The Nonlinear Library: LessWrong

Link to original articleWelcome 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: Sherlockian Abduction Master List, published by Cole Wyeth on July 13, 2024 on LessWrong. [Radically updated with many new entries around 07/10/24] Epistemic status: The List has been tested in the real world by me (with mixed results) and extensively checked for errors by many commenters. The Background section is mostly speculation and anecdotes, feel free to skip to The List once you understand its reason for existence. tldr: This is a curated list of observable details about a person's appearance that indicate something useful/surprising about them. Ideally, studying this list will be an efficient way to cultivate more insightful observational/abductive abilities, approaching the fictional example of Sherlock Holmes. Please contribute in the comments section after reading the Rules. Background Is it possible to develop observational abilities comparable to Sherlock Holmes? Sir Arthur Conan Doyle's fictional detective has many enviable skills, including mastery of disguise and some expertise at unarmed combat, as well as generally being a genius, but we will focus primarily on his more well known observational power. Though Holmes is often described as a master of logical "deduction," this power is better described as (possibly superhuman) abduction. That is, Holmes perceives tiny details that many people would miss, then constructs explanations for those details. By reasoning through the interacting implications of these explanations, he is able to make inferences that seem impossible to those around him. The final step is actually deductive, but the first two are perhaps more interesting. Holmes' ability to perceive more than others does seem somewhat realistic; it is always possible to actively improve one's situational awareness, at least on a short term basis, simply by focusing on one's surroundings. The trick seems to be the second step, where Holmes is able to work backwards from cause to effect, often leveraging slightly obscure knowledge about a wide variety of topics. I spent several of my naive teenage years trying to become more like Holmes. I carefully examined people's shoes (often I actually requested that the shoes be handed over) for numerous features: mud and dirt from walking outside, the apparent price of the shoe, the level of wear and tear, and more specifically the distribution of wear between heel and toe (hoping to distinguish sprinters and joggers), etc. I "read palms," studying the subtle variations between biking and weightlifting calluses. I looked for ink stains and such on sleeves (this works better in fiction than reality). I'm pretty sure I even smelled people. None of this worked particularly well. I did come up with some impressive seeming "deductions," but I made so many mistakes that these may have been entirely chance. There were various obstacles. First, it is time consuming and slightly awkward to stare at everyone you meet from head to toe. I think there are real tradeoffs here; you have only so much total attention, and by spending more on observing your surroundings, you have less left over to think. Certainly it is not possible to read a textbook at the same time, so practicing your observational techniques comes at a cost. Perhaps it becomes more habitual and easier over time, but I am not convinced it ever comes for free. Second, the reliability of inferences decays quickly with the number of steps involved. Many of Holmes' most impressive "deductions" come from combining his projected explanations for several details into one cohesive story (perhaps using some of them to rule out alternative explanations for the others) and drawing highly non-obvious, shocking conclusions from this story. In practice, one of the explanations is usually wrong, the entire story is base on false premises, and the conclusions are only sh...

Link to original articleWelcome 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: AI #72: Denying the Future, published by Zvi on July 12, 2024 on LessWrong. The Future. It is coming. A surprising number of economists deny this when it comes to AI. Not only do they deny the future that lies in the future. They also deny the future that is here, but which is unevenly distributed. Their predictions and projections do not factor in even what the AI can already do, let alone what it will learn to do later on. Another likely future event is the repeal of the Biden Executive Order. That repeal is part of the Republican platform, and Trump is the favorite to win the election. We must act on the assumption that the order likely will be repealed, with no expectation of similar principles being enshrined in federal law. Then there are the other core problems we will have to solve, and other less core problems such as what to do about AI companions. They make people feel less lonely over a week, but what do they do over a lifetime? Also I don't have that much to say about it now, but it is worth noting that this week it was revealed Apple was going to get an observer board seat at OpenAI… and then both Apple and Microsoft gave up their observer seats. Presumably that is about antitrust and worrying the seats would be a bad look. There could also be more to it. Table of Contents 1. Introduction. 2. Table of Contents. 3. Language Models Offer Mundane Utility. Long as you avoid GPT-3.5. 4. Language Models Don't Offer Mundane Utility. Many mistakes will not be caught. 5. You're a Nudge. You say it's for my own good. 6. Fun With Image Generation. Universal control net for SDXL. 7. Deepfaketown and Botpocalypse Soon. Owner of a lonely bot. 8. They Took Our Jobs. Restaurants. 9. Get Involved. But not in that way. 10. Introducing. Anthropic ships several new features. 11. In Other AI News. Microsoft and Apple give up OpenAI board observer seats. 12. Quiet Speculations. As other papers learned, to keep pace, you must move fast. 13. The AI Denialist Economists. Why doubt only the future? Doubt the present too. 14. The Quest for Sane Regulation. EU and FTC decide that things are their business. 15. Trump Would Repeal the Biden Executive Order on AI. We can't rely on it. 16. Ordinary Americans Are Worried About AI. Every poll says the same thing. 17. The Week in Audio. Carl Shulman on 80,000 hours was a two parter. 18. The Wikipedia War. One obsessed man can do quite a lot of damage. 19. Rhetorical Innovation. Yoshua Bengio gives a strong effort. 20. Evaluations Must Mimic Relevant Conditions. Too often they don't. 21. Aligning a Smarter Than Human Intelligence is Difficult. Stealth fine tuning. 22. The Problem. If we want to survive, it must be solved. 23. Oh Anthropic. Non Disparagement agreements should not be covered by NDAs. 24. Other People Are Not As Worried About AI Killing Everyone. Don't feel the AGI. Language Models Offer Mundane Utility Yes, they are highly useful for coding. It turns out that if you use GPT-3.5 for your 'can ChatGPT code well enough' paper, your results are not going to be relevant. Gallabytes says 'that's morally fraud imho' and that seems at least reasonable. Tests failing in GPT-3.5 is the AI equivalent of "IN MICE" except for IQ tests. If you are going to analyze the state of AI, you need to keep an eye out for basic errors and always always check which model is used. So if you go quoting statements such as: Paper about GPT-3.5: its ability to generate functional code for 'hard' problems dropped from 40% to 0.66% after this time as well. 'A reasonable hypothesis for why ChatGPT can do better with algorithm problems before 2021 is that these problems are frequently seen in the training dataset Then even if you hadn't realized or checked before (which you really should have), you need to notice that this says 2021, which is very much not ...

Link to original articleWelcome 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: Transformer Circuit Faithfulness Metrics Are Not Robust, published by Joseph Miller on July 12, 2024 on LessWrong. When you think you've found a circuit in a language model, how do you know if it does what you think it does? Typically, you ablate / resample the activations of the model in order to isolate the circuit. Then you measure if the model can still perform the task you're investigating. We identify six ways in which ablation experiments often vary.[1][2] How do these variations change the results of experiments that measure circuit faithfulness? TL;DR We study three different circuits from the literature and find that measurements of their faithfulness are highly dependent on details of the experimental methodology. The IOI and Docstring circuits in particular are much less faithful than reported when tested with a more precise methodology. The correct circuit for a set of prompts is undefined. The type of ablation you use to isolate the circuit determines the task that you are asking the circuit to perform - and therefore also the optimal circuit. This is especially important because previous work in automatic circuit discovery has tested algorithms by their ability to recover these "ground-truth" circuits from the literature - without considering these potential pitfalls and nuances. Case Studies We look at three circuits from the mech interp literature to demonstrate that faithfulness metrics are highly sensitive to the details of experimental setup. Indirect Object Identification Circuit The IOI circuit is the most well known circuit in a language model. It computes completions to prompts of the form: "When Mary and John went to the store, John gave a bottle of milk to ____" The circuit is specified as a graph of important attention heads (nodes) and the interactions between them (edges) as applied to a specific sequence of tokens. The authors report that the circuit explains 87% of the logit difference between the two name tokens. They find this number by passing some inputs to the model and ablating all activations outside of the circuit. Then they measure how much of the logit difference between the correct and incorrect name logits remains. However, an important detail is that they arrived at this number by ablating the nodes (heads) outside of the circuit, not by ablating the edges (interactions between heads) outside of the circuit. So they don't ablate, for example, the edges from the previous token heads to the name mover heads, even though these are not part of the circuit (effectively including more edges in the circuit). We calculate the logit difference recovered (defined below) when we ablate the edges outside of the circuit instead. They ablate the heads by replacing their activations with the mean value calculated over the "ABC distribution", in which the names in the prompts are replaced by random names.[3] In our experiments, we also try resampling the activations from different prompts (taking individual prompt activations instead of averaging). The first thing that jumps out from the box plots above is the very large range of results from different prompts. The charts here are cut off and some points are over 10,000%. This means that although the average logit difference recovered is reasonable, few prompts actually have a logit difference recovered close to 100%. And we see that ablating the edges instead of the nodes gives a much higher average logit difference recovered - close to 150% (which means that the isolated circuit has a greater logit difference between the correct and incorrect names than the un-ablated model). So the edge-based circuit they specified it is much less faithful than the node-based circuit they tested. The authors calculate the 87% result as the ratio of the expected difference (over a set of prompts) in the ...

Link to original articleWelcome 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: Superbabies: Putting The Pieces Together, published by sarahconstantin on July 12, 2024 on LessWrong. This post was inspired by some talks at the recent LessOnline conference including one by LessWrong user "Gene Smith". Let's say you want to have a "designer baby". Genetically extraordinary in some way - super athletic, super beautiful, whatever. 6'5", blue eyes, with a trust fund. Ethics aside[1], what would be necessary to actually do this? Fundamentally, any kind of "superbaby" or "designer baby" project depends on two steps: 1.) figure out what genes you ideally want; 2.) create an embryo with those genes. It's already standard to do a very simple version of this two-step process. In the typical course of in-vitro fertilization (IVF), embryos are usually screened for chromosomal abnormalities that would cause disabilities like Down Syndrome, and only the "healthy" embryos are implanted. But most (partially) heritable traits and disease risks are not as easy to predict. Polygenic Scores If what you care about is something like "low cancer risk" or "exceptional athletic ability", it won't be down to a single chromosomal abnormality or a variant in a single gene. Instead, there's typically a statistical relationship where many genes are separately associated with increased or decreased expected value for the trait. This statistical relationship can be written as a polygenic score - given an individual's genome, it'll crunch the numbers and spit out an expected score. That could be a disease risk probability, or it could be an expected value for a trait like "height" or "neuroticism." Polygenic scores are never perfect - some people will be taller than the score's prediction, some shorter - but for a lot of traits they're undeniably meaningful, i.e. there will be a much greater-than-chance correlation between the polygenic score and the true trait measurement. Where do polygenic scores come from? Typically, from genome-wide association studies, or GWAS. These collect a lot of people's genomes (the largest ones can have hundreds of thousands of subjects) and personal data potentially including disease diagnoses, height and weight, psychometric test results, etc. And then they basically run multivariate correlations. A polygenic score is a (usually regularized) multivariate regression best-fit model of the trait as a function of the genome. A polygenic score can give you a rank ordering of genomes, from "best" to "worst" predicted score; it can also give you a "wish list" of gene variants predicted to give a very high combined score. Ideally, "use a polygenic score to pick or generate very high-scoring embryos" would result in babies that have the desired traits to an extraordinary degree. In reality, this depends on how "good" the polygenic scores are - to what extent they're based on causal vs. confounded effects, how much of observed variance they explain, and so on. Reasonable experts seem to disagree on this.[2] I'm a little out of my depth when it comes to assessing the statistical methodology of GWAS studies, so I'm interested in another question - even assuming you have polygenic score you trust, what do you do next? How do you get a high-scoring baby out of it? Massively Multiplexed, Body-Wide Gene Editing? Not So Much, Yet. "Get an IVF embryo and gene-edit it to have the desired genes" (again, ethics and legality aside)[3] is a lot harder than it sounds. First of all, we don't currently know how to make gene edits simultaneously and abundantly in every tissue of the body. Recently approved gene-editing therapies like Casgevy, which treats sickle-cell disease, are operating on easy mode. Sickle-cell disease is a blood disorder; the patient doesn't have enough healthy blood cells, so the therapy consists of an injection of the patient's own blood cells which h...

Link to original articleWelcome 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: Yoshua Bengio: Reasoning through arguments against taking AI safety seriously, published by Judd Rosenblatt on July 12, 2024 on LessWrong. He starts by emphasizing The issue is so hotly debated because the stakes are major: According to some estimates, quadrillions of dollars of net present value are up for grabs, not to mention political power great enough to significantly disrupt the current world order. [...] The most important thing to realize, through all the noise of discussions and debates, is a very simple and indisputable fact: while we are racing towards AGI or even ASI, nobody currently knows how such an AGI or ASI could be made to behave morally, or at least behave as intended by its developers and not turn against humans. And goes on to do a pretty great job addressing "those who think AGI and ASI are impossible or are centuries in the future AGI is possible but only in many decades that we may reach AGI but not ASI, those who think that AGI and ASI will be kind to us that corporations will only design well-behaving AIs and existing laws are sufficient who think that we should accelerate AI capabilities research and not delay benefits of AGI talking about catastrophic risks will hurt efforts to mitigate short-term human-rights issues with AI concerned with the US-China cold war that international treaties will not work the genie is out of the bottle and we should just let go and avoid regulation that open-source AGI code and weights are the solution worrying about AGI is falling for Pascal's wager" Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

Link to original articleWelcome 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: Decomposing Agency - capabilities without desires, published by owencb on July 11, 2024 on LessWrong. What is an agent? It's a slippery concept with no commonly accepted formal definition, but informally the concept seems to be useful. One angle on it is Dennet's Intentional Stance: we think of an entity as being an agent if we can more easily predict it by treating it as having some beliefs and desires which guide its actions. Examples include cats and countries, but the central case is humans. The world is shaped significantly by the choices agents make. What might agents look like in a world with advanced - and even superintelligent - AI? A natural approach for reasoning about this is to draw analogies from our central example. Picture what a really smart human might be like, and then try to figure out how it would be different if it were an AI. But this approach risks baking in subtle assumptions - things that are true of humans, but need not remain true of future agents. One such assumption that is often implicitly made is that "AI agents" is a natural class, and that future AI agents will be unitary - that is, the agents will be practically indivisible entities, like single models. (Humans are unitary in this sense, and while countries are not unitary, their most important components - people - are themselves unitary agents.) This assumption seems unwarranted. While people certainly could build unitary AI agents, and there may be some advantages to doing so, unitary agents are just an important special case among a large space of possibilities for: Components which contain important aspects of agency (without necessarily themselves being agents); Ways to construct agents out of separable subcomponents (none, some, or all of which may be reasonably regarded agents in their own right). We'll begin an exploration of these spaces. We'll consider four features we generally expect agents to have[1]: Goals Things they are trying to achieve e.g. I would like a cup of tea Implementation capacity The ability to act in the world e.g. I have hands and legs Situational awareness Understanding of the world (relevant to the goals) e.g. I know where I am, where the kettle is, and what it takes to make tea Planning capacity The ability to choose actions to effectively further their goals, given their available action set and their understanding of the situation e.g. I'll go downstairs and put the kettle on We don't necessarily expect to be able to point to these things separately - especially in unitary agents they could exist in some intertwined mess. But we kind of think that in some form they have to be present, or the system couldn't be an effective agent. And although these features are not necessarily separable, they are potentially separable - in the sense that there exist possible agents where they are kept cleanly apart. We will explore possible decompositions of agents into pieces which contain different permutations of these features, connected by some kind of scaffolding. We will see several examples where people naturally construct agentic systems in ways where these features are provided by separate components. And we will argue that AI could enable even fuller decomposition. We think it's pretty likely that by default advanced AI will be used to create all kinds of systems across this space. (But people could make deliberate choices to avoid some parts of the space, so "by default" is doing some work here.) A particularly salient division is that there is a coherent sense in which some systems could provide useful plans towards a user's goals, without in any meaningful sense having goals of their own (or conversely, have goals without any meaningful ability to create plans to pursue those goals). In thinking about ensuring the safety of advanced AI systems, it ma...

Link to original articleWelcome 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: [EAForum xpost] A breakdown of OpenAI's revenue, published by dschwarz on July 10, 2024 on LessWrong. We estimate that, as of June 12, 2024, OpenAI has an annualized revenue (ARR) of: $1.9B for ChatGPT Plus (7.7M global subscribers), $714M from ChatGPT Enterprise (1.2M seats), $510M from the API, and $290M from ChatGPT Team (from 980k seats) (Full report in https://app.futuresearch.ai/reports/3Li1, methods described in https://futuresearch.ai/openai-revenue-report.) We looked into OpenAI's revenue because financial information should be a strong indicator of the business decisions they make in the coming months, and hence an indicator of their research priorities. Our methods in brief: we searched exhaustively for all public information on OpenAI's finances, and filtered it to reliable data points. From this, we selected a method of calculation that required the minimal amount of inference of missing information. To infer the missing information, we used the standard techniques of forecasters: fermi estimates, and base rates / analogies. We're fairly confident that the true values are relatively close to what we report. We're still working on methods to assign confidence intervals on the final answers given the confidence intervals of all of the intermediate variables. Inside the full report, you can see which of our estimates are most speculative, e.g. using the ratio of Enterprise seats to Teams seats from comparable apps; or inferring the US to non-US subscriber base across platforms from numbers about mobile subscribers, or inferring growth rates from just a few data points. Overall, these numbers imply to us that: Sam Altman's surprising claim of $3.4B ARR on June 12 seems quite plausible, despite skepticism people raised at the time. Apps (consumer and enterprise) are much more important to OpenAI than the API. Consumers are much more important to OpenAI than enterprises, as reflected in all their recent demos, but the enterprise growth rate is so high that this may change abruptly. Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

Link to original articleWelcome 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: Fluent, Cruxy Predictions, published by Raemon on July 10, 2024 on LessWrong. The latest in the Feedback Loop Rationality series. Periodically, people (including me) try to operationalize predictions, or bets, and... it doesn't seem to help much. I think I recently "got good" at making "actually useful predictions." I currently feel on-the-cusp of unlocking a host of related skills further down the rationality tech tree. This post will attempt to spell out some of the nuances of how I currently go about it, and paint a picture of why I think it's worth investing in. The takeaway that feels most important to me is: it's way better to be "fluent" at operationalizing predictions, compared to "capable at all." Previously, "making predictions" was something I did separately from my planning process. It was a slow, clunky process. Nowadays, reasonably often I can integrate predictions into the planning process itself, because it feels lightweight. I'm better at quickly feeling-around for "what sort of predictions would actually change my plans if they turned out a certain way?", and then quickly check in on my intuitive sense of "what do I expect to happen?" Fluency means you can actually use it day-to-day to help with whatever work is most important to you. Day-to-day usage means you can actually get calibrated for predictions in whatever domains you care about. Calibration means that your intuitions will be good, and you'll know they're good. If I were to summarize the change-in-how-I-predict, it's a shift from: "Observables-first". i.e. looking for things I could easily observe/operationalize, that were somehow related to what I cared about. to: "Cruxy-first". i.e. Look for things that would change my decisionmaking, even if vague, and learn to either better operationalize those vague things, or, find a way to get better data. (and then, there's a cluster of skills and shortcuts to make that easier) Disclaimer: This post is on the awkward edge of "feels full of promise", but "I haven't yet executed on the stuff that'd make it clearly demonstrably valuable." (And, I've tracked the results of enough "feels promise" stuff to know they have a <50% hit rate) I feel like I can see the work needed to make this technique actually functional. It's a lot of work. I'm not sure if it's worth it. (Alas, I'm inventing this technique because I don't know how to tell if my projects are worth it and I really want a principled way of forming beliefs about that. Since I haven't finished vetting it yet it's hard to tell!) There's a failure mode where rationality schools proliferate without evidence, where people get excited about their new technique that sounds good on paper, and publish exciting blogposts about it. There's an unfortunate catch-22 where: Naive rationality gurus post excitedly, immediately, as if they already know what they're doing. This is even kinda helpful (because believing in the thing really does help make it more true). More humble and realistic rationality gurus wait to publish until they're confident their thing really works. They feel so in-the-dark, fumbling around. As they should. Because, like, they (we) are. But, I nonetheless feel like the rationality community was overall healthier when it was full of excited people who still believed a bit in their heart that rationality would give them superpowers, and posted excitedly about it. It also seems intellectually valuable for notes to be posted along the way, so others can track it and experiment. Rationality Training doesn't give you superpowers - it's a lot of work, and then "it's a pretty useful skill, among other skills." I expect, a year from now, I'll have a better conception of the ideas in this post. This post is still my best guess about how this skill can work, and I'd like it if other people excitedl...

Link to original articleWelcome 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: Reliable Sources: The Story of David Gerard, published by TracingWoodgrains on July 10, 2024 on LessWrong. This is a linkpost for https://www.tracingwoodgrains.com/p/reliable-sources-how-wikipedia-admin, posted in full here given its relevance to this community. Gerard has been one of the longest-standing malicious critics of the rationalist and EA communities and has done remarkable amounts of work to shape their public images behind the scenes. Note: I am closer to this story than to many of my others. As always, I write aiming to provide a thorough and honest picture, but this should be read as the view of a close onlooker who has known about much within this story for years and has strong opinions about the matter, not a disinterested observer coming across something foreign and new. If you're curious about the backstory, I encourage you to read my companion article after this one. Introduction: Reliable Sources Wikipedia administrator David Gerard cares a great deal about Reliable Sources. For the past half-decade, he has torn through the website with dozens of daily edits - upwards of fifty thousand, all told - aimed at slashing and burning lines on the site that reference sources deemed unreliable by Wikipedia. He has stepped into dozens of official discussions determining which sources the site should allow people to use, opining on which are Reliable and which are not. He cares so much about Reliable Sources, in fact, that he goes out of his way to provide interviews to journalists who may write about topics he's passionate about, then returns to the site to ensure someone adds just the right quotes from those sources to Wikipedia articles about those topics and to protect those additions from all who might question them. While by Wikipedia's nature, nobody can precisely claim to speak or act on behalf of the site as a whole, Gerard comes about as close as anyone really could. He's been a volunteer Wikipedia administrator since 2004, has edited the site more than 200,000 times, and even served off and on as the site's UK spokesman. Few people have had more of a hand than him in shaping the site, and few have a more encyclopedic understanding of its rules, written and unwritten. Reliable sources, a ban on original research, and an aspiration towards a neutral point of view have long been at the heart of Wikipedia's approach. Have an argument, editors say? Back it up with a citation. Articles should cover "all majority and significant minority views" from Reliable Sources (WP:RS) on the topic "fairly, proportionately, and, as far as possible, without editorial bias" (WP:NPOV). The site has a color-coding system for frequently discussed sources: green for reliable, yellow for unclear, red for unreliable, and dark red for "deprecated" sources that can only be used in exceptional situations. The minutiae of Wikipedia administration, as with the inner workings of any bureaucracy, is an inherently dry subject. On the site as a whole, users sometimes edit pages directly with terse comments, other times engage in elaborate arguments on "Talk" pages to settle disputes about what should be added. Each edit is added to a permanent history page. To understand any given decision, onlookers must trawl through page after page of archives and discussions replete with tidily packaged references to one policy or another. Where most see boredom behind the scenes and are simply glad for mostly functional overviews of topics they know nothing about, though, a few see opportunity. Those who master the bureaucracy in behind-the-scenes janitorial battles, after all, define the public's first impressions of whatever they care about. Since 2017, when Wikipedia made the decision to ban citations to the Daily Mail due to "poor fact-checking, sensationalism, and flat-out fabrication," ed...

Link to original articleWelcome 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: New page: Integrity, published by Zach Stein-Perlman on July 10, 2024 on LessWrong. There's a new page collecting integrity incidents at the frontier AI labs. Also a month ago I made a page on labs' policy advocacy. If you have suggestions to improve these pages, or have ideas for other resources I should create, let me know. Crossposted from AI Lab Watch. Subscribe on Substack. Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org