The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - How I apply Non-Violent Communication by Kaj Sotala
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: How I apply Non-Violent Communication, published by Kaj Sotala on May 15, 2023 on LessWrong.
    In a discussion elsewhere, someone asked about simple communication patterns that they could use with their partner when they're having difficulties and want to avoid causing hurt or verbalizing projections. I ended up sharing an explanation of how I try to apply Non-Violent Communication (NVC), one of the most useful tools for this that I've ran across; and then I figured that I might as well also share those remarks more widely.
    Here's an edited version of things I ended up saying in that conversation.---I think one of the most important parts of NVC is the idea about distinguishing observations and interpretations, where an "observation" is defined as something that you could objectively verify (e.g. by capturing it on camera) and an interpretation is something that blends in more stuff, such as generalizations or assumptions about intent the other person's intent.For example, "You're always late" and "You don't care about my time" are interpretations, "On the last three times when we agreed to meet, you showed up 15 minutes after the agreed-upon time" is an observation.If you can separate those, you can then go into a potentially charged conversation by transforming something like "You are always late, why don't you care about my time" to something like "On the last three times when we agreed to meet, you showed up 15 minutes after the agreed-upon time.
    I found that frustrating because I made sure to be on time and could have spent that extra fifteen minutes to do something else", which is often quite helpful.This doesn't mean you'd need to keep detailed records to express things as observations. If you don't remember earlier specifics, you can just say something like "Hey you were fifteen minutes late today and I think that's happened before too".
    The main intents are to 1) avoid phrasings like "You're always late" or "You don't care about my time" that come across as accusatory or potentially unfair2) be sufficiently specific that the other person understands what you're talking about - the original NVC book has an example of a person being told "You have a big mouth" after he repeatedly started telling personal stories in the middle of meetings, but the feedback wasn't effective because he didn't understand what was meant by thisAs long as those goals are met, it doesn't matter even if the observations are a bit fuzzy.NVC has this famous formula of "observation, feeling, need, request" that it recommends phrasing things in terms of - "First state the observation, then say that when that happened you felt X because you have need Y, then make a request of how you'd like the person to act differently". But people fairly point out that it can make you sound a little robotic and it doesn't always sound very natural.
    What I use in practice is a simplified version which goes something like "State the observation, then explain what about it made you feel bad, taking care to frame things in terms of why it felt bad for you rather than making any claims about the other person's intent or character". (So "it feels bad to me that you're untrustworthy" is out, "it feels bad to me when I feel like I can't trust you" might be okay.) I don't normally put in an explicit request at this point, because just verbalizing my experience usually leads to a conversation about it anyway.Also, while NVC is normally taught as a way to express negative things, I've also found it useful to draw inspiration from it for expressing _positive_ things. For example, I told a friend, "It made me feel nice when you said [X] because I interpreted it to mean that you're prioritizing seeing me over other things that you could do, and I appreciate that". Here I am mentioning an interpretation, but it's a posit...
    4 min
  • EA - How much do markets value Open AI? by Ben West
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: How much do markets value Open AI?, published by Ben West on May 14, 2023 on The Effective Altruism Forum.
    Summary: A BOTEC indicates that Open AI might have been valued at 220-430x their annual recurring revenue, which is high but not unheard of. Various factors make this multiple hard to interpret, but it generally does not seem consistent with investors believing that Open AI will capture revenue consistent with creating transformative AI.
    Overview
    Epistemic status: revenue multiples are intended as a rough estimate of how much investors believe a company is going to grow, and I would be surprised if my estimated revenue multiple was off by more than a factor of 5. But the "strategic considerations" portion of this is a bunch of wild guesses that I feel much less confident about.
    There has been some discussion about how much markets are expecting transformative AI, e.g. here. One obvious question is "why isn't Open AI valued at a kajillion dollars?"
    I estimate that Microsoft's investment implicitly valued OAI at 220-430x their annual recurring revenue. This is high - average multiples are around 7x, but some pharmaceutical companies have multiples > 1000x. This would seem to support the argument that investors think that OAI is exceptional (but not "equivalent to the Industrial Revolution" exceptional).
    However, Microsoft received a set of benefits from the deal which make the EV multiple overstated. Based on adjustments, I can see the actual implied multiple being anything from -2,200x to 3,200x.
    (Negative multiples imply that Microsoft got more value from access to OAI models than the amount they invested and are therefore willing to treat their investment as a liability rather than an asset.)
    One particularly confusing fact is that OAI's valuation appears to have gone from $14 billion in 2021 to $19 billion in 2023. Even ignoring anything about transformative AI, I would have expected that the success of ChatGPT etc. should have resulted in a more than a 35% increase.
    Qualitatively, my guess is that this was a nice but not exceptional deal for OAI, and I feel confused why they took it. One possible explanation is “the kind of people who can deploy $10B of capital are institutionally incapable of investing at > 200x revenue multiples”, which doesn’t seem crazy to me. Another explanation is that this is basically guaranteeing them a massive customer (Microsoft), and they are willing to give up some stock to get that customer.
    Squiggle model here
    It would be cool if someone did a similar write up about Anthropic, although publicly available information on them is slim. My guess is that they will have an even higher revenue multiple (maybe infinite? I'm not sure if they had revenue when they first raised).
    Details
    Valuation: $19B
    A bunch of news sites (e.g. here) reported that Microsoft invested $10 billion to value OAI at $29 billion. I assume that this valuation is post money, meaning the pre-money valuation is 19 billion.
    Although this site says that they were valued at $14 billion in 2021, meaning that they only increased in value 35% the past two years. This seems weird, but I guess it is consistent with the view that markets aren’t valuing the possibility of TAI.
    Revenue: $54M/year
    Reuters claims they are projecting $200M revenue in 2023.
    FastCompany says they made $30 million in 2022.
    If the deal closed in early 2023, then presumably annual projections of their monthly revenue were higher than $30 million, though it's unclear how much.
    Let’s arbitrarily say MRR will increase 10x this year, implying a monthly growth rate of 10^(1/12) = 1.22
    Solving the geometric series of 200 = x (1-1.22^12) / (1 -1.22) we get that their first month revenue is $4.46M, a run rate of $53.52M/year
    Other factors:
    The vast majority of the investment is going to be spent on Micros...
    8 min
  • EA - Consider using the EA Gather Town for your online meetings and events by Siao Si
    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: Consider using the EA Gather Town for your online meetings and events, published by Siao Si on May 14, 2023 on The Effective Altruism Forum.
    Tl;dr
    The EA Gather town is cost-efficient to share among EA organisations - and gives you access to its own highly engaged EA community.
    Ways you can use the EA Gather space
    Have a permanent online ‘office’ for your organisation or group to cowork in
    Run meetings or internal events in the space
    Run public events, either for the broader EA community or other groups
    The space is run by a group of volunteers, so ping one of us on here if you want to talk about setting up a presence there - or just hop onto the space for a while and see how you find it!
    EA Gather Updates
    Gather has reduced their free plan from 25 concurrent users to 10, which has forced many EA groups using it to seek different services, which may also charge or have their own limitations.
    However, CEA has generously agreed to fund the EA Gather Town instance for 30-40 concurrent users, a capacity which EA organisations can freely piggyback on.
    The rest of this post will discuss the benefits and drawbacks of using the space.
    Benefits
    ‘Free’ virtual space
    Given the uneven distribution of usage over a month, almost all of the capacity we’re paying for goes unused. Each org might have a weekly meeting of 1-2 hours, spiking the usage to near capacity, then have 0-5 users online for the rest of the week. So it’s very likely you could run your own weekly meetings at whatever time suits you without any concern about capacity, and virtually certain if you have any flexibility to adjust the times.
    We have a shared event calendar so that you can track whether your usage spikes might overlap. In that event, we have some excess funding to boost capacity.
    Integration with a growing section of the EA community
    EA groups that have used the space regularly include EAGx Cambridge, Charity Entrepreneurship, Anima International, Training for Good, Alignment Ecosystem Development, EA France, EA Hong Kong, Metaculus, and more.
    Last year we were the meetup and hangout space for EAGxVirtual, and hopefully will host many more such events. We also have many independent regular users, who might be future staff of, donors to, beneficiaries of, or otherwise supporters of your group.
    It’s entirely up to you how public your office space is to other users. Some EA groups welcome guests, some are entirely private, some are in between. We’re currently exploring intuitive visual norms to clearly signal the preferences of different groups (feel free to suggest some!).
    Also, your office is not a prison - your members are always welcome and warmly encouraged to join us either for coworking or socialising in the common area :)
    Gather Town native functionality/default norms
    Gather Town has a number of nice features that led us to originally set up this group there and to stay there since:
    Intuitive video call protocol (if you stand near someone, you’re in conversation with them)
    Embeddable webpages (so you can eg have native access to pomodoro timers)
    Cute aesthetic - your office can look like a virtual office, a virtual park, a virtual pirate ship, or anything else you can imagine! You can traverse on foot, by go-kart, or by magic portal :)
    Various other bits of functionality and suggested norms
    Drawbacks
    The main reasons why you might not want to use the space:
    Buggyness
    Gather is relatively new, and has a few more moving parts than other video calling services. It has a few intermittent glitches (most of which can be resolved by refreshing the page). Twice in the last 13 months or so I’m aware of it having gone down for about an hour. If you need perfect uptime, Zoom is probably better. Note there’s both an app and browser version, so one might work substantially better than the other at any given po...
    5 min
  • LW - The Lightcone Theorem: A Better Foundation For Natural Abstraction? 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: The Lightcone Theorem: A Better Foundation For Natural Abstraction?, published by johnswentworth on May 15, 2023 on LessWrong.
    Credit to David Lorell for serving as an active sounding board as the ideas in this post were developed.
    For about a year and a half now, my main foundation for natural abstraction math has been The Telephone Theorem: long-range interactions in a probabilistic graphical model (in the long-range limit) are mediated by quantities which are conserved (in the long-range limit). From there, the next big conceptual step is to argue that the quantities conserved in the long-range limit are also conserved by resampling, and therefore the conserved quantities of an MCMC sampling process on the model mediate all long-range interactions in the model.
    The most immediate shortcoming of the Telephone Theorem and the resampling argument is that they talk about behavior in infinite limits. To use them, either we need to have an infinitely large graphical model, or we need to take an approximation. For practical purposes, approximation is clearly the way to go, but just directly adding epsilons and deltas to the arguments gives relatively weak results.
    This post presents a different path.
    The core result is the Lightcone Theorem:
    Start with a probabilistic graphical model on the variables X1,.,Xn.
    The graph defines adjacency, distance, etc between variables. For directed graphical models (i.e. Bayes nets), spouses (as well as parents and children) count as adjacent.
    We can model those variables as the output of a Gibbs sampler (that’s the MCMC process) on the graphical model.
    Call the initial condition of the sampler X0=(X01,.,X0n). The distribution of X0 must be the same as the distribution of X (i.e. the sampler is initialized “in equilibrium”).
    We can model the sampler as having run for any number of steps to generate the variables; call the number of steps T.
    At each step, the process resamples some set of nonadjacent variables conditional on their neighbors.
    The Lightcone Theorem says: conditional on X0, any sets of variables in X which are a distance of at least 2T apart in the graphical model are independent.
    Yes, exactly independent, no approximation.
    In short: the initial condition of the resampling process provides a latent, conditional on which we have exact independence at a distance.
    This was. rather surprising to me. If you’d floated the Lightcone Theorem as a conjecture a year ago, I’d have said it would probably work as an approximation for large T, but no way it would work exactly for finite T. Yet here we are.
    The Proof, In Pictures
    The proof is best presented visually. High-level outline:
    Perform a do() operation on the Gibbs sampler, so that it never resamples the variables a distance of T from XR.
    In the do()-operated process, X0 mediates between XTR and XTD(R,≥2T), where D(R,≥2T) indicates indices of variables a distance of at least 2T from XR.
    Since X0, XTR and XTD(R,≥2T) are all outside the lightcone of the do()-operation, they have the same joint distribution under the non-do()-operated sampler as under the do()-operated sampler.
    Therefore X0 mediates between XTR and XTD(R,≥2T) under the original sampler.
    We start with the graphical model:
    Within that graphical model, we’ll pick some tuple of variables XR (“R” for “region”). I’ll use the notation XD(R,t) for the variables a distance t away from R, XD(R,>t) for variables a distance greater than t away from R, XD(R,t) (everything more than distance t from XR).
    Next, we’ll draw the Gibbs resampler as a graphical model. We’ll draw the full state Xt at...
    11 min
  • AF - Reward is the optimization target (of capabilities researchers) by Max H
    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: Reward is the optimization target (of capabilities researchers), published by Max H on May 15, 2023 on The AI Alignment Forum.
    In Reward is not the optimization target, @TurnTrout writes:
    Therefore, reward is not the optimization target in two senses:
    Deep reinforcement learning agents will not come to intrinsically and primarily value their reward signal; reward is not the trained agent’s optimization target.
    Utility functions express the relative goodness of outcomes. Reward is not best understood as being a kind of utility function. Reward has the mechanistic effect of chiseling cognition into the agent's network. Therefore, properly understood, reward does not express relative goodness and is therefore not an optimization target at all.
    I think these claims are true and important, though in my own terms, I would rephrase and narrow as:
    Executing a policy trained through current reinforcement learning methods does not necessarily result in a system which takes actions to maximize the reward function.
    I find the claim stated this way more intuitively obvious, but also somewhat less interesting, because the restatement makes the narrowness more explicit.
    In this post, I want to highlight a fact which I did not see mentioned in the original post or the comments: in the field of reinforcement learning, there are agents which are pursuing the goal of building a system which maximizes a reward function, subject to some additional constraints. These agents are the capabilities researchers designing and implementing SotA reinforcement learning algorithms and other methods to build and test the most capable, general systems across a variety of domains.
    Capabilities researchers are not solely concerned with maximizing a particular reward function, since it is not very difficult or interesting to program a bot the old-fashioned way to beat a particular Atari game. For other games (e.g. Go) it is harder to beat human or machine performance by using traditional programming techniques, and trying doesn't make for a compelling AI research agenda.
    Aside from the headline metric of how well a new RL method does in terms of training policies which result in a high reward when executed, RL researchers place importance on:
    Efficiency, in many senses:
    How much computing power does it take to train a policy to perform well in a particular domain?
    How much (and what kind of) training data does it take to train a policy?
    How much computing power and space (i.e. size of the model) does it take to execute the policy?
    Generality of methods across domains: for example, can the same RL training process, applied to the same network architecture but with different training data, be applied to train policies and create systems which play many different types of games well? Dreamer is an example of a very general and powerful RL method.
    Generality of the trained system across domains: can a single system be trained to perform well in a variety of different domains, without retraining? Gato and LLMs are examples of this kind of generality.
    Why is this observation important?
    In my view, current RL methods have not yet advanced to the point of creating systems which can be indisputably described as agents which has any kind of values at all.
    I view most attempts to draw parallels between high-level processes that happen in current-day AI systems and human brains as looking for patterns which do not yet exist.
    Speculating on what kind of agents and values current DL-paradigm RL methods might produce in the future can be valuable and important research, but I think that it is important to remain grounded about what current systems are actually doing, and to be precise with terms.
    As an example of where I think a lack of grounding about current systems and methods leads to things going wrong, in Evolut...
    10 min
  • AF - The Lightcone Theorem: A Better Foundation For Natural Abstraction? 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: The Lightcone Theorem: A Better Foundation For Natural Abstraction?, published by johnswentworth on May 15, 2023 on The AI Alignment Forum.
    Credit to David Lorell for serving as an active sounding board as the ideas in this post were developed.
    For about a year and a half now, my main foundation for natural abstraction math has been The Telephone Theorem: long-range interactions in a probabilistic graphical model (in the long-range limit) are mediated by quantities which are conserved (in the long-range limit). From there, the next big conceptual step is to argue that the quantities conserved in the long-range limit are also conserved by resampling, and therefore the conserved quantities of an MCMC sampling process on the model mediate all long-range interactions in the model.
    The most immediate shortcoming of the Telephone Theorem and the resampling argument is that they talk about behavior in infinite limits. To use them, either we need to have an infinitely large graphical model, or we need to take an approximation. For practical purposes, approximation is clearly the way to go, but just directly adding epsilons and deltas to the arguments gives relatively weak results.
    This post presents a different path.
    The core result is the Lightcone Theorem:
    Start with a probabilistic graphical model on the variables X1,.,Xn.
    The graph defines adjacency, distance, etc between variables. For directed graphical models (i.e. Bayes nets), spouses (as well as parents and children) count as adjacent.
    We can model those variables as the output of a Gibbs sampler (that’s the MCMC process) on the graphical model.
    Call the initial condition of the sampler X0=(X01,.,X0n). The distribution of X0 must be the same as the distribution of X (i.e. the sampler is initialized “in equilibrium”).
    We can model the sampler as having run for any number of steps to generate the variables; call the number of steps T.
    At each step, the process resamples some set of nonadjacent variables conditional on their neighbors.
    The Lightcone Theorem says: conditional on X0, any sets of variables in X which are a distance of at least 2T apart in the graphical model are independent.
    Yes, exactly independent, no approximation.
    In short: the initial condition of the resampling process provides a latent, conditional on which we have exact independence at a distance.
    This was. rather surprising to me. If you’d floated the Lightcone Theorem as a conjecture a year ago, I’d have said it would probably work as an approximation for large T, but no way it would work exactly for finite T. Yet here we are.
    The Proof, In Pictures
    The proof is best presented visually. High-level outline:
    Perform a do() operation on the Gibbs sampler, so that it never resamples the variables a distance of T from XR.
    In the do()-operated process, X0 mediates between XTR and XTD(R,≥2T), where D(R,≥2T) indicates indices of variables a distance of at least 2T from XR.
    Since X0, XTR and XTD(R,≥2T) are all outside the lightcone of the do()-operation, they have the same joint distribution under the non-do()-operated sampler as under the do()-operated sampler.
    Therefore X0 mediates between XTR and XTD(R,≥2T) under the original sampler.
    We start with the graphical model:
    Within that graphical model, we’ll pick some tuple of variables XR (“R” for “region”). I’ll use the notation XD(R,t) for the variables a distance t away from R, XD(R,>t) for variables a distance greater than t away from R, XD(R,t) (everything more than distance t from XR).
    Next, we’ll draw the Gibbs resampler as a graphical model. We’ll draw the ful...
    11 min
  • LW - Helping your Senator Prepare for the Upcoming Sam Altman Hearing by Tiago de Vassal
    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: Helping your Senator Prepare for the Upcoming Sam Altman Hearing, published by Tiago de Vassal on May 14, 2023 on LessWrong.
    Hello Everyone,
    Next Tuesday, the US Senate will hold a hearing on AI (). OpenAI CEO Sam Altman will testify before Congress for the first time.
    We believe this is a great opportunity to bring awareness to some of the risks posed by AI technologies, and build common knowledge among US policymakers on possible solutions.
    If you live in the United States, you can play an important role as a constituent. Please reach out to your senator who will be present at this hearing, and voice your concerns about AI safety. To help them navigate the debate, we propose a list of questions, initially drafted by Siméon to share with the senators present at the hearing ().
    We’ve made a webpage where you can find out if a senator from your state will be present at the hearing. This page also includes the list of questions you can send them, a structured script for both call and email, as well as the necessary contact information./#senate
    AI Safety Tour is an initiative dedicated to mitigating AI existential risks through the active engagement of experts and related public figures in the public debate. You can learn more about us here: aisafetytour.com. If you click the join button, you’ll get to a form where you can reach out to us. You can also email us at [email protected]. And of course we’ll read every single comment on this page.
    Thanks,
    Tiago & Siméon
    AI Safety Tour
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min
  • AF - Why don't quantilizers also cut off the upper end of the distribution? by Alex Altair
    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 don't quantilizers also cut off the upper end of the distribution?, published by Alex Altair on May 15, 2023 on The AI Alignment Forum.
    It seems to me that the main goal of quantilization is to reduce the extreme unintended outcomes of maximizing (by sampling from something like a human-learned distribution over actions) while still remaining competitive (by sampling from only the upper quantile of said distribution).
    But that still leaves open the possibility of sampling one of those super highly maximized actions! Why not just hard-cut off the upper portion of the distribution as well? Why not have a quantilizer that samples between the 90%ile and the 99%ile?
    Or, while we're at it, why are we sampling at all? Why not just say that we're taking the 99%ile action?
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min
  • AF - Difficulties in making powerful aligned AI by DanielFilan
    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: Difficulties in making powerful aligned AI, published by DanielFilan on May 14, 2023 on The AI Alignment Forum.
    Here’s my breakdown of the difficulties involved in ensuring powerful AI makes our lives radically better, rather than taking over the world, as well as some reasons why I think they’re hard. Here are things it’s not:
    It’s not primarily a justification of why very powerful AI is possible or scary (altho it briefly discusses why very powerful AI would be scary).
    It’s not primarily a list of underlying factors that cause these difficulties (altho it does include and gesture to some of those).
    It’s not at all original - basically everything here has been said many times before, plausibly more eloquently.
    That said, it is my attempt to group the problems in my own words, in a configuration that I haven’t seen before, with enough high-level motivation that one can hopefully tell the extent to which advances in the state of the art address them.
    1. What sort of thinking do we want?
    The first difficulty: we don’t have a sense of what sort of thinking we would want AI systems to use, in sufficient detail that one could (for instance) write python code to execute it. Of course, some of the difficulty here is that we don’t know how smart machines think, but we can give ourselves access to subroutines like “do perfect Bayesian inference on a specified prior and likelihood” or “take a function from vectors to real numbers and find the vector that minimizes the function” and still not solve the problem. To illustrate:
    Take a hard-coded goal predicate, consider a bunch of plans you could take, and execute the plan that best achieves the goal? Unfortunately, the vast majority of goals you could think of writing down in an executable way will incentivize behaviour like gaining control over sources of usable energy (so that you definitely have enough to achieve your goal, and to double- and triple-check that you’ve really achieved it) and stopping other agents from being able to meddle with your plans (because if they could, maybe they’d stop you from achieving your goal).
    Do things that maximize the number of thumbs up you get from humans?1 Best plan: take control of the humans, force them to give you a thumbs up, or trick them into doing so. Presumably this is possible if you’re much smarter than humans, and it’s more reliable than doing good things - some people might not see why your good thing is actually good if left to their own devices.
    Look at humans, figure out what they want based on what they’re doing, and do whatever that is? Main problem: people don’t do the literally optimal thing for what they want. For instance, when people play chess, they usually don’t play perfect moves - even if they’re experts! You need some rule that tells you what people would do if they wanted some goal or another, but it’s not clear what this rule would be, it’s not clear how you make this rule more in line with reality if you never observe “wanting”, and so this ends up having essentially the same problems as plans 1 and 2.
    Read some text written by humans about what they’d like you to do, and do that?2 This is passing the buck to the text written by humans to specify how we want the AI to think, but that’s precisely the problem we’re trying to solve. Concretely, one way you could imagine doing this is to write something relatively informal like “Please be helpful and harmless to your human operators”, and have your AI correctly understand what we mean by that. That (a) presumes that there is a coherent thing that we mean by that (which doesn’t seem obvious to me, given our difficulty in explicitly formalizing this request), and (b) passes the specification buck to the problem of specifying how you should understand this request.
    It’s not a priori definitely impossible to build a ...
    15 min
  • LW - Jaan Tallinn's 2022 Philanthropy Overview by jaan
    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: Jaan Tallinn's 2022 Philanthropy Overview, published by jaan on May 14, 2023 on LessWrong.
    to follow up my philantropic pledge from 2020, i've updated my philanthropy page with 2022 results.
    in 2022 i made $23M worth of endpoint grants ($22.9M after various admin costs), exceeding my commitment of $19.9M (20k times $993.64 — the minimum price of ETH in 2022).
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min