The Nonlinear Library

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: Possible OpenAI's Q* breakthrough and DeepMind's AlphaGo-type systems plus LLMs, published by Burny on November 23, 2023 on LessWrong. tl;dr: OpenAI leaked AI breakthrough called Q*, acing grade-school math. It is hypothesized combination of Q-learning and A*. It was then refuted. DeepMind is working on something similar with Gemini, AlphaGo-style Monte Carlo Tree Search. Scaling these might be crux of planning for increasingly abstract goals and agentic behavior. Academic community has been circling around these ideas for a while. https://www.reuters.com/technology/sam-altmans-ouster-openai-was-precipitated-by-letter-board-about-ai-breakthrough-2023-11-22/ https://twitter.com/MichaelTrazzi/status/1727473723597353386 "Ahead of OpenAI CEO Sam Altman's four days in exile, several staff researchers sent the board of directors a letter warning of a powerful artificial intelligence discovery that they said could threaten humanity Mira Murati told employees on Wednesday that a letter about the AI breakthrough called Q* (pronounced Q-Star), precipitated the board's actions. Given vast computing resources, the new model was able to solve certain mathematical problems. Though only performing math on the level of grade-school students, acing such tests made researchers very optimistic about Q*'s future success." https://twitter.com/SilasAlberti/status/1727486985336660347 "What could OpenAI's breakthrough Q* be about? It sounds like it's related to Q-learning. (For example, Q* denotes the optimal solution of the Bellman equation.) Alternatively, referring to a combination of the A* algorithm and Q learning. One natural guess is that it is AlphaGo-style Monte Carlo Tree Search of the token trajectory. It seems like a natural next step: Previously, papers like AlphaCode showed that even very naive brute force sampling in an LLM can get you huge improvements in competitive programming. The next logical step is to search the token tree in a more principled way. This particularly makes sense in settings like coding and math where there is an easy way to determine correctness. https://twitter.com/mark_riedl/status/1727476666329411975 "Anyone want to speculate on OpenAI's secret Q* project? Something similar to tree-of-thought with intermediate evaluation (like A*)? Monte-Carlo Tree Search like forward roll-outs with LLM decoder and q-learning (like AlphaGo)? Maybe they meant Q-Bert, which combines LLMs and deep Q-learning Before we get too excited, the academic community has been circling around these ideas for a while. There are a ton of papers in the last 6 months that could be said to combine some sort of tree-of-thought and graph search. Also some work on state-space RL and LLMs." https://www.theverge.com/2023/11/22/23973354/a-recent-openai-breakthrough-on-the-path-to-agi-has-caused-a-stir OpenAI spokesperson Lindsey Held Bolton refuted it: "refuted that notion in a statement shared with The Verge: "Mira told employees what the media reports were about but she did not comment on the accuracy of the information."" https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt/ Google DeepMind's Gemini, that is currently the biggest rival with GPT4, which was delayed to the start of 2024, is also trying similar things: AlphaZero-based MCTS through chains of thought, according to Hassabis. Demis Hassabis: "At a high level you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models. We also have some new innovations that are going to be pretty interesting." https://twitter.com/abacaj/status/1727494917356703829 Aligns with DeepMind Chief AGI scientist Shane Legg saying: "To do really creative problem solving you need to start searching." https://twitter.com/iamgingertrash/status/1727482695356494132 "...

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 to publish research in peer-reviewed journals, published by Ren Springlea on November 23, 2023 on The Effective Altruism Forum. This is a guide to publishing research in peer-reviewed journals, particularly for people in the EA community who might be keen to publish research but haven't done so before. About this article: This guide is aimed at people who know how to conduct research but are new to publishing in peer-reviewed journals. I've had a few people ask me about this, so I thought I'd compile all of my knowledge in this article. I assume that the audience of this article already: know what academic research is; know how to conduct academic research (e.g. at the level of a Masters degree or perhaps a research-focused Bachelors degree); have a basic grasp of what a journal is; and have the skills to read scientific publications. This article is based on my own views and my own experiences in academic publishing. I expect that there will be many academics who have a different view or a different strategy. My own experiences come from publishing in ecology, economics, agriculture/fisheries, psychology, and science communication. Should you publish your work in a peer-reviewed journal? Advantages of publishing in peer-reviewed journals Publication in a journal can make your research appear more credible. This won't always matter, but my colleagues and I have encountered a few instances during animal advocacy where stakeholders (e.g. food retail companies, government policymakers) find research more compelling if it is published in a journal. Publication in a journal can make you appear more credible. This is particularly true in non-EA circles, like if you want to apply for research jobs or funding from outside of the EA community. Key ideas can be more easily noticed and adopted by other academics, and perhaps other stakeholders like government policymakers. I don't know how influential this effect is. Your research can be more easily criticised by academics. This can provide an important voice of critique from experts outside of the EA community, which could be one way to detect if a piece of research is flawed in some way. This is my motivation for publishing a study I conducted where I used an economic model to estimate the impact of slow-growing broilers on aggregate animal welfare - submitting a paper like this for peer review is a great way to get feedback from experts in that specific branch of economics. Drawbacks of publishing in peer-reviewed journals Publications are not impact. Publishing your work is a tool that can sometimes help you to achieve impact. If we're trying to do the most good in the world, that may sometimes involve publishing peer-reviewed research (see above). In other cases, it'll be better to spend that time on more impactful work. Not all research is suitable for peer-reviewed journals. For example, in the EA community, it is common to conduct prioritisation research to determine the most promising interventions or strategies, like our recent work on fish advocacy in Denmark. That fish advocacy report would only be interesting to advocacy organisations and doesn't contribute any new understanding of reality beyond the strategy of one organisation, so it is probably not be publishable in a journal (though a smaller version focused on the Appendix of that report may be publishable). Publishing in peer-reviewed journals costs time and energy. When I publish in a peer-reviewed journal (compared to when I publish a report on my organisation's website), I usually spend a couple of extra days writing the draft, and then a few extra days addressing peer review comments over the following months. Peer-reviewed papers have significantly longer timelines. After the first draft of a manuscript is done, it might take 6 or 12 months or even l...

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: 3. Uploading, published by Roger Dearnaley on November 23, 2023 on The AI Alignment Forum. Part 3 of AI, Alignment, and Ethics. This will probably make more sense if you start with Part 1. One Upload, One Vote Let us suppose that, sooner or later, we have some means of doing whole-brain uploading and simulation. Possibly the reading process is destructive, and produces about a liter of greyish-pink goo as a by-product, or possibly not. Or perhaps it's more of an emulation effect, just one based on gigabytes, terabytes, or petabytes of data about an individual, rather than actually simulating every synapse and neurotransmitter flow, such that you end up with a sufficiently accurate emulation of their behavior. Bearing in mind the Bitter Lesson, this might even be based on something like a conventional transformer architecture that doesn't look at all like the inner complexities of human brain, but which has been trained to emulate it in great detail (possibly pre-trained for humans in general, then fine-tuned on a specific individual), perhaps even down to some correlates of detailed neural firing patterns. The technical details don't really matter much. I am not in fact a carbon-chauvinist. These uploads are person-like, intelligent and agentic systems, they have goals, they can even talk, and, importantly, their goals and desires are exactly what you'd expect for a human being. They are a high fidelity-copy of a human, they will have all the same desires and drives as a human, and they will get upset if they're treated as slaves or second-class citizens, regardless of how much carbon there may or many not be in the computational substrate they're running on. Just like you or I would (or indeed as likely would any member of pretty-much any sapient species evolved via natural selection). If someone knew in advance that uploading themself meant becoming a slave or a second-class citizen, they presumably wouldn't do it, perhaps short of this being the only way to cheat death. They'd also campaign, while they were still alive, for upload rights. So we need to either literally or effectively forbid uploading, or else we need give uploads human rights, as close as we reasonably can. Unlike the situation for AIs, there is a very simple humna-fairness-instinct-compatible solution for how to count uploads in an ethical system. They may or may not have a body now, but they did once. So that's what gets counted: the original biological individual, back when they were individual. Then, if you destructively upload yourself, your upload inherits your vote and your human rights, is counted once in utility summations, and so forth. If your upload then duplicates themself, backs themself up, or whatever, there's still only one vote/one set of human rights/one unit of moral worth to go around between the copies, and they or we need some rules for how to split or assign this. Or, if you non-destructively upload yourself, you still only have one vote/set of human rights/etc, and it's now somehow split or assigned between the biological original of you still running on your biological brain and the uploaded copy of you, or even multiple copies of your upload. With this additional rule, then the necessary conditions for the human fairness instinct to make sense are both still obeyed in the presence of uploads: they care about the same good or bad things as us, and via this rule they can be counted. So that's really the only good moral solution that fir the human sense of fairness. OK, so we give uploads votes/human rights/moral worth. What could go wrong? I have seen people on Less Wrong assume that humans must automatically aligned with human values - I can only imagine on the basis that "they have human values, so they must be aligned to them" This is flat out, dangerously false. Please...

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: GWWC is funding constrained (and prefers broad-base support), published by Luke Freeman on November 23, 2023 on The Effective Altruism Forum. Giving What We Can is making giving effectively and significantly a cultural norm - and raising a lot of funds for highly effective charities. We're currently seeking funding to continue our work and ensure that we can inspire many more people to give effectively in the future. In 2024, we're hoping to hit 10,000 lifetime pledges. At Giving We We Can, we encourage people to give more and give better. Give more: We encourage people to pledge to give at least 10% of their income until the day they retire. Give better: We provide a donation platform that makes it easy for people to donate to our recommended high-impact charities. Over 8,500 people have taken the Giving What We Can Pledge to donate at least 10% of their income, and have collectively donated over $300 million. By 2030, we want to get to 100,000 pledgers and well over $1 billion of donations. Our ultimate mission is to make donating at least 10%, as effectively as possible, the global norm. We do this in three key ways: Our pledge: which has inspired a movement of donors to give more significantly, more sustainably, & more effectively. Our expertise: which helps donors to give more effectively across a diversity of causes and worldviews. Our donation platform: which makes effective giving easy & accessible for half a billion people on our expanding list of countries (more coming in 2024!). Our audience We believe that many people are in a position to do a lot of good by giving effectively. We aim to change the norms around giving, encouraging people to be more impactful and generous. Our pitch A decade of charity research has revealed something huge: The best charitable interventions often have 100x more impact per dollar than average ones At GWWC, we help donors find those opportunities (leveraging thousands of hours of research) & make them easy to donate to via our donation platform. Our impact From 2020 to 2022, we estimate that we caused $45 million to go to charity. Once we account for the value of new pledge commitments, we estimate we generated $62 million in value. These figures are our best guess of how much we caused to go to highly effective charities - they don't count money that would have been given anyway or money given to charities we aren't sure are effective. The monetary impact of GWWC is best documented in our most recent Impact Evaluation, which suggests that from 2020 to 2022: GWWC generated an additional $62 million in value for highly-effective charities. GWWC had a giving multiplier of 30x, meaning that for each $1 spent on our operations, we generated $30 of value to highly-effective charities on average. Please note that this isn't a claim that your additional dollar will have a 30x multiplier, even though we think it will still add a lot of value. Read more on how to interpret our results. Each new GWWC Pledge generates >$20,000 of value for highly-effective charities that would not have happened without GWWC. This evaluation suggests something we long suspected: If your goal is to get resources into the hands of highly-effective charities, we believe supporting Giving What We Can is a great funding opportunity. The cultural impact of GWWC (although harder to quantify) has also been significant by making the idea of giving 10% effectively more accessible and compelling to a broader audience. "Pledging 10% to effective charities" has become a touchstone of the effective giving community - inspiring TED talks, launching clubs, & drawing curiosity & praise from press around the world. Our plans We believe most of our impact lies in the coming decades, and Giving What We Can has spent the past 3.5 years building a sustainable foundation for...

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: Thomas Kwa's research journal, published by Thomas Kwa on November 23, 2023 on The AI Alignment Forum. Thomas Kwa Why I'm writing this Research feedback loops for junior AI safety researchers are pretty poor right now. They're better than in the past due to the recent explosion in empirical work, but AI safety is still a very new field whose methodology is not quite nailed down, and which cannot admit the same level of aimlessness as other sciences, or ML at large. There are very likely mistakes being repeated by scientists after scientist, and hopefully I can alleviate the problem slightly by publicly writing about successes and failures in my research process. This is a monologue containing my ~daily research thoughts, using the LW dialogue format because it allows continuous publication. Hopefully this lets some people either give me feedback or compare their research process to mine. If you do have some thoughts, feel free to leave a comment! People I'm collaborating with might occasionally leave dialogue entries. With that, let's share the state of my work as of 11/12. Currently I have a bunch of disconnected projects: Characterize planning inside KataGo by retargeting it to output the worst move (with Adria Garriga-Alonso). Improve circuit discovery by implementing edge-level subnetwork probing on sparse autoencoder features (with Adria and David Udell). Create a tutorial for using TransformerLens on arbitrary (e.g. non-transformer) models by extending 'HookedRootModule', which could make it easy to use TransformerLens for e.g. ARENA 3.0 projects. Create proofs for the accuracy of small neural nets in Coq (with Jason Gross and Rajashree Agrawal). Create demonstrations of catastrophic regressional Goodhart and possibly strengthen theoretical results. Help Peter Barnett and Jeremy Gillen wrap up some threads from MIRI, including editing an argument that misaligned mesa-optimizers are very likely. I plan to mostly write about the first three, but might write about any of these if it doesn't make things too disorganized. Thomas Kwa Monday 11/13 I did SERI MATS applications and thought about Goodhart, but most of my time was spent on the KataGo project. I might more about it later, but the idea is to characterize the nature of planning in KataGo. Early training runs had produced promising results-- a remarkably sparse mask lets the network output almost the worst possible move as judged by the value network-- but I was a bit suspicious that the hooked network was implementing some trivial behavior, like always moving in the corner. I adapted some visualization code previously used for FAR's adversarial Go attack paper to see what the policy was doing, and well... Turns out the network is doing the trivial behavior of moving in either the top left or bottom left corner. I wish I had checked this earlier (it took ~14 days of work to get here), but it doesn't kill this project-- I can just redo the next training run to only allow moves on the 3rd line or above, and hopefully the worst behavior here won't be so trivial. Tomorrow I'm going to table this project and start implementing edge-level subnetwork probing-- estimate is 2 days for the algorithm and maybe lots more effort to run benchmarks. Thomas Kwa Wednesday 11/15 Updates from the last two days: I finished the basic edge-level subnetwork probing code over the last two days. This is exciting because it might outperform ACDC and even attribution patching for circuit discovery. The original ACDC paper included a version of subnetwork probing, but that version was severely handicapped because it operated on the node level (structural pruning) rather than edge level. Adria is now on vacation, so I'm planning to get as far as I can running experiments before getting stuck somewhere and coming back to this after Thank...

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: AMA: GWWC research team, published by Sjir Hoeijmakers on November 22, 2023 on The Effective Altruism Forum. We're the research team at Giving What We Can: Alana (research communicator) Michael (researcher) Sjir (research director) Ask us anything! We'll be answering questions Monday the 27th from 2pm UTC until Tuesday the 28th at 9pm UTC. Please post your questions in advance as comments to this post. And please upvote the questions you'd like us to answer most. We'll do our best to answer as many as we can, though we can't guarantee we'll be able to answer all of them. You might want to ask about the evaluations of evaluators we just published, or - relatedly - about our new fund and charity recommendations and GWWC cause area funds (which we will launch Monday alongside other website updates). We are also happy to answer any questions you may have about our research plans for next year, about the impact evaluation we did earlier this year, about GWWC more broadly, or about anything else you are interested in! Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

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: so you want to save the world? an account in paladinhood, published by Tamsin Leake on November 22, 2023 on LessWrong. Aroden, who gave me the strength to be a holy warrior, give me also the wisdom to choose my fights wisely … and help me grow strong enough to save everyone, literally everyone, and do not let my strength come coupled with contempt for weakness, or my wisdom with contempt for foolishness … and look after everyone … in this strange country and in all the worlds You travelled to and any worlds You didn't. And make Hell cease. Iomedae, paladin of Aroden; from lintamande's glowfic "in His strength, I will dare and dare and dare until I die" introduction. a couple years ago, i was struggling to get myself to think abount alignment instead of making my video game and reading doujins. today, i find that my main remaining bottleneck to my dignity-point output is my physical stamina, and that i might make an actual significant difference to p(doom). this post is about how i got there. it is told in the imperative tense, because i hope that this serves as advice; but be aware that the space of human minds is vastly diverse, even just within lesswrong rationalists, and what worked for me might not work for you. (that said, it doesn't look like there's very many people striving for paladinhood-or-similar out there.) what's an iomedae? (if you don't know what a glowfic is, maybe check out this review of yudkowsky and lintamande's glowfic, planecrash.) in glowfics involving linta!golarion (lintamande's interpretation of golarion, the setting of the tabletop RPG pathfinder), iomedae is the lawful-good goddess of "defeating evil" - or, as lantalótë puts it: Iomedae's a Good goddess but she's not the goddess of any of those. She's not a goddess of nice things. She's the goddess of prioritization, of looking at the world with all its horrors and saying 'There will be time enough for love and beauty and joy and family later. But first we must make the world safe for them.' before ascending to godhood, iomedae was a paladin of Aroden, the god of civilization. various lintamande glowfics feature here in various setting, but "in His strength, I will dare and dare and dare until I die" in particular features her not long after she became a paladin, getting isekai'd in america and then adopted by child protective services, and trying to understand this strange world she's appeared in, and what it means for her to do good there. (throughout this post, when i paste a quote without mentioning its source, its source will be that particular glowfic.) what do i mean by "paladin"? what linta!iomedae's paladinhood represents to me, is being willfully entirely devoted to doing what matters; to do whatever it takes to save the world. her paladinhood - that is, the thing of her being a paladin - is not a burden that she bears; it is what she wants to do, deep in her heart. i would've cheerfully pressed a button that instantly transformed my mindset into that of iomedae; i reflectively endorsed becoming someone who does whatever it takes to save the world, even at very large costs to myself. but human brains are strange apparatus! we have no such button; if we want to change something in ourselves, we have to do a lot of hard work. i find myself mostly on the other side of that hard work, now, satisfied with what i've become. it is immensely wasteful that i didn't go through this seven years earlier, when i first became familiar with the sequences and other rationalist work. but i don't dwell on it, because dwelling on it does not, in fact, help save the world. (note that the notion of paladinhood is not one that i originally was aware of, when i decided to try hard to save the world. it is one that i learned after-the-fact, and thought it fit me well, as an aesthetic. my brain likes ae...

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: Donation Election rewards, published by Will Howard on November 22, 2023 on The Effective Altruism Forum. It's been great to see everyone contributing to the Donation Election, both with your posts and with your donations. If you weren't finding it exciting enough already, we're now offering rewards for different donation tiers! There are both individual rewards, and collective ones for if the fundraiser reaches certain milestones overall. You can donate to the fund here , and you can learn more about the Donation Election in the giving portal or in these posts. Now here's a more detailed description of the rewards: Individual rewards $10: You can add a badge next to your name saying that you donated. This will last until around the end of December when the fundraiser closes. I've set this on my own profile so you can see what it looks like[1]. $50: Someone who is bad at drawing will draw a picture of the animal or being of your choice. This will be a digital drawing[2], probably done by one of the CEA Online team. $100: Someone who is good at drawing will draw a picture of the animal or being of your choice (otherwise as above). $250: We will draw a portrait of you (or whoever you like) with Giving Season vibes - hopefully worthy of being used as your profile picture here or on other sites. To claim your individual rewards you can DM @EA Forum Team with a screenshot of your donation receipt[3], saying which rewards you want to claim (i.e. you don't have to claim the higher rewards if you don't want them). You can do this even if you donated before these were announced. Note that we may only do a limited number of drawings if it turns out we get loads of requests, but I'll warn you here beforehand if it looks like this is going to be the case. Collective rewards $40,000: We'll make a Forum yearbook photo, where anyone can be included if they want (by submitting a picture or telling us to use your profile pic), and we'll make this the banner image of the Community page for the next year. $50,000: To celebrate this win for democracy with even more democracy, we'll let Forum users vote for the next[4] small feature we build. This will be chosen from a set of 5+ that we select beforehand. Here's some that might end up on the list to give you an idea: A dark mode toggle on the Frontpage (currently it's hidden in account settings) Forum-native polls The ability to mute/block people (so you wouldn't see their posts or comments, and they wouldn't be able to message you) Reactions on DMs The ability to sign up for job ads based on your preferences AI generated preview images for posts $75,000: To celebrate this huge win for democracy, we'll let you vote for the next big feature we build. As above this will be decided from a fixed list, and here are some examples of big features we might put on the list: Posts private to logged-in users Private notes/highlights on posts Being able to react (agree/disagree/heart etc) to a specific passage in a post An easily accessible page for browsing editions of the Digest (a weekly curated list of top posts) A native way to limit your usage of the Forum to prevent doomscrolling The ability to set reminders and/or "snooze" posts to read later $100,000: To celebrate tail risk events, we will host a "Lesswrong Freaky Friday", where we dress up the Forum as LessWrong for the day and all try and act like LessWrongers. For the vote-on-a-feature ones, please feel free to comment here (or DM me) with features that you would like to be included in the vote, and we'll include them if they're popular and the right size. ^ For if you're reading this far in the future: ^ Unless you happen to come to Trajan House in Oxford in which case you may be able to get a physical version ^ You should get a receipt by email, we just need a screenshot of the part t...

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: OpenAI: The Battle of the Board, published by Zvi on November 22, 2023 on LessWrong. Previously: OpenAI: Facts from a Weekend. On Friday afternoon, OpenAI's board fired CEO Sam Altman. Overnight, an agreement in principle was reached to reinstate Sam Altman as CEO of OpenAI, with an initial new board of Brad Taylor (ex-co-CEO of Salesforce, chair), Larry Summers and Adam D'Angelo. What happened? Why did it happen? How will it ultimately end? The fight is far from over. We do not entirely know, but we know a lot more than we did a few days ago. This is my attempt to put the pieces together. This is a Fight For Control; Altman Started it This was and still is a fight about control of OpenAI, its board, and its direction. This has been a long simmering battle and debate. The stakes are high. Until recently, Sam Altman worked to reshape the company in his own image, while clashing with the board, and the board did little. While I must emphasize we do not know what motivated the board, a recent power move by Altman likely played a part in forcing the board's hand. OpenAI is a Non-Profit With a Mission The structure of OpenAI and its board put control in doubt. Here is a diagram of OpenAI's structure: Here is OpenAI's mission statement, the link has intended implementation details as well: This document reflects the strategy we've refined over the past two years, including feedback from many people internal and external to OpenAI. The timeline to AGI remains uncertain, but our Charter will guide us in acting in the best interests of humanity throughout its development. OpenAI's mission is to ensure that artificial general intelligence (AGI) - by which we mean highly autonomous systems that outperform humans at most economically valuable work - benefits all of humanity. We will attempt to directly build safe and beneficial AGI, but will also consider our mission fulfilled if our work aids others to achieve this outcome. OpenAI warned investors that they might not make any money: The way a 501(c)3 works is essentially that the board is answerable to no one. If you have a majority of the board for one meeting, you can take full control of the board. But does the board have power? Sort of. It has a supervisory role, which means it can hire or fire the CEO. Often the board uses this leverage to effectively be in charge of major decisions. Other times, the CEO effectively controls the board and the CEO does what he wants. A critical flaw is that firing (and hiring) the CEO, and choosing the composition of a new board, is the board's only real power. The board only has one move. It can fire the CEO or not fire the CEO. Firing the CEO is a major escalation that risks disruption. But escalation and disruption have costs, reputational and financial. Knowing this, the CEO can and often does take action to make them painful to fire, or calculates that the board would not dare. Sam Altman's Perspective While his ultimate goals for OpenAI are far grander, Sam Altman wants OpenAI for now to mostly function as an ordinary Big Tech company in partnership with Microsoft. He wants to build and ship, to move fast and break things. He wants to embark on new business ventures to remove bottlenecks and get equity in the new ventures, including planning a Saudi-funded chip factory in the UAE and starting an AI hardware project. He lobbies in accordance with his business interests, and puts a combination of his personal power, valuation and funding rounds, shareholders and customers first. To that end, over the course of years, he has remade the company culture through addition and subtraction, hiring those who believe in this mission and who would be personally loyal to him. He has likely structured the company to give him free rein and hide his actions from the board and others. Normal CEO did n...

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: GWWC's evaluations of evaluators, published by Sjir Hoeijmakers on November 22, 2023 on The Effective Altruism Forum. The Giving What We Can research team is excited to share the results of our first round of evaluations of charity evaluators and grantmakers! After announcing our plans for a new research direction last year, we have now completed five[1] evaluations that will inform our donation recommendations for this giving season. There are substantial limitations to these evaluations, but we nevertheless think that this is a significant improvement on the status quo, in which there were no independent evaluations of evaluators' work. We plan to continue to evaluate evaluators, extending the list beyond the five we've covered so far, improving our methodology, and regularly renewing our existing evaluations. In this post, we share the key takeaways from each of these evaluations, and link to the full reports. Our website will be updated to reflect the new fund and charity recommendations that came out of these evaluations (alongside many other updates) on Monday, the 27th. We are sharing these reports in advance of our website update so those interested have time to read them and can ask questions before our AMA next Monday and Tuesday. We're also sharing some context about why and how we evaluate evaluators, which will be included in our Monday website update as well. One other exciting (and related) announcement: we'll be launching our new GWWC cause area funds on Monday! These funds (which you'll see referenced in the reports) will make grants based on our latest evaluations of evaluators, advised by the evaluators we end up working with.[2] We are launching them to provide a strong and easy default donation option for donors, and one that will stay up-to-date over time (i.e., donors can set up a recurring donation to these funds knowing that it will always be allocated based on GWWC's latest research). donation platform as well. We look forward to your questions and comments, and in particular to engaging with you in our AMA! (Please note that we may not be able to reply to many comments until then, as we are finalising the website updates and some of us will be on leave.) Global health and wellbeing GiveWell (GW) Based on our evaluation, we've decided to continue to rely on GW's charity recommendations and to ask GW to advise our new GWWC Global Health and Wellbeing Fund. Some takeaways that inform this decision include: GW's overall processes for charity recommendations and grantmaking are generally very strong, reflecting a lot of best practices in finding and funding the most cost-effective opportunities. GW's cost-effectiveness analyses stood up to our quality checks. We thought its work was remarkably evenhanded (we never got the impression that the evaluations were exaggerated), and we generally found only minor issues in the substance of its reasoning, though we did find issues with how well this reasoning was presented and explained. We found it noteworthy how much subjective judgement plays a role in its work, especially with how GW compares different outcomes (like saving and improving lives), and also in some key parameters in its cost-effectiveness analyses supporting deworming. We think reasonable people could come to different conclusions than GW does in some cases, but we think GW's approach is sufficiently well justified overall for our purposes. For more, please see the evaluation report. Happier Lives Institute (HLI) We stopped this evaluation short of finishing it, because we thought the costs of finalising it outweighed the potential benefits at this stage. For more on this decision and on what we did learn about HLI, please see the evaluation report. Animal welfare EA Funds' Animal Welfare Fund (AWF) Based on our evaluation, we've decide...