The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - No, You Need to Write Clearer by NicholasKross
    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: No, You Need to Write Clearer, published by NicholasKross on April 29, 2023 on LessWrong.
    This post is aimed solely at people in AI alignment/safety.
    So I was reading this post, which basically asks "How do we get Eliezer Yudkowsky to realize this obviously bad thing he's doing, and either stop doing it or go away?"
    That post was linking this tweet, which basically says "Eliezer Yudkowsky is doing something obviously bad."
    Now, I had a few guesses as to the object-level thing that Yudkowsky was doing wrong. The person who made the first post said this:
    he's burning respectability that those who are actually making progress on his worries need. he has catastrophically broken models of social communication and is saying sentences that don't mean the same thing when parsed even a little bit inaccurately. he is blaming others for misinterpreting him when he said something confusing. etc.
    A-ha! A concrete explanation!
    Buried in the comments. As a reply to someone innocently asking what EY did wrong.
    Not in the post proper. Not in the linked tweet.
    The Problem
    Something about this situation got under my skin, and not just for the run-of-the-mill "social conflict is icky" reasons.
    Specifically, I felt that if I didn't write this post, and directly get it in front of every single person involved in the discussion... then not only would things stall, but the discussion might never get better at all.
    Let me explain.
    Everyone, everyone, literally everyone in AI alignment is severely wrong about at least one core thing, and disagreements still persist on seemingly-obviously-foolish things.
    This is because the field is "pre-paradigmatic". That is, we don't have many common assumptions that can all agree on, no "frame" that we all think is useful.
    In biology, they have a paradigm involving genetics and evolution and cells. If somebody shows up saying that God created animals fully-formed... they can just kick that person out of their meetings! And they can tell them "go read a biology textbook".
    If a newcomer disagrees with the baseline assumptions, they need to either learn them, challenge them (using other baseline assumptions!), or go away.
    We don't have that luxury.
    AI safety/alignment is pre-paradigmatic. Every word in this sentence is a hyperlink to an AI safety approach. Many of them overlap. Lots of them are mutually-exclusive. Some of these authors are downright surprised and saddened that people actually fall for the bullshit in the other paths.
    Many of these people have even read the same Sequences.
    Inferential gaps are hard to cross. In this environment, the normal discussion norms are necessary but not sufficient.
    What You, Personally, Need to Do Differently
    Write super clearly and super specifically.
    Be ready and willing to talk and listen, on levels so basic that without context they would seem condescending. "I know the basics, stop talking down to me" is a bad excuse when the basics are still not known.
    Draw diagrams. Draw cartoons. Draw flowcharts with boxes and arrows. The cheesier, the more "obvious", the better.
    "If you think that's an unrealistic depiction of a misunderstanding that would never happen in reality, keep reading."
    Eliezer Yudkowsky, about something else.
    If you're smart, you probably skip some steps when solving problems. That's fine, but don't skip writing them down! A skipped step will confuse somebody. Maybe that "somebody" needed to hear your idea.
    Read The Sense of Style by Steven Pinker. You can skip chapter 6 and the appendices, but read the rest. Know the rules of "good writing". Then make different tradeoffs, sacrificing beauty for clarity. Even when the result is "overwritten" or "repetitive".
    Make it obvious which (groupings of words) within (the sentences that you write) belong together. This helps people "parse" your
    Explain the sam...
    8 min
  • LW - LLMs and computation complexity by Jonathan Marcus
    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: LLMs and computation complexity, published by Jonathan Marcus on April 28, 2023 on LessWrong.
    Epistemic status: Speculative. I've built many large AI systems in my previous HFT career but have never worked with generative AIs. I am leveling up in LLMs by working things out from base principles and observations. All feedback is very welcome.
    Tl;dr: An LLM cannot solve computationally hard problems. Its ability to write code is probably its skill of greatest potential. I think this reduces p(near term doom).
    An LLM takes the same amount of computation for each generated token, regardless of how hard it is to predict. This limits the complexity of any problem an LLM is trying to solve.
    Consider two statements:
    "The richest country in North America is the United States of ______"
    "The SHA1 of 'abc123', iterated 500 times, is _______"
    An LLM's goal is to predict the best token to fill in the blank given its training and the previous context. Completing statement 1 requires knowledge about the world but is computationally trivial. Statement 2 requires a lot of computation. Regardless, the LLM performs the same amount of work for either statement.
    It cannot correctly solve computationally hard statements like #2. Period. If it could, that would imply that all problems can be solved in constant time, which is provably (and obviously) false.
    Why does this matter? It puts some bounds on what an LLM can do.
    Zvi writes:
    Eliezer Yudkowsky does not see any of this as remotely plausible. He points out that in order to predict all the next word in all the text on the internet and all similar text, you need to be able to model the processes that are generating that text. And that predicting what you would say is actually a good bit harder than it is to be a being that says things - predicting that someone else would say is tricker and requires more understanding and intelligence than the someone else required to say it, the problem is more constrained.
    And then he points out that the internet contains text whose prediction outright requires superhuman capabilities, like figuring out hashes, or predicting the results of scientific experiments, or generating the result of many iterations of refinement. A perfect predictor of the internet would be a superintelligence, it won’t ‘max out’ anywhere near human.
    I interpret this the opposite way. Being a perfect predictor of the internet would indeed require a superintelligence, but it cannot be done by an LLM.
    How does an LLM compute?
    What kinds of problems fall into category 2 (i.e., clearly unanswerable by an LLM)? Let's dig in to how an LLM computes.
    For each token, it reviews all the tokens in its context window "at least once", call it O(1) time. To produce n tokens, it does O(n^2) work. Without being too precise about the details, this roughly means it can't solve problems that are more complex than O(n^2).
    Consider some examples (all tested with GPT-4):
    Addition, O(1)
    It's not always accurate, but it's usually able to do addition correctly.
    Sorting, O(n log n)
    I asked it to sort 100 random integers that I'd generated, and it got it right.
    My guess is that it doesn't have the internal machinery to do a quick sort, and was probably doing something more like O(n^2), but either way that's within its powers to get right, and it got it.
    Matrix multiplication, O(n^3)
    I generated a 3x3 matrix called A and told it to compute AA. This was interesting, let's look at what it did:
    Pretty cool! It executed the naive matrix multiplication algorithm by using O(n^3) tokens to do it step-by-step. If I ask it to do it without showing its work, it hallucinates an incorrect answer:
    The result was the right shape, and the elements had approximately the right number of digits. Slight problem: the elements are all incorrect. Whoops. This makes sense though....
    10 min
  • LW - The Social Alignment Problem by irving
    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 Social Alignment Problem, published by irving on April 28, 2023 on LessWrong.
    TLDR: I think public outreach is a very hopeful path to victory. More importantly, extended large-scale conversation around questions such as whether public outreach is a hopeful path to victory would be very likely to decrease p(doom).
    You’re a genius mechanical engineer in prison. You stumble across a huge bomb rigged to blow in a random supply closet. You shout for two guards passing by, but they laugh you off. You decide to try to defuse it yourself.
    This is arguably a reasonable response, given that this is your exact skill set. This is what you were trained for. But after a few hours of fiddling around with the bomb, you start to realize that it's much more complicated than you thought. You have no idea when it’s going to go off, but you start despairing that you can defuse it on your own. You sink to the floor with your face in your hands. You can’t figure it out. Nobody will listen to you.
    Real Talking To The Public has never been tried
    Much like the general public has done with the subject of longevity, I think many people in our circle have adopted an assumption of hopelessness toward public outreach social alignment, before a relevant amount of effort has been expended. In truth, there are many reasons to expect this strategy to be quite realistic, and very positively impactful too. A world in which the cause of AI safety is as trendy as the cause of climate change, and in which society is as knowledgeable about questions of alignment as it it about vaccine efficacy (meaning not even that knowledgeable), is one where sane legislation designed to slow capabilities and invest in alignment becomes probable, and where capabilities research is stigmatized and labs find access to talent and resources harder to come by.
    I’ve finally started to see individual actors taking steps towards this goal, but I’ve seen a shockingly small amount of coordinated discussion about it. When the topic is raised, there are four common objections: They Won’t Listen, Don’t Cry Wolf, Don’t Annoy the Labs, and Don’t Create More Disaster Monkeys.
    They won’t listen/They won’t understand
    I cannot overstate how clearly utterly false this is at this point.
    It’s understandable that this has been our default belief. I think debating e/accs on Twitter has broken our brains. The experience of explaining again and again why something smarter than you that doesn’t care about you is dangerous, and being met with these arguments, is a soul-crushing experience. It made sense to expect that if it’s this hard to explain to a fellow computer enthusiast, then there’s no hope of reaching the average person. For a long time I avoided talking about it with my non-tech friends (let's call them "civilians") for that reason. However, when I finally did, it felt like the breath of life. My hopelessness broke, because they instantly vigorously agreed, even finishing some of my arguments for me. Every single AI safety enthusiast I’ve spoken with who has engaged with civilians has had the exact same experience.
    I think it would be very healthy for anyone who is still pessimistic about convincing people to just try talking to one non-tech person in their life about this. It’s an instant shot of hope.
    The truth is, if we were to decide that getting the public on our side is our goal, I think we would have one of the easiest jobs any activists social alignment researchers have ever had.
    Far from being closed to the idea, civilians in general literally already get it. It turns out, Terminator and the Matrix have been in their minds this whole time. We assumed they'd been inoculated against serious AI risk concern - turns out, they walked out of the theaters thinking “wow, that’ll probably happen someday”. They’ve been thinking that th...
    13 min
  • EA - Better weather forecasting: Agricultural and non-agricultural benefits in low- and lower-middle-income countries by Rethink Priorities
    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: Better weather forecasting: Agricultural and non-agricultural benefits in low- and lower-middle-income countries, published by Rethink Priorities on April 28, 2023 on The Effective Altruism Forum.
    Editorial note
    This report is a “shallow” investigation, as described here, and was commissioned by Open Philanthropy and produced by Rethink Priorities. Open Philanthropy does not necessarily endorse our conclusions.
    The primary focus of the report was to investigate whether improving weather forecasting could have benefits for agriculture in low- and lower-middle income countries, and evaluate how cost-effective this might be. Note that this means we did not evaluate improvements in weather forecasting against other potential interventions to achieve the same aims, such as the development of climate-resilient crops.
    We reviewed the academic and gray literature, and also spoke to seven experts. In our report, we provide a brief description of weather forecasting and the global industry, before evaluating which farmers might most benefit from improved forecasts. We then explore how predictions are currently made in countries of interest, and how accurate they are. We evaluate the cost-effectiveness of one intervention that was often mentioned by experts, and highlight other potential opportunities for grantmaking and further research.
    We don’t intend this report to be Rethink Priorities’ final word on this topic and we have tried to flag major sources of uncertainty in the report. We are open to revising our views as more information is uncovered.
    Key takeaways
    Weather forecasting consists of three stages.
    Data assimilation: to understand the current state of the atmosphere, based on observations from satellites and surface-based stations. All forecasts beyond 4-5 days require global observations.
    Forecasting: to model how the atmosphere will change over time. Limits to supercomputing power necessitates tradeoffs, e.g., between forecast length and resolution.
    Communication: packaging relevant information and sharing this with potential users.
    The global annual spending on weather forecasting is over $50 billion.
    Around 260-305 million smallholder farms in South Asia, sub-Saharan Africa and Southeast Asia stand to benefit the most.
    A wide range of farming decisions benefit from weather forecasts, from strategic seasonal or annual decisions like crop choice, to day-to-day decisions like irrigation timing.
    There is some evidence that farmers can benefit from forecasts in terms of increased yields and income.
    For smallholder farmers, cereals are likely the most important crop group, constituting 90% of their agricultural output.
    Medium-range and seasonal forecasts of rainfall and temperature are most important to these farmers.
    In the lower-middle-income countries and low-income countries1 of interest, weather forecasting quality remains poor.
    Global numerical weather prediction (NWP) is a methodology that underlies much of weather forecasting. Seasonal forecasts of temperature seem more accurate than those for precipitation. At shorter timescales, forecasts in the tropics may be useful with a lead time of up to two weeks, and are generally less accurate than forecasts for the mid-latitudes.
    Public sector forecasting in these LMICs is generally informed by global NWPs, meaning that accuracy and resolution remain low.
    LMICs do not improve on global NWPs, as they lack resources and access to raw data.
    We have not found any evidence to suggest that private sector forecasts are better, though Ignitia’s approach targets one of the main issues with global NWPs.
    A small sample of public and private organizations we reviewed spends about $300 million each year on improving forecasting.
    It’s likely that advisories are needed, especially for seasonal forecasts.
    Improving weather forecast...
    6 min
  • LW - [Linkpost] Sam Altman's 2015 Blog Posts Machine Intelligence Parts 1 & 2 by Olivia Jimenez
    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: [Linkpost] Sam Altman's 2015 Blog Posts Machine Intelligence Parts 1 & 2, published by Olivia Jimenez on April 28, 2023 on LessWrong.
    I'm often surprised more people haven't read Open AI CEO Sam Altman's 2015 blog posts Machine Intelligence Part 1 & Part 2. In my opinion, they contain some of the most strong, direct, and clear articulations of why AGI is dangerous from a person at an AGI company.
    (Note that the posts were published before OpenAI was founded. There's a helpful wiki of OpenAI history here.)
    Hence: a linkpost. I've copied both posts directly below for convenience. I've also bolded a few of the lines I found especially noteworthy.
    Machine intelligence, part 1
    This is going to be a two-part post—one on why machine intelligence is something we should be afraid of, and one on what we should do about it. If you’re already afraid of machine intelligence, you can skip this one and read the second post tomorrow—I was planning to only write part 2, but when I asked a few people to read drafts it became clear I needed part 1.
    WHY YOU SHOULD FEAR MACHINE INTELLIGENCE
    Development of superhuman machine intelligence (SMI) [1] is probably the greatest threat to the continued existence of humanity. There are other threats that I think are more certain to happen (for example, an engineered virus with a long incubation period and a high mortality rate) but are unlikely to destroy every human in the universe in the way that SMI could. Also, most of these other big threats are already widely feared.
    It is extremely hard to put a timeframe on when this will happen (more on this later), and it certainly feels to most people working in the field that it’s still many, many years away. But it’s also extremely hard to believe that it isn’t very likely that it will happen at some point.
    SMI does not have to be the inherently evil sci-fi version to kill us all. A more probable scenario is that it simply doesn’t care about us much either way, but in an effort to accomplish some other goal (most goals, if you think about them long enough, could make use of resources currently being used by humans) wipes us out. Certain goals, like self-preservation, could clearly benefit from no humans. We wash our hands not because we actively wish ill towards the bacteria and viruses on them, but because we don’t want them to get in the way of our plans.
    (Incidentally, Nick Bostrom’s excellent book “Superintelligence” is the best thing I’ve seen on this topic. It is well worth a read.)
    Most machine intelligence development involves a “fitness function”—something the program tries to optimize. At some point, someone will probably try to give a program the fitness function of “survive and reproduce”. Even if not, it will likely be a useful subgoal of many other fitness functions. It worked well for biological life. Unfortunately for us, one thing I learned when I was a student in the Stanford AI lab is that programs often achieve their fitness function in unpredicted ways.
    Evolution will continue forward, and if humans are no longer the most-fit species, we may go away. In some sense, this is the system working as designed. But as a human programmed to survive and reproduce, I feel we should fight it.
    How can we survive the development of SMI? It may not be possible. One of my top 4 favorite explanations for the Fermi paradox is that biological intelligence always eventually creates machine intelligence, which wipes out biological life and then for some reason decides to makes itself undetectable.
    It’s very hard to know how close we are to machine intelligence surpassing human intelligence. Progression of machine intelligence is a double exponential function; human-written programs and computing power are getting better at an exponential rate, and self-learning/self-improving software will improve i...
    14 min
  • LW - Locating Fulcrum Experiences by LoganStrohl
    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: Locating Fulcrum Experiences, published by LoganStrohl on April 28, 2023 on LessWrong.
    Note: There are two appendices at the end of this post. The first contains a summary of the steps I’ve described here. The second is a glossary of key terms I’ve introduced in this essay.
    Once you’ve booted up some curiosity and original seeing—perhaps even identified a question that’s crucial to your story—it’s time to start making observations. At this stage, the purpose of these observations is not so much to find an answer to your question, as to get in closer contact with the world so you’re well positioned to ask better questions.
    But how do you determine which parts of the world are relevant to your topic, especially when you know that your basic conceptualization of the issue may be flawed?
    Standard approaches to this problem include using a working model to make testable predictions, and seeking expert advice. Here we'll take a more bottom-up approach; instead of investigating large models from the start, or relying on the models of others, naturalism focuses first and foremost on the immediate sense data you can personally gather. You'll direct your data-gathering (even in the absence of any coherent model!) by identifying something I call a fulcrum experience.
    What is a fulcrum experience?
    A fulcrum experience is a collection of sensations that would lead you to relate differently to your topic if you observed it closely.
    For example, when I studied courage, “fear” turned out to be a fulcrum experience for me. After really paying attention to experiences of fear in detail, I automatically thought about courage, bravery, cowardice, and related topics quite differently from how I had before. In the past, I’d implicitly assumed that being courageous meant taking actions I am afraid of. By observing fear, I learned about a range of cognitive reactions to the experience of being afraid; I came to think instead that courage has much more to do with where my actions come from, and how that relates to my overall system of values, than with which actions I ultimately take.
    Once I was really paying attention to my experiences surrounding fear, my previous conceptualization simply crumbled. After enough study, a new way of thinking gradually emerged, founded more on direct observation than on vague storytelling. Therefore, fear was a fulcrum experience.
    Locating fulcrum experiences is the first main phase of naturalist study. By “locating fulcrum experiences”, I mean 1) anticipating which experiences will have the power to shift your perspective in such a way that key features of your topic become clearly visible to you, and 2) anticipating where to find those experiences so that you can study them in depth. If you’re able to identify and explore fulcrum experiences, you’ll likely find yourself asking questions based on thoughts you were not previously capable of entertaining.
    Suppose I’m concerned about something to do with where my beliefs come from. My current story is that, “Sometimes I believe things because I think they’re true, but sometimes I believe things for other reasons than that, and I’m worried about the beliefs that are indifferent to truth.” My current quest is, “What leads me to form beliefs other than attempts to guess the truth?”
    It is of course possible to try answering a question like this by just consulting existing models of myself, or of people in general (“Well, probably motivated reasoning!”), but that’s not what comes next in the particular strategy I’m discussing. In this strategy, rather than doing my best to answer the question with whatever information I already have, the purpose of the question is to lead me toward fresh observations.
    It’s often easy to make fresh observations: On the chair in front of me, the pattern of the weave makes many little ...
    27 min
  • EA - Report: Food Security in Argentina in the event of an Abrupt Sunlight Reduction Scenario (ASRS) by JorgeTorresC
    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: Report: Food Security in Argentina in the event of an Abrupt Sunlight Reduction Scenario (ASRS), published by JorgeTorresC on April 27, 2023 on The Effective Altruism Forum.
    In order to prepare in case of an Abrupt Sunlight Reduction Scenario (ASRS), it is necessary to understand the threats and vulnerabilities of the agri-food system and the current ecosystem of risk management in Argentina.
    Through research, modeling, and interviews, the RCG team, in collaboration with ALLFED, has produced a report titled "Food Security in Argentina in the event of an Abrupt Sunlight Reduction Scenario (ASRS)". This strategic proposal offers a comprehensive overview of risk management in Argentina, the current state of risk, and presents eight main recommendations divided into two categories: communication and supplies, and the production and redirection of food in the event of an ASRS. Additionally, it highlights priority actions for implementing the proposed solutions.
    Download full report
    If you are interested in attending our research findings presentation on May 8, 2023 at 19:00 (GMT+1) sign up at the following link
    Executive Summary
    Abrupt Sunlight Reduction Scenarios (ASRS) result from events that eject particulate matter into the upper atmosphere, reflecting and absorbing sunlight that would otherwise reach the Earth's surface. This decrease in sunlight causes a drop in global temperatures and precipitation, with devastating consequences for agriculture. Potential causes of an ASRS include large volcanic eruptions, nuclear winter, and asteroid or comet impact (ALLFED , 2022). The impact of such events is likely to last for several years, even a decade, with global implications for agriculture and food security.
    According to certain researchers, in a severe nuclear winter scenario such as the one described, an estimated 75% of the world's population could starve to death (Xia et al., 2022). If the atmosphere were to collect 150 million tons of soot, it would cause a decrease in temperature ranging between 7ºC and 15ºC. This temperature drop would be accompanied by a reduction in sunlight and precipitation, leading to a collapse in caloric production. Specifically, caloric production would fall to 10%-20% of its current value.
    Some regions of the world appear to have better conditions for surviving an abrupt sunlight reduction scenario (ASRS). These include island nations like New Zealand or Australia (Boyd & Wilson, 2022) and continental countries such as Argentina, Uruguay, and Paraguay (Xia et al., 2022). After evaluating different countries in Latin America, we realized that Argentina is one of the world's leading producers and exporters of food, especially grains and oilseeds. Therefore, in the event of an ASRS, Argentina would play a crucial role in the distribution and food exportation, even if its production decreased during the scenario. It would still have greater food availability than other countries that would be more severely affected.
    Adapting the country's food systems quickly and effectively would make the difference between a national famine situation and producing sufficient, varied, and nutritious food with a surplus to export, thus avoiding a regional humanitarian crisis and a foreign refugee crisis.
    Considering the importance of Argentina's geographical location, it is essential for the government to actively participate in the development of contingency plans aimed at addressing possible threats in the region, and the creation of an interdepartmental working group is recommended to investigate the threat posed by an ASRS and how to deal with it. The strategic initiatives aim to contribute to the strengthening of preparedness and specific recommendations for its location. This report consists of 8 main recommendations, divided into communication and supp...
    7 min
  • EA - Proposals for the AI Regulatory Sandbox in Spain by Guillem Bas
    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: Proposals for the AI Regulatory Sandbox in Spain, published by Guillem Bas on April 27, 2023 on The Effective Altruism Forum.
    Translated by Daniela Tiznado.
    Summary: The European Union is designing a regulatory framework for artificial intelligence (AI) that could be approved by the end of 2023. This regulation prohibits unacceptable practices and stipulates requirements for AI systems in critical sectors. These obligations consist of a risk management system, a quality management system, and post-market monitoring. The legislation enforcement will be tested for the first time in Spain, in a regulatory sandbox of approximately three years. This will be a great opportunity to prepare the national ecosystem and influence the development of AI governance internationally. In this context, we present several policies to consider, including third-party auditing, the detection and evaluation of frontier AI models, red teaming exercises, and creating an incident database.
    Introduction
    Everything indicates that the European Union will become the first major political entity to approve a comprehensive regulatory framework for artificial intelligence (AI). On April 21, 2021, The European Commission presented the Regulation laying down harmonised rules on AI –henceforth AI Act or Act–. This legislative proposal covers all types of AI systems in all sectors except the military, making it the most ambitious plan to regulate AI.
    As we will explain below, Spain will lead the implementation of this regulation in the context of a testing ground or sandbox. This is an opportunity for the Spanish Government to contribute to establishing good auditing and regulatory practices that can be adopted by other member states.
    This article is divided into six sections. Firstly, we provide a brief history of the Act. The second part summarizes the legislative proposal of the European Commission. The third section details the first sandbox of this regulation, carried out in Spain. The fourth lists the public bodies involved in the testing environment. The fifth part explains the relevance of this exercise. Finally, we present proposals to improve the governance of risks associated with AI in this context. We conclude that this project provides an excellent opportunity to develop a culture of responsible AI and determine the effectiveness of various policies.
    Brief History of the Act
    The foundations of the text date back to 2020, when the European Commission published the White Paper on Artificial Intelligence. This was the beginning of a consultation process and a subsequent roadmap that involved the participation of hundreds of stakeholders, resulting in the aforementioned proposal.
    After its publication, the Commission received feedback from 304 actors and initiated a review process involving the European Parliament and the Council of the European Union as legislative bodies. In December 2022, the Council adopted a common approach. In the case of the Parliament, the vote to agree on a joint position is scheduled for May (Bertuzzi, 2023). The trilogue will begin immediately afterward, and the final version could be approved by the end of 2023, entering into force at the beginning of 2024.
    Summary of the Act
    The main starting point of the proposed law is the classification of AI systems according to the level of risk they entail. Specifically, the proposal is based on a hierarchy distinguishing between unacceptable, high, limited, and minimal risks. The first two are the main focus of the regulation.
    As part of the category of unacceptable risks, practices that pose a clear threat to the safety, livelihoods, and rights of people will be banned. Currently, three practices have been deemed unacceptable as they go against European values: distorting human behavior to cause harm; evaluating and classi...
    22 min
  • EA - Life in a Day: The film that opened my heart to effective altruism by Aaron Gertler
    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: Life in a Day: The film that opened my heart to effective altruism, published by Aaron Gertler on April 27, 2023 on The Effective Altruism Forum.
    There are at least two kinds of "EA origin story":
    Some people discover it for themselves. They have EA-flavored goals, they Google something like "do the most good", and they find EA.
    Other people are introduced by others. They have EA-flavored goals, but they don't find EA until someone tells them about it.
    My story is the second kind. I saw Peter Singer give a talk, which I attended because I'd read GiveWell's website, which I'd seen on LessWrong, which I heard about from Harry Potter and the Methods of Rationality.
    So I owe part of my ethical system to the guy who heard me discussing Harry Potter in a college dining hall, interrupted my conversation, gave me a five-minute pitch for HPMOR, and left. (I never saw him again.)
    But what about the rest of the story?
    Peter Singer convinced me to join Giving What We Can and earn to give after college. Hundreds of other people watched the same talk, in the same auditorium, and didn't do those things. There must be some other factor that made me especially receptive to EA (that is, gave me "EA-flavored goals").
    Of course, there are actually dozens of factors, because life is complicated. For example:
    I speak English, and live in a country that had EA presence early on.
    My parents are empathetic people who shaped me into an empathetic person.
    My parents are financially secure people who didn't need my support, so I could easily afford to give 10% of my income.
    I went to the kind of college where Peter Singer gives talks and students advertise HPMOR to total strangers.
    I saw a movie called Life in a Day.
    These are hard to replicate, except the last one, which anyone can do right now:
    The English (foreign language) subtitles seem broken, but English (full text) works.
    How the film works
    YouTube decided to make a documentary. They asked people to film themselves on July 24th, 2010, and share the footage. They received 80,000 submissions and 4500 hours of film from 192 countries. They used it to make a 90-minute film about life on Earth.
    We start and end at midnight. For each part of the day, we jump around the world to see what people are doing. Because people are similar, we see similar actions, in parallel.
    For example, the sun comes up eight minutes into the movie. For the next 90 seconds, we watch people wake up. Some have alarms; some have roosters; some rise with the sun. Some are woken up by parents or lovers. Others wake up alone. One person sleeps on the street and wakes to the passing of cars. Everyone is different. But we all wake up.
    Submitters also had the option to answer questions: "What do you love? What do you fear? What do you have in your pockets or handbag?"
    We get three minutes of fear (ten for love). People are afraid of ghosts, spiders, lions, and small noises in the middle of the night. They are afraid of God, Hell, and people different from themselves. They are afraid of losing childhood, losing their hair, and losing the people they love. We are all afraid of something.
    This is what Life in a Day shows, over and over: In so many ways, we are the same.
    What the film did to me
    This isn't a new idea. You can hear echoes of it in the Golden Rule, "all men are created equal", and "workers of the world, unite!".
    But I didn't really feel the idea until I saw Life in a Day, which conveys it more powerfully than anything else I've ever seen. I felt strange on the drive home; I could no longer look at the world in the same way.
    When I later heard about a philosophy that was dedicated to helping people as much as possible, no matter where they lived, it struck a chord. Without Life in a Day, I'm not sure I'd have felt the same deep sense of "yes, this is obviously righ...
    10 min
  • EA - What are work practices that you’ve adopted that you now think are underrated? 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: What are work practices that you’ve adopted that you now think are underrated?, published by Lizka on April 27, 2023 on The Effective Altruism Forum.
    Basically every time I’ve worked with new people or on a new kind of project, I’ve learned a practice or method that now seems quite important to how I work. I want to see if we can crowd-source more (and discuss them).
    So share things you’ve learned! I’m sharing some as answers on the thread.
    Note: Please don’t hesitate to share things that you think are common. I expect that fewer people know about them than you might think — especially if you’re from a field or industry where the practice is normal. (Relevant xkcd comic.)
    See also:
    Are there robustly good and disputable leadership practices?
    Personal development and practical advice pages on the Forum
    (There's probably more relevant content — please feel free to let me know!)
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min