The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - Short timelines and slow, continuous takeoff as the safest path to AGI by rosehadshar
    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: Short timelines and slow, continuous takeoff as the safest path to AGI, published by rosehadshar on June 21, 2023 on LessWrong.
    There are various different timelines/takeoff dynamics scenarios:
    In this post, we’re going to set out our understanding of the case for short timelines and slow, continuous takeoff as the safest path to AGI.
    A few reasons why laying out this case seems worth doing:
    OpenAI’s strategy assumes that short timelines and slow, continuous takeoff are the safest path to AGI. Whether or not this is the safest path, the fact that OpenAI thinks it’s true and is one of the leading AI labs makes it a path we’re likely to take. Humanity successfully navigating the transition to extremely powerful AI might therefore require successfully navigating a scenario with short timelines and slow, continuous takeoff.
    Short timelines and slow, continuous takeoff might indeed be the safest path to AGI. After spending some time trying to state the strongest version of the case, we think it’s stronger than we initially expected (and than many in the AI safety community seem to assume).
    If it’s not true that short timelines and slow, continuous takeoff are the safest path to AGI, it might be really important to convince people at OpenAI that this is the case. It seems unlikely that it’s possible to do this without deeply understanding their current view.
    The arguments we’re going to lay out are:
    Slow, continuous takeoff is safer than fast, discontinuous takeoff
    Slow, continuous takeoff is more likely given short timelines than long ones
    Time during takeoff is more valuable than time before, so it’s worth trading time now for time later
    It’s possible to safely navigate short timelines and slow, continuous takeoff
    Note that Sam Altman usually refers just to slow takeoff rather than slow continuous takeoff, but in this piece we discuss slow continuous takeoff, because we think that it’s the strongest way of making this case and that it’s likely that this is what Altman means.
    Disclaimers:
    Neither of us work at OpenAI or are in touch with OpenAI leadership. This post represents our best guess charitable interpretation of the case. We’ve done a lot of filling in the blanks on our side in order to make the arguments as coherent as possible. Our ideal in writing this post is to pass an ideological Turing Test, where Sam Altman could read this post and think ‘yes, exactly, that’s why I think that’. Probably we fall short of this as we haven’t engaged with Sam directly.
    Neither of us fully buys that short timelines and slow, continuous takeoff are the safest path to AGI. We’re writing this post to improve our (and hopefully other people’s) understanding of the position. In this post, we don’t give our overall evaluation of the case - just present what seem to us like the strongest arguments for it.
    We hope this post will serve as a jumping off point for critiques which engage more deeply with OpenAI’s strategy. (And where the post doesn’t faithfully capture OpenAI views, we’d love to get comments pointing this out.)
    Slow, continuous takeoff is safer than fast, discontinuous takeoff
    To successfully navigate the transition to extremely powerful AI, we want AI safety and governance efforts to keep pace with AI capabilities (or ideally, to exceed them). When compared to fast discontinuous takeoff, slow continuous takeoff seems much safer from this perspective:
    Fast takeoff doesn’t give society time to figure out how to respond to advanced capabilities and the destabilisation which could follow from them. It also doesn’t give us to coordinate once the danger is clearer and more imminent.
    Discontinuous takeoff entails jumps where capabilities could suddenly exceed existing levels of safety/governance.
    Figuring out how to make systems safe will likely depend on bootstrapping up safety...
    15 min
  • EA - Upcoming speaker series on emerging tech, national security & US policy careers by kuhanj
    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: Upcoming speaker series on emerging tech, national security & US policy careers, published by kuhanj on June 21, 2023 on The Effective Altruism Forum.
    There is an upcoming virtual speaker series on emerging tech, national security, and US policy careers and wanted to share this opportunity. I’ve heard some of these speakers before think the series could be really helpful for anyone interested in working on AI/bio policy in the US. I've pasted the announcement of the speaker series below.
    Summer webinar series on emerging technology & national security policy careers
    The Horizon Institute for Public Service, in collaboration with partners at the Scowcroft Center for International Affairs at the Texas A&M Bush School and SeedAI, is excited to announce an upcoming webinar series on US emerging technology policy careers to help individuals decide if they should pursue careers in this field. In line with Horizon’s and our partners’ focus areas, the series will focus primarily on policy opportunities related to AI and biosecurity and run from late June to early August.
    Sessions will not be recorded and individuals must sign up to receive event access — you can express interest in attending here.
    Horizon’s mission is to help the US government navigate our era of rapid technological change by fostering the next generation of public servants with emerging technology expertise. The policy opportunities and challenges related to emerging technology are interdisciplinary and will require talent from a range of backgrounds and communities, including many that don’t have ready access to information about what policy work is like and what a career transition might look like. As a result, this series will cover:
    Examples of individuals with non-traditional (e.g. technical or legal) backgrounds transitioning into emerging technology policy
    Examples of what a “day in a life” is like at a think tank, advocacy organization, executive agency, or in Congress
    Examples of ongoing policy efforts and debates related to AI or biosecurity policy
    All sessions will involve interactive conversations with experienced policy practitioners and opportunities for audience questions. Some of the sessions will be useful for individuals from all fields and career stages, while others are more focused — you may choose to attend all or only some of the sessions. Currently scheduled sessions include:
    Q&A with Jason Matheny, CEO of the RAND Corporation
    Q&A with Nikki Teran, Institute for Progress Fellow, on biosecurity challenges and policy careers to address them
    Q&A with Helen Toner‚ Director of Strategy and Foundational Research Grants at CSET, on AI challenges and policy careers to address them
    What’s it like working in the executive branch?
    What’s it like working in a think tank?
    What’s it like working in Congress?
    Choosing graduate schools for policy careers (master’s, PhD, JD)
    Transitioning from law to policy
    Transitioning from science and tech to policy
    Advancing in policy from underrepresented backgrounds
    Horizon Fellowship info session
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org
    4 min
  • EA - Five Years of Rethink Priorities: What We've Learned by Peter Wildeford
    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: Five Years of Rethink Priorities: What We've Learned, published by Peter Wildeford on June 21, 2023 on The Effective Altruism Forum.
    The post contains a reflection on our journey as co-founders of Rethink Priorities. We are Peter Wildeford and Marcus A. Davis.
    In 2017, we were at a crossroads.
    We had been working on creating new global health and development interventions, co-founding an organization that used text message reminders to encourage new parents in India to get their children vaccinated.
    However, we felt there was potentially more value in creating an organization that would help tackle important questions within cause and intervention prioritization. We were convinced that farmed and wild animal welfare were very important, but we didn’t know which approaches to helping those animals would be impactful. Hits-based giving seemed like an important idea, but we were unsure how to empirically compare that type of approach to the mostly higher-certainty outcomes available from funding GiveWell’s top charities.
    So, we chose to create a research organization. Our aim was to take the large amount of evidence base and strong approaches used to understand global health interventions and apply them to other neglected cause areas, such as animal welfare and reducing risks posed by unprecedented new technologies like AI. We wanted to identify neglected interventions and do the research needed to make them happen.
    Five years later, Rethink Priorities is now a research and implementation group that works with foundations and impact-focused non-profits to identify pressing opportunities to make the world better, figures out strategies for working on those problems, and does that work.
    Reflecting on everything the organization has accomplished and everything we want to happen in the next five years, we’re proud of a lot of the work our team has done.
    For example, we went from being unsure if invertebrates were capable of suffering to researching the issue and establishing invertebrate welfare as a proposition worth taking seriously. Following through, we helped create some of the first groups in the effective animal advocacy space working on interventions targeting invertebrates. Our team did the deep philosophical work and the practical research needed to establish specific interventions, and we incubated groups to implement them.
    Building on this work, our ambitious Moral Weight Project improved our understanding of both capacity for welfare and intensity of valenced experiences across species, and the moral implications of those possible differences. By doing so, the Moral Weight Project laid the foundation for cross-animal species cost-effectiveness analyses that inform important decisions regarding how many resources grantmakers and organizations should tentatively allocate towards helping each of these species.
    We have also produced dozens of in-depth research pieces. Our global health and development team, alone, has produced 23 reports commissioned by Open Philanthropy that increased the scope of impactful interventions considered in their global health and development portfolio. This work has influenced decisions directing millions of dollars towards the most effective interventions.
    Our survey and data analysis team also worked closely with more than a dozen groups in EA including the Centre for Effective Altruism, Open Philanthropy, 80,000 Hours, and 1 Day Sooner to help them fine-tune their messaging, improve their advertising, and have better data analysis for their impact tracking.
    RP has provided 23 fellowships to aspiring researchers, building a robust talent pipeline. Many talented people have remained in successful careers at Rethink Priorities. RP staff have gone on to work at Open Philanthropy, Founders Pledge, the Centre for Effective Altruism, 80,00...
    20 min
  • EA - Concrete projects for reducing existential risk by Buhl
    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: Concrete projects for reducing existential risk, published by Buhl on June 21, 2023 on The Effective Altruism Forum.
    This is a blog post, not a research report, meaning it was produced quickly and is not to Rethink Priorities’ typical standards of substantiveness and careful checking for accuracy.
    Super quick summary
    This is a list of twenty projects that we (Rethink Priorities’ Existential Security Team) think might be especially promising projects for reducing existential risk, based on our very preliminary and high-level research to identify and compare projects.
    You can see an overview of the full list here. Here are five ideas we (tentatively) think seem especially promising:
    Improving info/cybersec at top AI labs
    AI lab coordination
    Field building for AI policy
    Facilitating people’s transition from AI capabilities research to AI safety research
    Finding market opportunities for biodefence-relevant technologies
    Introduction
    What is this list?
    This is a list of projects that we (the Existential Security Team at Rethink Priorities) think are plausible candidates for being top projects for substantially reducing existential risk.
    The list was generated based on a wide search (resulting in an initial list of around 300 ideas, most of which we did not come up with ourselves) and a shallow, high-level prioritization process (spending between a few minutes and an hour per idea). The process took about 100 total hours of work, spread across three researchers. More details on our research process can be found in the appendix. Note that some of the ideas we considered most promising were excluded from this list due to being confidential, sensitive or particularly high-risk.
    We’re planning to prioritize projects on this list (as well as other non-public ideas) for further research, as candidates for projects we might eventually incubate. We’re planning to focus exclusively on projects aiming to reduce AI existential risk in 2023 but have included project ideas in other cause areas on this list as we still think those ideas are promising and would be excited about others working on them. More on our team’s strategy here.
    We’d be potentially excited about others researching, pursuing and supporting the projects on this list, although we don't think this is a be-all-end-all list of promising existential-risk-reducing projects and there are important limitations to this list (see “Key limitations of this list”).
    Why are we sharing this list and who is it for?
    By sharing this list, we’re hoping to:
    Give a sense of what kinds of projects we’re considering incubating and be transparent about our research process and results.
    Provide inspiration for projects others could consider working on.
    Contribute to community discussion about existential security entrepreneurship – we’re excited to receive feedback on the list, additional project suggestions, and information about the project areas we highlight (for example, existing projects we may have missed, top ideas not on this list, or reasons that some of our ideas may be worse than we think).
    You might be interested in looking at this list if you’re:
    Considering being a founder or early employee of a new project. This list can give you some inspiration for potential project areas to look into. If you’re interested in being a (co-)founder or early employee for one of the projects on this list, feel free to reach out to Marie Buhl at [email protected] so we can potentially provide you with additional resources or contacts when we have them.
    Note that our plan for 2023 is to zoom in on just a few particularly promising projects targeting AI existential risk. This means that we’ll have limited bandwidth to provide ad hoc feedback and support for projects that aren’t our main focus, and that we might not be able to respond to ev...
    37 min
  • LW - My side of an argument with Jacob Cannell about chip interconnect losses by Steven Byrnes
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: My side of an argument with Jacob Cannell about chip interconnect losses, published by Steven Byrnes on June 21, 2023 on LessWrong.
    Context / How I came to write this
    Jacob Cannell (@jacob_cannell) made some claims about fundamental limits of interconnect losses on chips at Brain Efficiency: Much More Than You Wanted To Know, and in particular claimed that current chips have interconnect losses close to as low as they will ever get.
    When I read that claim I thought it was wrong, as was Jacob’s discussion of interconnect losses more generally, but I didn’t (and still don’t) think the disagreement directly mattered for AI x-risk, so I mostly didn’t want to spend time arguing about it.
    But then later Alexander Gietelink Oldenziel wrote $250 prize for checking Jake Cannell's Brain Efficiency, and I wound up in a 15-comment back-and-forth with Jacob about it, before ducking out. (Other people continued that thread afterwards).
    Unfortunately, I quit the discussion while still being confused about where Jacob was coming from. So this post will not be maximally good and useful, sorry. Nevertheless, here’s a summary of my current perspective and understanding, in case anyone cares.
    I believe that Jacob plans to write at least one response comment in the comment section at the bottom of this post; if he hasn’t yet, you should check back later.
    (Jargon level: medium-low maybe? There is still some unexplained physics & EE jargon, but hopefully I made the centrally important parts accessible to non-experts. DM or email me if something is confusing, and I will try to fix it.)
    (All numbers in this post should be treated as Fermi estimates.)
    (Thanks very much to Jacob for his extraordinary patience in trying to explain to me his perspective on this topic. And also his perspective on many other topics!)
    Background to the technical disagreement
    “Interconnects” send information from one point to another on a chip.
    The fundamental thermodynamic limit for the power required to send a bit of information from point A to point B is 0. As a stupid example, there is a lot of digital information on Earth, and it all travels 1e12 meters in orbit around the sun each year for roughly zero energy cost.
    Chip interconnect losses are obviously much much higher than the thermodynamic limit of “zero”—they might even constitute a majority of chip power consumption these days. Everyone knows that, and so does Jacob. So what is he saying?
    I think Jacob divides the world of interconnects into two categories, “reversible” and “irreversible” interconnects, with the former including optical interconnects and superconducting wires, and the latter including normal wires and brain axons. (I’m stating this categorization without endorsing it.)
    I think Jacob takes “reversible” interconnects (optical interconnects & superconducting wires) to have a fundamental interconnect loss limit of zero, but to have practical limits such that we’re not expecting to cut orders of magnitude from the total interconnect loss budget this way. I agree with his conclusion here, although we had some disagreements in how we got there. But anyway, that’s off-topic for this post.
    (See my brief discussion of optical interconnects here—basically, nobody seems to have even a roadmap to making optical interconnects with such low power that they could replace almost all (say, >90%) of the aggregate on-chip interconnect length.)
    Instead our main dispute was about voltages-on-wires, the workhorse of within-chip communication. Pause for:
    On-chip wire interconnects for dummies: As background, here is the oversimplified cartoon version of integrated circuits. There are a bunch of metal wires, and there are transistors that act as switches that connect or disconnect pairs of wires from each other. Depending on which transistors are “on” versus “off...
    22 min
  • EA - Rethink Priorities’ Worldview Investigation Team: Introductions and Next Steps by Bob Fischer
    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: Rethink Priorities’ Worldview Investigation Team: Introductions and Next Steps, published by Bob Fischer on June 21, 2023 on The Effective Altruism Forum.
    Some months ago, Rethink Priorities announced its interdisciplinary Worldview Investigation Team (WIT). Now, we’re pleased to introduce the team’s members:
    Bob Fischer is a Senior Research Manager at Rethink Priorities, an Associate Professor of Philosophy at Texas State University, and the Director of the Society for the Study of Ethics & Animals. Before leading WIT, he ran RP’s Moral Weight Project.
    Laura Duffy is an Executive Research Coordinator for Co-CEO Marcus Davis and works on the Worldview Investigations Project. She is a graduate of the University of Chicago, where she earned a Bachelor of Science in Statistics and co-facilitated UChicago Effective Altruism’s Introductory Fellowship.
    Arvo Muñoz Morán is a Quantitative Researcher working on the Worldview Investigations Team at Rethink Priorities and a research assistant at Oxford's Global Priorities Institute. Before that, he was a Research Analyst at the Forethought Foundation for Global Priorities Research and earned an MPhil in Economics from Oxford. His background is in mathematics and philosophy.
    Hayley Clatterbuck is a Philosophy Researcher at Rethink Priorities and an Associate Professor of Philosophy at the University of Wisconsin-Madison. She has published on topics in probability, evolutionary biology, and animal minds.
    Derek Shiller is a Philosophy Researcher at Rethink Priorities. He has a PhD in philosophy and has written on topics in metaethics, consciousness, and the philosophy of probability. Before joining Rethink Priorities, Derek worked as the lead web developer for The Humane League.
    David Bernard is a Quantitative Researcher at Rethink Priorities. He will soon complete his PhD in economics at the Paris School of Economics, where his research focuses on forecasting and causal inference in the short and long-run. He was a Fulbright Scholar at UC Berkeley and a Global Priorities fellow at the Global Priorities Institute.
    Over the next few months, the team will be working on cause prioritization—a topic that raises hard normative, metanormative, decision-theoretic, and empirical issues. We aren’t going to resolve them anytime soon. So, we need to decide how to navigate a sea of open questions. In part, this involves making our assumptions explicit, producing the best models we can, and then conducting sensitivity analyses to determine both how robust our models are to uncertainty and where the value of information lies.
    Accordingly, WIT’s goal is to make several contributions to the broader conversation about global priorities. Among the planned contributions, you can expect:
    A cross-cause cost-effectiveness model. This tool will allow users to compare interventions like corporate animal welfare campaigns with work on AI safety, the Against Malaria Foundation with attempts to reduce the risk of nuclear war, biosecurity projects with community building, and so on. We’ve been working on a draft of this model in recent months and we recently hired two programmers—Chase Carter and Agustín Covarrubias—to accelerate its public release. While this tool won’t resolve all disputes about resource allocation, we hope it will help the community reason more transparently about these issues.
    Surveys of key stakeholders about the inputs to the model. Many people have thought long and hard about how much x-risk certain interventions can reduce, the relative importance of improving human and animal welfare, and the cost of saving lives in developing countries. We want to capture and distill those insights.
    A series of reports on the cruxes. The model has three key cruxes: animals’ “moral weights,” the expected value of the future, and your preference for ...
    5 min
  • EA - Why Altruists Can't Have Nice Things by lincolnq
    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 Altruists Can't Have Nice Things, published by lincolnq on June 21, 2023 on The Effective Altruism Forum.
    Gesturing at a thing to mostly avoid. My personal opinion.
    This topic has been discussed on the EA Forum before, e.g. Free-spending EA might be a big problem (2022) and The biggest risk of free-spending EA is grift (2022). I also wrote What's Your Hourly Rate? in 2013, and Value of time as an employee in 2019. This piece mostly stands on its own.
    There's a temptation, when solving the world's toughest and most-important problems, to throw money around.
    Lattes on tap! Full-time massage team! Business class flights! Retreat in the Bahamas! When you do the cost/benefit analysis it comes out positive: "An extra four hours of sleep on the plane is worth four thousand dollars, because of how much we're getting paid and how tight the time is."
    The problem, which we always underindex on, is that our culture doesn't stand up to this kind of assault on normalcy. No altruistic, mission-oriented culture can. "I have never witnessed so much money in my life." [1]
    What is culture? I often phrase it as "lessons from the early days of an org." How we survive; how we make it work despite the tough times; our story of how we started with something small and ended up with something great. That knowledge fundamentally pervades everything we do. It needs upkeep and constant reinforcement. "It is always Day One" [2] refers to how Amazon is trying hard, even as they have grown huge, to preserve their culture of scrappiness and caring.
    What perks say
    Fancy, unusual, expensive perks are costly signals. They're saying or implying the following:
    Your time is worth a lot of money
    You are special and important, you deserve this
    We are rich and successful; we are elite
    We are generous and you are lucky to be in our orbit
    You're in the inner ring; you're better than people who aren't part of this
    We desperately want to keep you around
    You are free from menial tasks
    You would never pay for this on your own—but through us, you can have it anyway
    We're just like Google!
    Some of these things might be locally true, but when I zoom out, I get a villainous vibe: this story flatters, it manipulates, it promotes hubris, it tells lies you want to believe. It's a Faustian trade: in exchange for these perks you "just" have to distort your reality, and we're not even asking you to believe hard scary things, just nice ego-boosting things about how special, irreplaceable, on-the-right-track we all are.
    Signals you might want to send instead
    The work cultures I prefer would signal something like the following:
    We're normal people who have chosen to take on especially important work
    We have an angle / insight that most people haven't realized/acted on yet
    We might be wrong, and are constantly seeking evidence that would change our minds
    We should try to be especially virtuous whenever we find ourselves setting a moral example for others
    (We aren't morally better by default, although we may have a bit more information; we are always learning as we go)
    We are focused on the long term good for the world
    Tons of people are suffering / will suffer in the future
    We remind ourselves of this regularly
    (We regularly acknowledge suffering in the world; we definitely don't do things that are cushy)
    We invest a lot in our people and their relationships, and act to preserve that value
    Everyone works hard, sometimes they have to do things they don't want to do
    We are in it for the long haul so don't burn out
    (We might use money in various ways—to make our work relationships stronger or stave off burnout—but we aren't profligate.)
    Instantiations of this culture will vary a decent amount—but I expect that a lot of altruistic orgs have/want to promote at least some subset of these values. Fancy perks push (perniciously, te...
    12 min
  • LW - My tentative best guess on how EAs and Rationalists sometimes turn crazy by habryka
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: My tentative best guess on how EAs and Rationalists sometimes turn crazy, published by habryka on June 21, 2023 on LessWrong.
    Epistemic status: This is a pretty detailed hypothesis that I think overall doesn’t add up to more than 50% of my probability mass on explaining datapoints like FTX, Leverage Research, the LaSota crew etc., but is still my leading guess for what is going on. I might also be really confused about the whole topic.
    Since the FTX explosion, I’ve been thinking a lot about what caused FTX and, relatedly, what caused other similarly crazy- or immoral-seeming groups of people in connection with the EA/Rationality/X-risk communities.
    I think there is a common thread between a lot of the people behaving in crazy or reckless ways, that it can be explained, and that understanding what is going on there might be of enormous importance in modeling the future impact of the extended LW/EA social network.
    The central thesis: "People want to fit in"
    I think the vast majority of the variance in whether people turn crazy (and ironically also whether people end up aggressively “normal”) is dependent on their desire to fit into their social environment. The forces of conformity are enormous and strong, and most people are willing to quite drastically change how they relate to themselves, and what they are willing to do, based on relatively weak social forces, especially in the context of a bunch of social hyperstimulus (lovebombing is one central example of social hyperstimulus, but also twitter-mobs and social-justice cancelling behaviors seem similar to me in that they evoke extraordinarily strong reactions in people).
    My current model of this kind of motivation in people is quite path-dependent and myopic. Just because someone could leave a social context that seems kind of crazy or abusive to them and find a different social context that is better, with often only a few weeks of effort, they rarely do this (they won't necessarily find a great social context, since social relationships do take quite a while to form, but at least when I've observed abusive dynamics, it wouldn't take them very long to find one that is better than the bad situation in which they are currently in). Instead people are very attached, much more than I think rational choice theory would generally predict, to the social context that they end up in, with people very rarely even considering the option of leaving and joining another one.
    This means that I currently think that the vast majority of people (around 90% of the population or so) are totally capable of being pressured into adopting extreme beliefs, being moved to extreme violence, or participating in highly immoral behavior, if you just put them into a social context where the incentives push in the right direction (see also Milgram and the effectiveness of military drafts).
    In this model, the primary reason for why people are not crazy is because social institutions and groups that drive people to extreme action tend to be short lived. The argument here is an argument from selection, not planning. Cults that drive people to extreme action die out quite quickly since they make enemies, or engage in various types of self-destructive behavior. Moderate religions that include some crazy stuff, but mostly cause people to care for themselves and not go crazy, survive through the ages and become the primary social context for a large fraction of the population.
    There is still a question of how you do end up with groups of people who end up taking pretty crazy beliefs extremely seriously. I think there are a lot of different attractors that cause groups to end up with more of the crazy kind of social pressure. Sometimes people who are more straightforwardly crazy, who have really quite atypical brains, end up in positions of power and se...
    13 min
  • LW - Never Fight The Last War by ChristianKl
    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: Never Fight The Last War, published by ChristianKl on June 20, 2023 on LessWrong.
    Lieutenant Colonel J. L. Schley, of the Corps of Engineers wrote:
    “There is a tendency in many armies to spend the peace time studying how to fight the last war”
    After the Chinchilla paper, everyone got focused on the idea that large language models can be more effective.
    The GPT3.5 we saw in ChatGPT was scary and got people to think about how we can prevent bigger models from being trained. There was a lot of capability gain when OpenAI released GPT4 but if we can trust Sam Altman, most of it was not because of an increase in model size but by doing a lot of different improvements.
    From that perspective, many people think about whether or not GPT5 or GPT6 will be scary and dangerous by improving along the same axis as the GPT3.5 ->GPT4 improvement.
    Right now, it seems that most of the danger is not in model improvement but in model utilization.
    Palantir argues that models are commodities and released an AI product without providing any models that go beyond the existing open-source models. The value that Palantir does provide is that it provides interfaces of how a model can interact with the existing data a company has and how to create multi-step chains of data transformation. It provides a system to manage agents that use models to make decisions.
    I read some people argue that Palantir's product is worse than GPT4 and thus should not be seen as an advance in AI, but that's flawed thinking. Autonomous agents are more dangerous than anything you can do in ChatGPT.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min
  • EA - The impact of mobile phones & mobile money for people in poverty by GiveDirectly
    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 impact of mobile phones & mobile money for people in poverty, published by GiveDirectly on June 21, 2023 on The Effective Altruism Forum.
    In our main program, GiveDirectly gives those who don’t already have a mobile phone the option to buy one, subtracting the ~$15 cost from their first transfer, an offer taken by 90% of recipients. Your donations then go directly to the SIM card in their phone, a technology called mobile money.
    The primary benefits of direct cash – improved earnings, health, education, etc. – may obscure a secondary benefit of our work: connecting people to a mobile network. Below we lay out what the evidence shows impact and how we specifically extend impact.
    Access to mobile money and phones has distinct benefits
    A large analysis found mobile money access alone (without cash grants or other aid) has lifted 194K or 2% of Kenyan households out of extreme poverty by increased savings, resilience, and access to better business opportunities. Other research finds mobile money usage can help African families be more resilient to economic shocks, access healthcare more often, and have more social closeness and lending reciprocity than similar households without mobile money.
    Even without access to mobile money, simply owning a phone improves resilience and earnings by reducing travel costs and increasing social connection. Phone ownership also allows for innovative responses to shocks and disasters. During COVID-19 lockdowns, the government of Togo targeted and sent cash aid fully remotely, taking advantage of the fact that some 90% of their citizens already owned a phone. Health officials in low income countries have harnessed phone data to track disease outbreaks.
    GiveDirectly overcomes structural obstacles to bring mobile access
    We issue mobile phones and free SIM cards to households that need one. Our staff walk recipients through how to use this new technology, guiding them to set their unique PIN and check their balance. We also sensitize them to fraud risks and inform them what fees they should expect when they visit an agent to cash out. A government ID is required to register for mobile money, so in some cases we coordinate with local governments to hold ID-issuing campaigns ahead of our enrollment.
    But even with phones, many villages will still struggle to access the benefits of mobile devices due to other structural hurdles. Here are some solutions we implemented to help:
    We bring cell network where it never reached before. 16% of Africans do not live within reach of a mobile network, with the biggest gaps in the poorest regions due to a lack of demand – telcos only build towers where they believe they’ll have customers. However, when GiveDirectly starts a project, we’re creating thousands of new subscribers. Our project reaching 15K families in Kiryandongo, Uganda motivated two telcos (MTN & Airtel) to extend coverage and mobile money agents to the area. In Maryland, Liberia, where we’ve enrolled 11.5K households in our basic income program, we co-financed 10 new cell towers with local telco MTN, bringing cell coverage for the first time to over 2,400 adults across 21 villages (see video below).
    We get cash to remote communities. Over 75% of recipients choose to cash out rather than keeping their transfers in their mobile money account, as many rural merchants only accept cash. In some rural areas, banks and agents do not have enough cash on hand for hundreds of families to withdraw large amounts in a matter of days. GiveDirectly collaborates with national banks and mobile money operators to ensure sufficient liquidity in the regions where payments are about to go out. In some of the most remote areas, we arrange for mobile money agents to travel to village centers to cash out recipients.
    The final challenge of charging phones without an ...
    6 min