clarifications on diminishing returns and risk aversion in giving, published by Robert Wiblin on November 25, 2022 on The Effective Altruism Forum.
    In April when we released my interview with SBF, I attempted to very quickly explain his views on expected value and risk aversion for the episode description, but unfortunately did so in a way that was both confusing and made them sound more like a description of my views rather than his.
    Those few paragraphs have gotten substantial attention because Matt Yglesias pointed out where it could go wrong, and wasn't impressed, thinking that I'd presented an analytic error as "sound EA doctrine".
    So it seems worth clarifying what I actually do think. In brief, I entirely agree with Matt Yglesias that:
    Returns to additional money are certainly not linear at large scales, which counsels in favour of risk aversion.
    Returns become sublinear more quickly when you're working on more niche cause areas like longtermism, relative to larger cause areas such as global poverty alleviation.
    This sublinearity becomes especially pronounced when you're considering giving on the scale of billions rather than millions of dollars.
    There are other major practical considerations that point in favour of risk-aversion as well.
    (SBF appears to think the effects above are smaller than Matt or I do, but it's hard to know exactly what he believes, so I'll set that aside here.)
    The offending paragraphs in the original post were:
    "If you were offered a 100% chance of $1 million to keep yourself, or a 10% chance of $15 million — it makes total sense to play it safe. You’d be devastated if you lost, and barely happier if you won.
    But if you were offered a 100% chance of donating $1 billion, or a 10% chance of donating $15 billion, you should just go with whatever has the highest expected value — that is, probability multiplied by the goodness of the outcome [in this case $1.5 billion] — and so swing for the fences.
    This is the totally rational but rarely seen high-risk approach to philanthropy championed by today’s guest, Sam Bankman-Fried. Sam founded the cryptocurrency trading platform FTX, which has grown his wealth from around $1 million to $20,000 million.
    The point from the conversation that I wanted to highlight — and what is clearly true — is that for an individual who is going to spend the money on themselves, the fact that one quickly runs out of any useful way to spend the money to improve one's well-being makes it far more sensible to receive $1 billion with certainty than to accept a 90% chance of walking away with nothing.
    On the other hand, if you plan to spend the money to help others, such as by distributing it to the world's poorest people, then the good one does by dispersing the first dollar and the billionth dollar are much closer together than if you were spending them on yourself. That greatly strengthens the case for taking the risk of receiving nothing in return for a larger amount on average, relative to the personal case.
    But: the impact of the first dollar and the billionth dollar aren't identical, and in fact could be very different, so calling the approach 'totally rational' was somewhere between an oversimplification and an error.
    Before we get to that though, we should flag a practical consideration that is as important, or maybe more so, than getting the shape of the returns curve precisely right.
    As Yglesias points out, once you have begun a foundation and people are building organisations and careers in the expectation of a known minimum level of funding for their field, there are particular harms to risking your entire existing endowment in a way that could leave them and their work stranded and half-finished.
    While in the hypothetical your downside is meant to be capped at zero, in reality, 'swinging for the fences' with...
    8 min
  • EA - Sentience Institute 2022 End of Year Summary by MichaelDello
    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: Sentience Institute 2022 End of Year Summary, published by MichaelDello on November 25, 2022 on The Effective Altruism Forum.
    Summary
    Sentience Institute is a 501(c)3 nonprofit think tank aiming to maximize positive impact through longtermist research on social and technological change, particularly moral circle expansion. Our main focus in 2022 has been conducting high-quality empirical research, primarily surveys and behavioral experiments, to build the field of digital minds research (e.g., How will humans react to AI that seems agentic and intentional? How will we know when an AI is sentient?). Our two most notable publications this year are a report and open-access data for our 2021 Artificial Intelligence, Morality, and Sentience (AIMS) survey and our paper in Computers in Human Behavior on “Predicting the Moral Consideration of Artificial Intelligences,” and we have substantial room for more funding to continue and expand this work in 2023 and beyond.
    AIMS is the first nationally representative longitudinal survey of attitudes on these topics. Even today with limited AI capabilities, we are already seeing the social relationship we have with these digital minds bearing on public discourse, funding, and other events in the trajectory of AI. The CiHB predictors paper is a deep dive into demographic and psychological predictors of AI attitudes, so we can understand why certain people view these topics in the way that they do. This follows up on our 2021 conceptual paper in Futures and literature review in Science and Engineering Ethics. We have also been working to build this new field through hosting a podcast, an AI summit at the University of Chicago, and a regular intergroup call between organizations working on this topic (e.g., Center on Long-Term Risk, Future of Humanity Institute).
    The urgency of building this field has been underscored this year in two ways in 2022. First, AIs are rapidly becoming more advanced, as illustrated in the amazing performance of image generation models — OpenAI’s DALL-E 2, Midjourney, and Stable Diffusion — as well as DeepMind’s Gato as a general-purpose transformer and high-performing language models Chinchilla and Google’s PaLM. Second, the topic of AI sentience had one of its first spikes in the mainstream news cycle, first as OpenAI chief scientist Ilya Sutskever tweeted “it may be that today's large neural networks are slightly conscious” in February, and then a much larger spike as Google Engineer Blake Lemoine was fired after claiming their language model LaMDA is “sentient.”
    In 2023, we hope to continue our empirical research as well as develop these findings into a digital minds “cause profile,” making the case for it as a highly neglected, tractable cause area in effective altruism with an extremely large scale. The perceptions, nature, and effects of digital minds are quickly becoming an important part of the trajectory of advanced AI systems and seem like they will continue to be so in medium- and long-term futures. This cause profile would be in part a more rigorous version of our blog post, “The Importance of Artificial Sentience.”
    2022 has been a challenging year for effective altruism funding. There was already an economic downturn when this month’s collapse of FTX, one of the two largest EA funders, left many EA organizations like us scrambling for funding and many other funders stretched thin. We expect substantially less funding in the coming months, and we need your help to continue work in this time-sensitive area, perhaps more than any other year since we were founded in 2017. We hope to raise $90,000 this giving season for our work on digital minds, such as surveys, experimental research, the cause profile, and other field-building projects. We will also continue some work on factory farming thanks to ge...
    13 min
  • AF - Refining the Sharp Left Turn threat model, part 2: applying alignment techniques by Victoria Krakovna
    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: Refining the Sharp Left Turn threat model, part 2: applying alignment techniques, published by Victoria Krakovna on November 25, 2022 on The AI Alignment Forum.
    A Sharp Left Turn (SLT) is a rapid increase in model capabilities (such as planning and world modeling). In a previous post, we broke down the SLT threat model into 3 claims:
    Capabilities will generalize far (i.e. to many domains)
    Alignment techniques that worked previously will fail during this transition
    Humans can’t intervene to prevent or align this transition
    We then proposed some possible mechanisms for Claim 1. This follow-up post will investigate possible arguments and mechanisms for Claim 2. In particular, we will outline our current understanding of the most promising plan for getting through an SLT and how it could fail (conditional on an SLT occurring).
    Plan: we use alignment techniques to find a goal-aligned model before SLT occurs, and the model preserves its goals during the SLT.
    We can try to learn a goal-aligned model before SLT occurs: a model that has beneficial goals and is able to reason about its own goals. This requires the model to have two properties: goal-directedness towards beneficial goals, and situational awareness (which enables the model to reason about its goals). Here we use the term "goal-directedness" in a weak sense (that includes humans and allows incoherent preferences) rather than a strong sense (that implies expected utility maximization).
    One can argue that the goal-aligned model has an incentive to preserve its goals, which would result in an aligned model after SLT. Since preserving alignment during SLT is largely outsourced to the model itself, arguments for alignment techniques failing during an SLT don't imply that the plan fails (e.g. it might be fine if interpretability or ELK techniques no longer work reliably during the transition if we can trust the model to manage the transition).
    Step 1: Finding a goal-aligned model before SLT
    We want to ensure that the model is goal-oriented with a beneficial goal and has situational awareness before SLT. It's important that the model acquires situational awareness at the right time: after it acquires beneficial goals. If situational awareness arises when the model has undesirable goals, this leads to deceptive alignment.
    Thus, our model search process would follow a decision tree along these lines:
    If situational awareness is detected without goal-directedness, restart the search.
    If undesirable goal-directedness or early signs of deceptive alignment are detected, restart the search.
    If an upcoming phase transition in capabilities is detected, and the model is not goal-aligned, restart the search.
    If beneficial goal-directedness is detected without situational awareness, train the model for situational awareness.
    Here are some ways alignment techniques could be applied in this plan:
    Use ELK techniques to identify whether the model has situational awareness (i.e. whether it believes it's inside a training process) and what goals it believes itself to have.
    Use interpretability techniques to pick up on early signs of deceptive alignment and restart and improve the search. (If we develop more advanced interpretability techniques that make it possible to retarget the internal search of the model, we can retarget a goal-directed model to a beneficial goal instead of restarting the search.)
    Use foundational progress on abstraction to determine whether the goals learned by the system are beneficial.
    Use foundational research insights (e.g. selection theorems) to make the search process more effective at avoiding models with undesirable goals and finding goal-aligned models
    Use insights from studying capability transitions to predict sharp increases in problematic capabilities, and make sure we only cross that threshold after we...
    11 min
  • EA - Some feelings, and what’s keeping me going by Michelle Hutchinson
    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: Some feelings, and what’s keeping me going, published by Michelle Hutchinson on November 25, 2022 on The Effective Altruism Forum.
    It seems weird to start a public blog post with how I’m feeling, because that seems so irrelevant to what’s happening right now, and to anyone who doesn’t know me. But a couple of people suggested it might be good to write a post like this. Also, over some of the past week I couldn’t be around others (covid), and found that pretty isolating and alienating while the FTX crisis was intensifying. For people who feel less surrounded by like-minded people in daily life, I thought it might be helpful to have some more voices on the forum expressing how we’re feeling. Sharing our coping strategies also seems particularly useful right now, given that many of us are currently dealing both with difficult emotions, more life uncertainty and greatly increased stress.
    In what follows, I’m not going to touch much on what specifically happened around FTX. I’ve haven’t been following the reporting as closely as others and don’t have a background in finance. I’m also not going to say that much about what I’ve learned from the debacle. I want to come back to what I should learn going forward when the situation is clearer, the most urgent response work has settled down and I can think more clearly. As usual for my posts on the forum, I’m writing in a personal capacity.
    [Written 19/11/2022]
    How I’m feeling
    I was hesitant to write about my feelings not just because they seem irrelevant, but also because it feels hard even to know how I’m feeling, let alone describe it. I also honestly really don’t want to inhabit a bunch of my emotions right now. But I’ll have a go at describing them here.
    Please don’t take any of these as in any way a prescription on how people should feel. Some of my friends feel angry and betrayed right now, some feel confused about their place in the world, some feel totally fine and largely unaffected by last week’s events. All of those sound totally reasonable to me.
    Here are a few of the kinds of things I’ve been feeling over the last week, to give a flavour of how much it feels like I’ve bounced around: Panic and urgency over figuring out how the things I’m responsible for need to be handled and done differently in light of this. Frustration and anger at how powerless I feel to really help. Anguish that whatever I tried to work on I somehow seemed to upset someone or let someone down. Gratitude for being able to go home to my son and how excited toddlers are to see you even if you can’t do anything more in the world than provide food and cuddles.
    I’ve been trying not to inhabit my feelings of sadness too much. There’s a lot to do, now of all times, so being too upset to contribute seems worth avoiding if I can. I’ve found two things particularly salient, and their sadness impossible to avoid. One is interpersonal conflict. It’s an extremely high stress time in which people are even busier than usual, so it’s easier than ever to cause friction, whether with friends/colleagues/online acquaintances. I’m also finding that harder to deal with than usual, I think because my refuge is usually caring relationships. The other sadness I’m finding impossible to avoid inhabiting is that some of my dearest friends are feeling deeply sad in a way I can do almost nothing to alleviate.
    Those are the things that I can’t get away from. But I know that in some light they don’t even seem that significant. There are so many sadnesses now that I really want to avoid looking at. One is sadness for the people who lost their money, despite thinking it was safely deposited rather than even being invested. Like a lot of us, I know people who lost money to this, including most of their life savings. Another is sadness for people who had been planning important e...
    10 min
  • EA - The 2022 Giving What We Can Donor Lottery is now open by Fabio Kuhn
    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 2022 Giving What We Can Donor Lottery is now open, published by Fabio Kuhn on November 25, 2022 on The Effective Altruism Forum.
    The 2022 Giving What We Can donor lottery (previously managed by EA Funds) is now open, with block sizes of $100k, $500k, and $2 million. To enter a lottery, or to learn more about donor lotteries, head to the Giving What We Can donor lottery page.
    Why enter a donor lottery?
    We’ve written before about why you should give to a donor lottery this giving season. If you’re intrigued about what a donor lottery is, or why you might want to enter, I’d recommend reading that article, including the comments (note that it’s from last year, so some specifics like block sizes and dates are different). I’ve copied in some of the key points below:
    A donor lottery allows you to turn your donation into a larger donation with some probability, while holding the expected donation amount constant. E.g., you can trade a $1,000 donation for a 1% chance of allocating $100,000 worth of donations. Your expected donation size stays constant at $1,000.
    If you win, it will be worth the time to think more carefully about where to allocate the money. Because extra time thinking is more likely to lead to better (rather than worse) decisions, this leads to more (expected) impact overall, even though your expected donation size stays the same.
    For this reason, we believe that a donor lottery is the most effective way for most smaller donors to give the majority of their donations, for those who feel comfortable with it.
    If you win, we can put you in touch with experienced grantmakers who can help you with the decision.
    You should only participate in a donor lottery if you think there’s a good chance you (or someone who you trust) will spend additional time thinking about your donation if you win.
    We also think there’s a good case for continuing to make some fraction of your donations directly, to keep engaged with EA donation opportunities.
    You can participate anonymously if you like.
    Continue reading the original article here.
    Practical information about the 2022/2023 Donor Lottery
    The Giving What We Can Donor Lottery is now open. The lottery will close to new entries on Monday, Jan 10 2023, 12:00 PM UTC. Any payments not confirmed by Giving What We Can by Monday, Jan 17, 2023, 12:00 PM UTC will not be accepted as entries.
    The lotteries will be drawn starting at Mon, Jan 24, 2023, 12:00 PM UTC (drawings for each block size will be spaced five minutes apart).
    There will be three block sizes:
    $100,000
    $500,000
    $2,000,000
    Which block you decide to enter is up to you (there are no minimum entry sizes on any of the blocks). If you’re not sure, we suggest that you aim to enter with a 1%-30% chance of winning.
    Donations are tax-deductible in the US, the UK, and the Netherlands. However, if you live somewhere else, you should still consider entering if you think the expected value of the lottery (including the potential to allocate winnings to projects that are more effective than the most effective charity that’s tax-deductible where you live) is a larger value-add than tax-deductibility.
    It is possible to participate anonymously, such that your personal details will only be visible to Giving What We Can and EV operational staff, even if you win. By default, all grants will be made public, unless winners or recipients request otherwise.
    For more in-depth information about the lottery process (including the important Caveats and Limitations section), please see the donor lottery website.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    4 min
  • EA - Part 1: The AI Safety community has four main work groups, Strategy, Governance, Technical and Movement Building by PeterSlattery
    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: Part 1: The AI Safety community has four main work groups, Strategy, Governance, Technical and Movement Building, published by PeterSlattery on November 25, 2022 on The Effective Altruism Forum.
    Epistemic status
    Written as a non-expert to develop and get feedback on my views, rather than persuade. It will probably be somewhat incomplete and inaccurate, but it should provoke helpful feedback and discussion.
    Aim
    This is the first part of my series ‘A proposed approach for AI safety movement building’. Through this series, I outline a theory of change for AI Safety movement building. I don’t necessarily want to immediately accelerate recruitment into AI safety because I take concerns (e.g., 1,2) about the downsides of AI Safety movement building seriously. However, I do want to understand how different viewpoints within the AI Safety community overlap and aggregate.
    I start by attempting to conceptualise the AI Safety community. I originally planned to outline my theory of change in my first post. However, when I got feedback, I realised that i) I conceptualised the AI Safety community differently from some of my readers, and ii) I wasn’t confident in my understanding of all the key parts.
    TLDR
    I argue that the AI Safety community mainly comprises four overlapping, self-identifying, groups: Strategy, Governance, Technical and Movement Building.
    I explain what each group does and what differentiates it from the other groups
    I outline a few other potential work groups
    I integrate these into an illustration of my current conceptualisation of the AI Safety community
    I request constructive feedback.
    My conceptualisation of the AI Safety community
    At a high level of simplification and low level of precision, the AI Safety community mainly comprises four overlapping, self-identifying, groups who are working to prevent an AI-related catastrophe. These groups are Strategy, Governance, Technical and Movement Building. These are illustrated below.
    We can compare the AI Safety community to a government and relate each work group to a government body. I think this helps clarify how the parts of the community fit together (though of course, the analogies are imperfect).
    Strategy
    The AI Safety Strategy group seeks to mitigate AI risk by understanding and influencing strategy.
    Their work focuses on developing strategies (i.e., plans of action) that maximise the probability that we achieve positive AI-related outcomes and avoid catastrophes. In practice, this includes researching, evaluating, developing, and disseminating strategy (see this for more detail).
    They attempt to answer questions such as i) ‘how can we best distribute funds to improve interpretability?’, ii) ‘when should we expect transformative AI?’, or iii) “What is happening in areas relevant to AI?”.
    Due to a lack of ‘strategic clarity/consensus’ most AI strategy work focuses on research. However, Toby Ord’s submission to the UK parliament is arguably an example of developing, and disseminating an AI Safety related strategy.
    We can compare the Strategy group to the Executive Branch of a government, which sets a strategy for the state and parts of the government, while also attempting to understand and influence the strategies of external parties (e.g., organisations and nations).
    AI Safety Strategy exemplars: Holden Karnofsky, Toby Ord and Luke Muehlhauser.
    AI Safety Strategy post examples (1,2,3,4).
    Governance
    The AI Safety Governance group seeks to mitigate AI risk by understanding and influencing decision-making.
    Their work focuses on understanding how decisions are made about AI and what institutions and arrangements help those decisions to be made well. In practice, this includes consultation, research, policy advocacy and policy implementation (see 1 & 2 for more detail).
    They attempt to answer questions such as i) ...
    12 min