The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • AF - Real-Time Research Recording: Can a Transformer Re-Derive Positional Info? by Neel Nanda
    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: Real-Time Research Recording: Can a Transformer Re-Derive Positional Info?, published by Neel Nanda on November 1, 2022 on The AI Alignment Forum.
    New experiment: Recording myself real-time as I do mechanistic interpretability research! I try to answer the question of what happens if you train a toy transformer without positional embeddings on the task of "predict the previous token" - turns out that a two layer model can rederive them! You can watch me do it here, and you can follow along with my code here. This uses a transformer mechanistic interpretability library I'm writing called EasyTransformer, and this was a good excuse to test it out and create a demo!
    This is an experiment in recording and publishing myself doing "warts and all" research - figuring out how to train the model and operationalising an experiment (including 15 mins debugging loss spikes...), real-time coding and tensor fuckery, and using my go-to toolkit. My hope is to give a flavour of what actual research can look like - how long do things actually take, how often do things go wrong, what is my thought process and what am I keeping in my head as I go, what being confused looks like, and how I try to make progress. I'd love to hear whether you found this useful, and whether I should bother making a second half!
    Though I don't want to overstate this - this was still a small, self-contained toy question that I chose for being a good example task to record (and I wouldn't have published it if it was TOO much of a mess).
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min
  • EA - Dark mode (Forum update November 2022) by Lizka
    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: Dark mode (Forum update November 2022), published by Lizka on November 1, 2022 on The Effective Altruism Forum.
    TL;DR: We now have dark mode on the EA Forum. There are some bugs, but you can turn it on!
    This is what the Frontpage looks like in dark mode:
    How to use it
    You need to be logged in — the Forum will respect browser settings for logged-out users.
    To change light/dark mode settings, hover over your username in the upper right-hand corner (or just go to your profile page). Then select “Account Settings.” In account settings, click on “Site Customizations” and click on the dropdown under “Theme.”
    You’ll see three options:
    “Auto” will respect your browser settings, even if you have a setting that varies your mode by time of day or the like (e.g. if you set your browser to be light during the day and dark in the evening and at night)
    “Light” will be light — the Forum you might be used to.
    “Dark” will always be in dark mode.
    Feedback is (always) welcome!
    Please also feel free to report bugs in the comments of this post.
    Thanks to everyone who worked on this. :)
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min
  • EA - Rethink Priorities’ Special Projects Team is hiring by Rachel Norman
    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’ Special Projects Team is hiring, published by Rachel Norman on November 1, 2022 on The Effective Altruism Forum.
    TL;DR
    The Special Projects Program launched in mid 2022 to provide fiscal sponsorship and operational support to (for the most part, longtermist) megaprojects.
    We are now hiring for Manager and Associate roles on the Special Projects team.
    Rethink Priorities is a remote organization that welcomes applicants from almost anywhere in the world.
    Compensation for the roles ranges from $74,000 USD (for Associate roles) up to $111,000 (for Manager roles).
    Examples of activities for Managers and Associates include day-to-day operations, running hiring rounds, developing budgets, writing business plans for entrepreneurial projects, fundraising, liaising with stakeholders, researching legal/operational issues, scouting for talent for the projects, helping run prizes or requests for proposals, and grantmaking.
    In lieu of holding a webinar for these roles, we will aim to answer all questions posted in the comments section of this post. Please ask away!
    Please apply here by November 13th at the end of the day in your local time.
    General information about all our roles
    We invite anyone who is interested to apply, regardless of background, experience, or credentials. We will select candidates almost entirely based on their performance in our selection process, putting only minimal weight on CVs and references.
    Most of our positions are equally open to part- or full-time candidates (anywhere between 20-40 hours per week).
    The full-time annual salary we offer varies based on title and prior experience. The range for the roles in our Special Projects Team is from $74,000 (for Associate roles) to $111,000 (for Manager roles). Salaries for part-time staff are prorated for the fraction of 40 hours/week they work.
    See Titles at Rethink Priorities for more information.
    See the "What we offer" section below for details.
    We are a remote organization and welcome applicants from anywhere in the world.
    We expect to be legally able to hire in most countries. If you are unsure whether we can hire in the country in which you are based, please email [email protected].
    Most of our staff are in time zones between UTC-8 and UTC+3. While we welcome candidates from outside of these time zones, we often have meetings during working hours in these time zones that employees are expected to attend.
    For permanent staff, Rethink Priorities offers comprehensive health coverage as regionally appropriate.
    Other benefits include stipends for workplace equipment, organization-wide mid- and end-of-year breaks, and generous leave policies.
    About Special Projects and the roles
    The Special Projects Program launched in mid 2022 to support activities outside of RP’s core research agenda. A lot of these projects require complicated operational support. RP’s Core Operations team is focused on ensuring RP's day-to-day functioning, and often these special projects require additional capacity beyond what the Core Ops team has or exploration of novel areas that are outside the purview of the Core Ops team.
    Although we currently expect Special Projects staff to primarily focus on operations activities, this is unlikely to solely be day-to-day operations tasks, and there are also many other activities we would be excited for the team to engage in. Some possible examples include developing budgets, writing business plans for entrepreneurial projects, fundraising, liaising with stakeholders, researching legal/operational issues, scouting for talent for the projects, helping run prizes or requests for proposals, and grantmaking.
    We expect that most of these projects will be in the longtermism space, although that is not a requirement. SP plans on fiscally sponsoring, incubating, or otherwise...
    12 min
  • EA - ML Safety Scholars Summer 2022 Retrospective by ThomasW
    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: ML Safety Scholars Summer 2022 Retrospective, published by ThomasW on November 1, 2022 on The Effective Altruism Forum.
    TLDR
    This is a report on the Machine Learning Safety Scholars program, organized over the summer by the Center for AI Safety. 63 students graduated from MLSS: the list of graduates and final projects is here. The program was intense, time-consuming, and at times very difficult, so graduation from the program is a significant accomplishment.
    Overall, I think the program went quite well and that many students have noticeably accelerated their AI safety careers. There are certainly many areas of improvement that could be made for a future iteration, and many are detailed here. We plan to conduct followup surveys to determine the longer-run effects of the program.
    This post contains three main sections:
    This TLDR, which is meant for people who just want to know what this document is and see our graduates list.
    The executive summary, which includes a high-level overview of MLSS. This might be of interest to students considering doing MLSS in the future or anyone else interested in MLSS.
    The full report, which was mainly written for future MLSS organizers, but I’m publishing here because it might be useful to others running similar programs.
    The report was written by Thomas Woodside, the project manager for MLSS. “I” refers to Thomas, and does not necessarily represent the opinion of the Center for AI Safety, its CEO Dan Hendrycks, or any of our funders.
    Visual Executive Summary
    MLSS Overview
    MLSS was a summer program for mostly undergraduate students aimed to teach the foundations of machine learning, deep learning, and ML safety. The program ended up being ten weeks long and included an optional final project. It incorporated office hours, discussion sections, speaker events, conceptual readings, paper readings, written assignments, and programming assignments. You can see our full curriculum here.
    Survey Results
    All MLSS graduates filled out an anonymous survey. What follows is a mostly visual depiction of the program through the lens of these survey results.
    Overall Experience in MLSS
    Of course, this sample is biased towards graduates of MLSS, since we required them to complete the survey to receive their final stipends (and non graduates didn’t get final stipends). However, it seems clear from the way that people responded to this survey that the majority had quite positive opinions of our program.
    We also asked students about their future plans:
    The results suggest that many students are actively trying to work in AI safety and that their participation in MLSS helped them become more confident in that choice. MLSS did decrease some students’ desire to research AI safety. We do not think this is necessarily a bad thing, as many students were using MLSS to test fit; presumably, some are not great fits for AI safety research but might be able to contribute in some other way.
    We asked students about why they chose to do the program:
    We conclude that the stipend was extremely useful to students, and allowed many students to complete the program who wouldn’t have otherwise been able to. Nearly all graduates said they were interested in learning about ML safety in particular.
    We asked students about the quality of the support from their TAs:
    We asked students a few questions about what they thought about AI x-risk:
    Lastly, we asked how many hours students spent in the course:
    Concluding Thoughts
    To a large extent, we think these results speak for themselves. Students said they got a lot out of MLSS, and I believe many of them are very interested in pursuing AI safety careers that have been accelerated by the program. The real test of our program, of course, will come when we survey students in the future to see what they are doing and how the course helped...
    36 min
  • AF - Threat Model Literature Review by Zachary Kenton
    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: Threat Model Literature Review, published by Zachary Kenton on November 1, 2022 on The AI Alignment Forum.
    TL;DR: This post provides a literature review of some threat models of how misaligned AI can lead to existential catastrophe. See our accompanying post for high-level discussion, a categorization and our consensus threat model.
    Where available we cribbed from the summary in the Alignment Newsletter.
    For other people's overviews of some threat models, see here and here.
    [ETA: When considering strengths and weaknesses of each threat model, this was done with respect to our goal of better generation/prioritization among alignment research projects. They shouldn't necessarily be read as an all-things-considered review of that work.]
    Categorization
    Threat Models
    Goal misgeneralization + Specification Gaming ➜ MAPS
    Is Power-Seeking AI an Existential Risk? (Carlsmith)
    Summary
    This report investigates the classic AI risk argument in detail, and decomposes it into a set of conjunctive claims. Here’s the quick version of the argument: We will likely build highly capable and agentic AI systems that are aware of their place in the world, and which will be pursuing problematic objectives. Thus, they will take actions that increase their power, which will eventually disempower humans, leading to an existential catastrophe. We will try and avert this, but will probably fail to do so since it is technically challenging and we are not capable of the necessary coordination.
    There’s a lot of vague words in the argument above, so let’s introduce some terminology to make it clearer:
    Advanced capabilities: We say that a system has advanced capabilities if it outperforms the best humans on some set of important tasks (such as scientific research, business/military/political strategy, engineering, and persuasion/manipulation).
    Agentic planning: We say that a system engages in agentic planning if it (a) makes and executes plans, (b) in pursuit of objectives, (c) on the basis of models of the world. This is a very broad definition and doesn’t have many of the connotations you might be used to for an agent. It does not need to be a literal planning algorithm -- for example, human cognition would count, despite (probably) not being just a planning algorithm.
    Strategically aware: We say that a system is strategically aware if it models the effects of gaining and maintaining power over humans and the real-world environment.
    PS-misaligned (power-seeking misaligned): On some inputs, the AI system seeks power in unintended ways due to problems with its objectives (if the system actually receives such inputs, then it is practically PS-misaligned).
    The core argument is then that AI systems with advanced capabilities, agentic planning, and strategic awareness (APS-systems) will be practically PS-misaligned, to an extent that causes an existential catastrophe.
    The key hypothesis underlying this argument is:
    Instrumental Convergence Hypothesis: If an APS AI system is less-than-fully aligned, and some of its misaligned behavior involves strategically-aware agentic planning in pursuit of problematic objectives, then in general and by default, we should expect it to be less-than-fully PS-aligned, too.
    The reason to believe the hypothesis is that power is useful for achieving objectives, because it increases the options available to the system. If the system shows unintended behavior in pursuit of a problematic objective then having more options available will tend to improve its ability to achieve that objective, hence we expect it to be PS-misaligned.
    Of course, we will try to prevent this -- why should we expect that we can’t fix the problem? The author considers possible remedies, and argues that they all seem quite hard:
    We could give AI systems the right objectives (alignment), but this seems quit...
    42 min
  • AF - Clarifying AI X-risk by Zachary Kenton
    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: Clarifying AI X-risk, published by Zachary Kenton on November 1, 2022 on The AI Alignment Forum.
    TL;DR: We give a threat model literature review, propose a categorization and describe a consensus threat model from some of DeepMind's AGI safety team. See our post for the detailed literature review.
    The DeepMind AGI Safety team has been working to understand the space of threat models for existential risk (X-risk) from misaligned AI. This post summarizes our findings. Our aim was to clarify the case for X-risk to enable better research project generation and prioritization.
    First, we conducted a literature review of existing threat models, discussed their strengths/weaknesses and then formed a categorization based on the technical cause of X-risk and the path that leads to X-risk. Next we tried to find consensus within our group on a threat model that we all find plausible.
    Our overall take is that there may be more agreement between alignment researchers than their disagreements might suggest, with many of the threat models, including our own consensus one, making similar arguments for the source of risk. Disagreements remain over the difficulty of the alignment problem, and what counts as a solution.
    Categorization
    Here we present our categorization of threat models from our literature review, based on the technical cause and the path leading to X-risk. It is summarized in the diagram below.
    In green on the left we have the technical cause of the risk, either specification gaming (SG) or goal misgeneralization (GMG). In red on the right we have the path that leads to X-risk, either through the interaction of multiple systems, or through a misaligned power-seeking (MAPS) system. The threat models appear as arrows from technical cause towards path to X-risk.
    The technical causes (SG and GMG) are not mutually exclusive, both can occur within the same threat model. The distinction between them is motivated by the common distinction in machine learning between failures on the training distribution, and when out of distribution.
    To classify as specification gaming, there needs to be bad feedback provided on the actual training data. There are many ways to operationalize good/bad feedback. The choice we make here is that the training data feedback is good if it rewards exactly those outputs that would be chosen by a competent, well-motivated AI. We note that the main downside to this operationalisation is that even if just one out of a huge number of training data points gets bad feedback, then we would classify the failure as specification gaming, even though that one datapoint likely made no difference.
    To classify as goal misgeneralization, the behavior when out-of-distribution (i.e. not using input from the training data), generalizes poorly about its goal, while its capabilities generalize well, leading to undesired behavior. This means the AI system doesn’t just break entirely, it still competently pursues some goal, but it’s not the goal we intended.
    The path leading to X-risk is classified as follows. When the path to X-risk is from the interaction of multiple systems, the defining feature here is not just that there are multiple AI systems (we think this will be the case in all realistic threat models), it’s more that the risk is caused by complicated interactions between systems that we heavily depend on and can’t easily stop or transition away from. (Note that we haven't analyzed the multiple-systems case very much, and there are also other technical causes for those kinds of scenarios.)
    When the path to X-risk is through Misaligned Power-Seeking (MAPS), the AI system seeks power in unintended ways due to problems with its goals. Here, power-seeking means the AI system seeks power as an instrumental subgoal, because having more power increases the options availab...
    8 min
  • EA - Insect farming might cause more suffering than other animal farming by ishankhire
    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: Insect farming might cause more suffering than other animal farming, published by ishankhire on November 1, 2022 on The Effective Altruism Forum.
    The edible insect market has been growing and is projected to continue increasing rapidly in the future. There has been significant research pointing to the fact that insects show signs of consciousness, and I think it’s important to seriously compare insect welfare interventions with other ones.
    Brian Tomasik's “How Much Direct Suffering Is Caused by Various Animal Foods?” compares the days of suffering per kilogram of different animals. I’d like to add on to this post by a guesstimate model of where insect suffering may lie in this list. I used cricket suffering to create this model and it’s important to note there are differences in the numbers depending on which insects are being farmed—mealworms and black soldier flies likely have a different weight and sentience multiplier.
    My reasoning for the numbers I used is given in the sheet: if you hover over the fields, you will see a detailed description.
    The model shows the mean value is 1300 days of suffering per kg of insects consumed, which is twice as high as the highest animal in Brian Tomasik’s list (farmed catfish). The median value is considerably lower (35), and the 5th and 95th percentile are approximately 0.32 and 2500.
    If correct, these estimates suggest that insect welfare deserves a high priority compared to other farmed animal welfare.
    Here are possible interventions. In addition to legislation to restrict insect farming (such as bans on using insect feed for livestock), I think there could be a lot of value in using insects other than crickets which have fewer neurons and therefore could have a smaller sentience multiplier (like mealworms). Interventions to ban likely more painful methods of killing insects such as boiling could also be impactful.
    I’d very much appreciate discussion and feedback on the model and where you think it could be improved.
    I made this model during a one-day hackathon at the Atlas Fellowship.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • AF - Auditing games for high-level interpretability by Paul Colognese
    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: Auditing games for high-level interpretability, published by Paul Colognese on November 1, 2022 on The AI Alignment Forum.
    This post was written under Evan Hubinger’s direct guidance and mentorship, as a part of the Stanford Existential Risks Institute ML Alignment Theory Scholars (MATS) program.
    Thanks to Evan Hubinger, Arun Jose, and Dane Sherburn for discussions/feedback.
    Introduction
    This post aims to extend the auditing games framework to high-level interpretability; a framework for adversarially testing future interpretability methods developed for understanding and working with a model’s internal properties at a high-level (e.g. understanding whether or not a model is performing optimization, etc.)
    We begin by discussing what we mean by high-level interpretability and why we should care about it. We then introduce an auditing games framework in this context along with some simple examples. We then argue that this framework provides good benchmarks for evaluating our high-level interpretability methods. Finally, we discuss a fundamental difficulty that we encounter when trying to create good tests within the auditing games framework described in this post.
    What is high-level interpretability and why should we care?
    Understanding what our models are doing internally seems necessary for us to have confidence that they will perform well during deployment.
    One key problem in prosaic alignment is that of distributional shifts: the set of models that perform well during and at the end of training may include models that perform poorly, potentially catastrophically, during deployment. Deceptive models fit this description, i.e., a model that performs well during the training process as an instrumental goal so that it can be deployed and pursue its unaligned terminal goal without fear of modification from the training process. Unfortunately, observing the external behavior of models during training can’t help us differentiate between an aligned model and a competent deceptive model.
    One way we can gain insight into how a model will behave in deployment is by understanding how the model operates internally/what algorithm(s) it’s running. For example, if we could use interpretability tools to look inside a model and understand whether it’s an optimizer, and if so what its terminal goals are, then we would better understand how it’s going to behave in deployment, in particular, whether it’s going to pursue aligned goals or not.
    Many alignment proposals require the capacity to interpret and work with a model’s internals at a high level, at least to some degree, for us to have confidence that the proposals will work. Some examples of the types of insight/leverage we might want to gain from using interpretability tools include:
    Knowing whether a model is deceptive.
    Knowing whether a model is doing optimization/search.
    If the model is doing optimization, what is it optimizing for?
    If we have a model that’s supposed to make predictions, how is it making its predictions?
    If the model is an optimizer with objectives, can we modify its objectives?
    If the model has a world model, can we modify the knowledge stored in its world model?
    Developing good benchmarks for interpretability methods that aim to tackle the above questions is important to ensure we’re progressing in the right direction. In the next section, we’ll explore a candidate for such a benchmark.
    What are auditing games?
    Auditing games are a framework for adversarially testing our interpretability methods. In this section, we begin by explaining what auditing games are in the context of high-level interpretability.
    Suppose an interpretability team (the Blue Team) has invented an interpretability method that they claim can do X, where X could be something like “detecting whether a model based on a transformer architectu...
    13 min
  • EA - A dozen doubts about GiveWell’s numbers by JoelMcGuire
    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 dozen doubts about GiveWell’s numbers, published by JoelMcGuire on November 1, 2022 on The Effective Altruism Forum.
    An entry to GiveWell’s Change Our Mind Contest
    Joel McGuire, Michael Plant, and Samuel Dupret
    [Note: We co-authored a post on deworming that has been considered for the contest.]
    Summary
    We raise twelve critiques of GiveWell’s cost-effectiveness analyses. Ten apply to specific inputs for malaria prevention, cash transfers, deworming, and two are relevant for more than one intervention. Our calculations are available in this modified spreadsheet of GiveWell’s cost-effectiveness analyses. We provide a summary of these first, then set them out in greater detail. We encountered these as part of our work to replicate GiveWell’s analysis. They do not constitute an exhaustive review
    Malaria prevention
    1. GiveWell assumes that a large fraction (34%) of the total effects of anti-malaria bed nets come from the extra income children will later earn due to reduced malaria exposure. However, GiveWell provides little explanation or justification for these numbers, which are strikingly, indeed suspiciously, large: taken at face value, they imply that bednets are 4x more cost-effective at increasing income than cash transfers.
    We were unsure what the income-increasing effect of bednet should be, so we investigated a number of factors, which we enumerate as the next six critiques.
    2. The evidence GiveWell uses to generate its income-improving figure for bednets is from contexts that seem very different from that of the charities it recommends. While GiveWell does discount this evidence it uses for ‘generalisability’, we suggest a larger discount would be consistent with GiveWell’s reasoning regarding deworming and would reduce the total cost-effectiveness by at least 10%.
    3. We present possible mechanisms for malaria prevention to income benefits, but they do not explain most of the effects. It’s unclear how this would change the cost-effectiveness results.
    4. We demonstrate how GiveWell’s analysis does not sufficiently adjust for differences in context by looking at malaria prevalence. Malaria prevalence was higher in the context of the evidence than in AMF’s context. When we adjust for this, we expect this will lead to around a 17% decrease in AMF’s total cost-effectiveness, with low confidence.
    5. GiveWell has missed some evidence for the effect of malaria prevention on income. We expect this decreases the total cost-effectiveness of AMF by 14%, with moderate confidence.
    6. GiveWell may underestimate the differences in benefits between women and men. We expect with moderate confidence that incorporating more evidence would decrease the total cost-effectiveness of AMF by 16%.
    7. GiveWell assumes, without clear justification, that the income effects of malaria prevention will endure in just the same way as they suppose the effects of deworming do. We’ve previously called deworming’s benefits into question (McGuire et al., 2022), incorporating the adjustment proposed there would decrease the total cost-effectiveness by AMF by 3.4% to 10.2% (low confidence)
    We’ve presented these changes in cost-effectiveness for changing only one variable at a time. If we implement all of them at once (excluding critiques 3 and 7 for which we are more uncertain how they’ll change the results), then the effects decline by 29% (see row 202 of “AMF total” tab).
    Cash transfers
    8. GiveWell’s analysis of the effectiveness of cash transfers only uses one study to estimate GiveDirectly’s immediate increases in consumption (Haushofer & Shapiro, 2013) when there are many more available. Using more data we show that recipients of GiveDirectly cash transfers are actually 70% richer than assumed in GiveWell’s cost-effectiveness analysis. Hence, the relative increase in consumption from GiveDirectl...
    43 min
  • LW - Why Aren't There More Schelling Holidays? by johnswentworth
    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 Aren't There More Schelling Holidays?, published by johnswentworth on October 31, 2022 on LessWrong.
    A common pattern when working in teams: when one person is out on vacation, there’s a disproportionate drop in productivity for the team as a whole. Lots of things end up blocked on the person who’s out. For instance, in a small software team, maybe the developer who owns a particular API is out, and nobody else knows that API well enough to confidently make changes, so anything involving changes in that API ends up blocked until the owner is back. Or, even if someone else steps in to handle changes to the API, they’re much more likely to introduce bugs.
    To some extent, we can structure teams to mitigate that kind of problem. “Everyone does everything” is very costly, but underrated. Weaker versions like “At least three people can cover any given thing” are also costly, but underutilized.
    But there’s a much less costly strategy which gains a decent chunk of the same benefits: coordinate vacation. If everyone goes on vacation 3 weeks per year on the same 3 weeks, then that’s only 3 weeks of nonproductivity for the team; the other 49 weeks are full steam. If everyone on a 10-person team goes on vacation 3 weeks per year at different times, then that’s 30 weeks per year of disproportionate productivity-loss; the team is at full steam less than half the time.
    Thus the case for national holidays: it’s not like everyone is required to get the day off (at least in the US), but it’s a Schelling point for lots of people to take a vacation at the same time. People get their much-needed recovery time simultaneously, so teams can spend more time working at full capacity.
    But in practice, most people take a lot more vacation days than there are national holidays - which suggests that the current number of holidays is too small.
    It seems like companies (or nonprofit orgs) could profit by filling that gap at the company level: declare official “company holiday” weeks, and offer incentives for employees to take their vacations during those weeks. Unlike national holidays, the company holidays won’t necessarily line up neatly with good vacation times for spouses/children/parents/friends. But on the upside, since a company holiday need not be at the same time as the rest of the country, traveling should be easier.
    Expanding from there, one could imagine a few companies/orgs which need to interface with each other a lot coordinating their Schelling holidays. For instance, when I worked at a trading company, the usual rule was “if the markets aren’t open today, we’re not working today”.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min