The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - Prize idea: Transmit MIRI and Eliezer's worldviews by elifland
    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: Prize idea: Transmit MIRI and Eliezer's worldviews, published by elifland on September 19, 2022 on LessWrong.
    Motivation
    Claim (80% confidence): At least 50% of the disagreement between people who align more with MIRI/Eliezer and those who align more with opposing clusters of views (Christiano, Garfinkel, etc.) is caused not by rational disagreement, but by more subconscious emotional stuff like tastes, vibes, culture, politics, etc.
    Elicit Prediction (forecast.elicit.org/binary/questions/28xD2oJyG)
    In particular, I think Eliezer’s writing tends to appeal more to people who:
    Enjoy reading fiction, and are okay with points primarily being made via parables
    Similarly, don’t mind long pieces with no summaries and a lack of clear organizational structure of the claims being made and the evidence for them
    Aren’t turned off by perceived arrogance (I’m not taking a stance in this post on whether the arrogance level is justified or not)
    Past attempts to communicate their worldview such as the MIRI conversations have helped some, but I think mostly weren’t attacking the core issue for which my current guess is large differences in communication styles.
    For people who aren’t high on the above axes, I think Eliezer’s writing often tends to be fairly hard to read and offputting, which is unfortunate. It leads to people not taking the points as seriously as they should (and perhaps has the opposite effect on those who the style resonates with). While I disagree with MIRI/Eliezer on a lot of topics, I agree with them to some extent on a lot and think it’s very valuable to understand their worldview and build a “MIRI-model”.
    I’ve written briefly about my personal experience taking Eliezer and MIRI less seriously than I should have here. I still haven’t read most of the sequences and don’t intend to read HPMOR, but I now take MIRI and Eliezer’s worldviews much more seriously than I used to.
    Proposal
    An idea to mitigate this is to give out substantial prizes for write-ups which transmit important aspects of the MIRI/Eliezer worldview in ways which are easier for people with different tastes to digest.
    A few possible candidates for this “transmission” include (not an exhaustive list!):
    The Sequences
    Eliezer’s and Nate’s portion of the MIRI conversations
    2022 MIRI Alignment Discussion
    AGI Ruin: A List of Lethalities may be a good candidate, though while extremely important I think this one is already decently structured and argued relative to some other pieces
    I propose prizes of ~$1-10k (weighted by importance) for pieces that do a good job at this, judged by some combination of clarity to those with different tastes from MIRI/Eliezer and how well they maintain a representation of MIRI/Eliezer’s views. I’d be able to commit ~$1-5k myself, and would be happy for commitments from others.
    But won’t the writeups not do a good job conveying MIRI’s intuitions?
    If a writeup is getting popular but contains serious mistakes, MIRI/Eliezer can chime in and say so.
    Isn’t MIRI already trying to communicate their view?
    I think so, but it seems great to also have outsiders working on the case. I’d guess there are advantages and disadvantages to both strategies.
    Have I talked to MIRI about this?
    Yes. Rob suggested going for it and re-iterated the above, saying MIRI’s involvement shouldn’t be a bottleneck to investigating MIRI’s worldview.
    Next steps
    I’d love to get:
    Feedback on whether this is a good idea, and suggestions for revision if so
    Volunteers to help actually push the prize to happen and be judged, which I’m not sure I really want to do myself
    Commitments to the prize pool
    Other ideas for good pieces to transmit
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    4 min
  • LW - Losing the root for the tree by Adam Zerner
    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: Losing the root for the tree, published by Adam Zerner on September 20, 2022 on LessWrong.
    1
    You know that being healthy is important. And that there's a lot of stuff you could do to improve your health: getting enough sleep, eating well, reducing stress, and exercising, to name a few.
    There’s various things to hit on when it comes to exercising too. Strength, obviously. But explosiveness is a separate thing that you have to train for. Same with flexibility. And don’t forget cardio!
    Strength is most important though, because of course it is. And there’s various things you need to do to gain strength. It all starts with lifting, but rest matters too. And supplements. And protein. Can’t forget about protein.
    Protein is a deeper and more complicated subject than it may at first seem. Sure, the amount of protein you consume matters, but that’s not the only consideration. You also have to think about the timing. Consuming large amounts 2x a day is different than consuming smaller amounts 5x a day. And the type of protein matters too. Animal is different than plant, which is different from dairy. And then quality is of course another thing that is important.
    But quality isn’t an easy thing to figure out. The big protein supplement companies are Out To Get You. They want to mislead you. Information sources aren’t always trustworthy. You can’t just hop on The Wirecutter and do what they tell you. Research is needed.
    So you listen to a few podcasts. Follow a few YouTubers. Start reading some blogs. Throughout all of this you try various products and iterate as you learn more. You’re no Joe Rogan, but you’re starting to become pretty informed.
    2
    You’re a product manager over at Danslist, an up-and-coming competitor to Craigslist. You’ve recently been promoted and are now in charge of the most important page of the site: the product listings page.
    This page is Danslist’s bread and butter. It’s very important that it is a good web page. And there’s a lot you can do to improve it. These improvements can be grouped into aesthetics, usability, functionality and speed.
    Speed is something that the VP of Product has been talking a lot about. And you recently had a few meetings with the tech lead and some senior engineers. They had various ideas about how the speed could be improved. One is to parallelize the data fetching instead of doing it serially. Everyone agrees that this would be a good idea and a pretty obvious next step.
    But in order to parallelize the fetching, well, some things need to be reworked. Currently the team is first fetching from Service A, then using the response from Service A to fetch from Service B, and then using the response from Service B to fetch from Service C. So given how things are currently structured, the fetching can’t be done in parallel. Service C needs the response from Service B, and Service B needs the response from Service A.
    It’d take some effort, but these services can be rewritten in such a way where they don’t depend on one another and the data fetching can be done in parallel. Seems logical enough.
    But unfortunately, things don’t stop there. It isn’t currently possible to restructure those services with the Postgres database the team is currently using. They need to switch to MySQL first, because MySQL has some features that Postgres just doesn’t offer.
    Again, this all seems logical enough to you. So you announce the migration to MySQL to the team, and with the help of the tech lead and one of your highly paid superstar backend engineers, you start writing some tickets.
    Does anyone notice a problem here? Or at least, does anyone have any questions? Personally I’m thinking to myself: wait, what are we actually trying to do here?
    Well, I guess we’re trying to improve the speed of the landing page. And parallelizing the data fetching is one w...
    15 min
  • LW - Quintin's alignment papers roundup - week 2 by Quintin Pope
    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: Quintin's alignment papers roundup - week 2, published by Quintin Pope on September 19, 2022 on LessWrong.
    Introduction
    Last week's paper roundup (more or less by accident) focused mostly on path dependence of deep learning and the order of feature learning. Going forwards, I've decided to have an explicit focus for each week's roundup. This week's focus is on the structure/redundancy of trained models, as well as linear interpolations through parameter space.
    I've also decided to publish each roundup on Monday morning.
    Papers
    Residual Networks Behave Like Ensembles of Relatively Shallow Networks
    In this work we propose a novel interpretation of residual networks showing that they can be seen as a collection of many paths of differing length. Moreover, residual networks seem to enable very deep networks by leveraging only the short paths during training. To support this observation, we rewrite residual networks as an explicit collection of paths. Unlike traditional models, paths through residual networks vary in length. Further, a lesion study reveals that these paths show ensemble-like behavior in the sense that they do not strongly depend on each other. Finally, and most surprising, most paths are shorter than one might expect, and only the short paths are needed during training, as longer paths do not contribute any gradient. For example, most of the gradient in a residual network with 110 layers comes from paths that are only 10-34 layers deep.
    Our results reveal one of the key characteristics that seem to enable the training of very deep networks: Residual networks avoid the vanishing gradient problem by introducing short paths which can carry gradient throughout the extent of very deep networks.
    My opinion:
    This paper suggests that neural nets are redundant by default, which gives some intuition for why it's often possible to prune large fractions of a network's parameters without much impact on the test performance, as well as the mechanism by which residual connections allow for training deeper networks: residual connections allow shallow nets to communicate directly with the input / output space, so they allow for deep nets to be built from ensembling shallow nets.
    I think it also points away from neural nets implementing a Kolmogorov or circuit simplicity prior.
    On the Effect of Dropping Layers of Pre-trained Transformer Models
    Transformer-based NLP models are trained using hundreds of millions or even billions of parameters, limiting their applicability in computationally constrained environments. While the number of parameters generally correlates with performance, it is not clear whether the entire network is required for a downstream task. Motivated by the recent work on pruning and distilling pre-trained models, we explore strategies to drop layers in pre-trained models, and observe the effect of pruning on downstream GLUE tasks. We were able to prune BERT, RoBERTa and XLNet models up to 40%, while maintaining up to 98% of their original performance. Additionally we show that our pruned models are on par with those built using knowledge distillation, both in terms of size and performance.
    Our experiments yield interesting observations such as, (i) the lower layers are most critical to maintain downstream task performance, (ii) some tasks such as paraphrase detection and sentence similarity are more robust to the dropping of layers, and (iii) models trained using a different objective function exhibit different learning patterns and w.r.t the layer dropping.
    My opinion:
    (see below)
    Of Non-Linearity and Commutativity in BERT
    In this work we provide new insights into the transformer architecture, and in particular, its best-known variant, BERT. First, we propose a method to measure the degree of non-linearity of different elements of transformers. Next, w...
    19 min
  • LW - Gene drives: why the wait? by Metacelsus
    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: Gene drives: why the wait?, published by Metacelsus on September 19, 2022 on LessWrong.
    (Crossposted from my Substack)
    If you’ve been following biology news over the last few years, you might have heard of an interesting concept called a “gene drive”. The overall idea is to engineer a genetic allele that transmits itself to all offspring of a sexually reproducing organism, instead of being inherited by 50% as usual. This allele can also perform some other biological function (a relevant example is causing female sterility).
    A gene drive spreads through a population. From Esvelt et al. 2014 (CC-BY)
    In multiple trials, modern CRISPR-based gene drives have shown high efficacy in spreading through populations of caged Anopheles mosquitoes and completely suppressing their reproduction. Since Anopheles mosquitoes are the only ones that transmit malaria, causing their extinction would directly save hundreds of thousands of lives per year. Similar gene drives targeted to other types of mosquitoes (Aedes, Culex, etc.) could also eliminate diseases such as dengue fever, Zika virus, and West Nile virus.
    However, in spite of promising laboratory trials, gene drives have not yet been deployed in the wild. But why not?
    History of gene drives
    Although the technology to build effective gene drives did not exist until recently, the idea has been around for a while. In fact, gene drives occur naturally. Some well-known examples are transposons in flies, homing endonucleases in algae, and segregation distorters in mice.
    The idea of engineering a site-specific nuclease as a gene drive was developed as early as 2003, and in the decade that followed there were several efforts to develop these, with the labs of Austin Burt and Andrea Crisanti taking a lead role. These early systems showed some biased inheritance, but were not stable for more than a few generations.
    The advent of CRISPR as a gene editing system opened up a new opportunity. A paper in 2014 by Kevin Esvelt and co-workers proposed Cas9 as a nuclease for a gene drive, with several properties making it ideal for the task.
    It lacks repetitive sequences that caused problems with earlier gene drives using zinc-finger nucleases or TALENs.
    It has a very high efficiency of cutting.
    It is easy to target a new site by simply changing the guide RNA.
    Several nearby sites could be targeted at once, using different guide RNAs.
    From Esvelt et al. 2014 (CC-BY)
    CRISPR-based gene drives quickly gained popularity in the field, and by 2018 the Crisanti lab had demonstrated a working gene drive that could efficiently suppress populations of Anopheles gambiae by targeting an exon of the doublesex gene required for female development. At the time this was announced, I was studying at the University of Cambridge, and attended a public lecture by Prof. Crisanti about his lab’s work. The overall mood in the room was almost euphoric: here was a technology that could save millions of lives, the best thing since Borlaug’s wheat!
    Since that lecture, about 2 million people, mostly children in Africa, have died of malaria. Gene drive research has not stood still: the Crisanti lab tested their doublesex drive in larger cages of mosquitoes, and it again completely eliminated the populations. But given the millions of lives at stake, what’s taking so long for this gene drive to be released?
    See also: the battle against malaria in Africa has stalled
    Why the wait?
    There are two good arguments against the immediate release of gene drives to eliminate mosquitoes.
    First, nuclease gene drives have the possibility of generating resistant alleles, making future gene drives not work against the same target. Therefore, it’s important to get it right the first time, otherwise the potential of gene drives could be wasted. The goal of the large cage trials I mentioned earli...
    6 min
  • LW - Some notes on solving hard problems by Joe Rocca
    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: Some notes on solving hard problems, published by Joe Rocca on September 19, 2022 on LessWrong.
    Caveats/Notes:
    This is a cleaned-up version of some personal notes that I've had sitting in Google Docs for a while. There's a lot of hand-waving, probably some over-generalising, and many points could do with some/more examples. Please take each suggestion with a large grain of salt, and feel free to add your thoughts via the comments.
    I'm not sure how far these thoughts will transfer outside of my particular problem domain, and problem-solving style. I tend to solve problems by building up visual intuitions/models/algebras, rather than leaning heavily on symbols/equations.
    There are some overlaps between various suggestions below - they shouldn't be read as completely orthogonal points. Some ideas are special cases of other ideas, or are basically the same idea from a different perspective.
    If you're also interested in the "meta" of problem solving and want to chat about this, please feel free to reach out to me here on LW or on Twitter at @rocca27 - would be great to meet others who are interested in this!
    Develop multiple perspectives on the problem
    You should work hard to "see" your problem and potential solutions from multiple perspectives. By "perspective" I mean a high-level mental model that you use to frame your thinking about the problem. A perspective might center around an analogy, or a particular data structure, or a line of reasoning rooted in some low-level principles.
    Once you've got multiple perspectives, you can play the perspectives against one another to get a more fundamental understanding of the problem. One perspective will give you insights which unblock your progress on another perspective.
    Each perspective is just a mental tool to help you think about the problem - it is specialised for understanding certain types of behavior/constraints that are difficult to understand with the other perspectives.
    Good perspectives make important aspects and constraints of your problem "obvious". They make it easy to quickly reason about your problem, and neatly encapsulate constraints so that it requires little effort to keep them all in your head. If you find yourself having to "strain" your brain to pull implications out of a mental model, then it may be that you're using the wrong one for that particular task.
    A quote from Claude Shannon that I found on this topic:
    [...] try to restate it in just as many different forms as you can. Change the words. Change the viewpoint. Look at it from every possible angle. After you’ve done that, you can try to look at it from several angles at the same time and perhaps you can get an insight into the real basic issues of the problem, so that you can correlate the important factors and come out with the solution. It’s difficult really to do this, but it is important that you do. If you don’t, it is very easy to get into ruts of mental thinking. You start with a problem here and you go around a circle here and if you could only get over to this point, perhaps you would see your way clear; but you can’t break loose from certain mental blocks which are holding you in certain ways of looking at a problem. That is the reason why very frequently someone who is quite green to a problem will sometimes come in and look at it and find the solution like that, while you have been laboring for months over it.
    You’ve got set into some ruts here of mental thinking and someone else comes in and sees it from a fresh viewpoint.
    Start the jigsaw puzzle with the corner pieces
    The corner pieces of a jigsaw puzzle are the easiest to place. Lock in the "easy certainties" first, and then build more "solid ground" outwards from them - work in the heavily-constrained areas. Speculating on the super "fuzzy" areas (e.g. the center of the jigsaw puzzl...
    47 min
  • EA - Safety timelines: How long will it take to solve alignment? by Esben Kran
    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: Safety timelines: How long will it take to solve alignment?, published by Esben Kran on September 19, 2022 on The Effective Altruism Forum.
    TLDR; no one talks about how far we are towards safe AGI and the focus is on when AGI doom arrives. We want to readjust this focus and measure progress, guide and facilitate research, and evaluate projects in AI safety for impact. We also ask you to add your views to this survey.
    Prelude
    It was a hot London Summer day and fourteen people had gathered to discuss the future of AI safety. To ascertain the mission criticality of our endeavor, we drew everyone’s AGI timelines on a whiteboard. The product of our estimates was close to the Metaculus estimate and some expressed concern at the stress of impending doom.
    We then proceeded to draw our expected timelines for when alignment would be solved and very few had ever analyzed this. The few that expressed their timelines ended at 15 years hence, in 2037. And I thought this was a remarkably hopeful message!
    A case for hope
    Let us be optimistic and expect the median arrival for the solution to alignment is 2037 estimated as a Gaussian with a standard deviation of 10 years. If we estimate the probability that the solution will come before AGI based on sampling the probability mass of the Metaculus’ forecast for AGI and our “Safety Timeline” (or “alignment solution timeline”), we can calculate the probability that alignment will be solved before an AGI is released:
    P(solution=75/85/95%
    > Will a panel of 15 recognized experts in AI safety expect a pro...
    13 min
  • LW - Biden should be applauded for appointing Renee Wegrzyn for ARPA-H by ChristianKl
    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: Biden should be applauded for appointing Renee Wegrzyn for ARPA-H, published by ChristianKl on September 18, 2022 on LessWrong.
    When Biden first announced ARPA-H (Advanced Research Projects Agency for Health) I was skeptical. At the time the proposal was to make ARPA-H a normal agency within the NIH. When critics argued that the NIH bureaucracy would likely reduce the amount of innovation that ARPA-H could produce the administration listened and US health secretary Xavier Becerra decided to make ARPA-H an independent entity within the NIH.
    Appointing Renee Wegrzyn as the head of ARPA-H is an excellent choice. Having previously worked at DARPA and IARPA and not in the NIH makes her a good person to copy the structure of how DARPA and IARPA get things done.
    Being 45, she is twenty years younger than the average director of an institute at the NIH who is 65. When science sometimes evolves from gravestone to gravestone, it’s great to see a director for the ARPA-H who’s in her forties. Especially, in contrast to the newly appointed NIH director Lawrence A. Tabak who is a 71-year-old (a fact that his NIH biography omits) having a 45-year-old lead ARPA-H is a welcome surprise.
    While at DARPA, Wegrzyn worked on biosecurity and even gave a talk about the importance of biosecurity to the Long Now Foundation. It’s valuable to have people who care about biosecurity at the top of organizations that direct a lot of research dollars.
    A lot of the research focuses either on directly curing diseases or on basic research but the research that focuses on producing tools for better research is historically underfunded. Wegrzyn experience as vice president of business development at Gingko Bioworks which is a unicorn biotech company that historically doesn’t focus on curing diseases directly but providing tools to help other companies, gives me hope that ARPA-H will be able to fund research into producing better tools to deal with biological problems. During COVID-19 they worked on producing COVID tests which was likely a lesson in the flaws of government bureaucracy holding back innovation.
    In an interview she described her own research as being guided by the question of ‘what do we know about protein folding that can help us develop a totally new type of diagnostic?’ I’m happy to see that and think it will lead to better research paths than having someone who focuses more on creating FDA approved drugs in charge of allocating research funding.
    She coedited Alzheimer’s Disease Targets for New Clinical Diagnostic and Therapeutic Strategies in 2012. It seems to me that one of the more important issues in Alzheimers research funding is the overinvestment in the amyloid cascade hypothesis. The book gets that issue right when it argues “Until recently, the amyloid cascade hypothesis had been the predominant working hypothesis for AD pathogenesis. However, this hypothesis has been supplanted by what some have called the oligomer cascade hypothesis, which posits that Aβ oligomers, rather than fibrils, are the proximate neurotoxic agents in AD” I think she showed excellent judgment by writing a book that argued for directing our Alzheimer research dollar to other paths than the amyloid cascade hypothesis (that part isn’t written by her directly).
    In the essay that she wrote herself, she argues “Drug discovery and development for the treatment of Alzheimer’s disease would greatly benefit from biomarkers that are predictive of clinical outcomes to inform the selection of effective drug candidates, monitor dose safety and efficacy on treated cohorts, and identify or confirm the mechanisms of drug action in hopes of developing treatments that go beyond symptomatic relief and are disease modifying. The ability to monitor disease progression longitudinally during patient treatment will also be key to...
    5 min
  • LW - Do bamboos set themselves on fire? by Malmesbury
    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: Do bamboos set themselves on fire?, published by Malmesbury on September 19, 2022 on LessWrong.
    Cross-posted from Telescopic Turnip.
    As we all know, the best place to have a kung-fu fight is a bamboo forest. There are just so many opportunities to grab pieces of bamboos and manufacture improvised weapons, use them to catapult yourself in the air and other basic techniques any debutant martial artist ought to know. A lesser-known fact is that bamboo-forest fights occur even when the cameras of Hong-Kong filmmakers are not present. They may even happen without the presence of humans at all. The forest itself is the kung-fu fight.
    It's often argued that humans are the worst species on Earth, because of our limitless potential for violence and mutual destruction. If that's the case, bamboos are second. Bamboos are sick. The evolution of bamboos is the result of multiple layers of shear brutality, with two imbricated levels of war, culminating in an apocalypse of combustive annihilation. At least, according to some hypotheses.
    Bamboo wars: bamboo-made famine
    Recommended soundtrack for this part
    If you enter a bamboo forest and wait for a long time, you may have the chance to witness their strange mating ritual. And by "a long time", I mean you'll probably have to wait for decades before you see anything happen at all. But eventually, all the bamboos in the forest will start to produce an absolutely massive amount of seeds, all at the same time.
    And then they all die. The swarm of seeds will give birth to a new generation of bamboos, and it will take another few decades before you see any flower again. Based on historical texts relating the event, we can estimate that some species like Phyllostachis bambusoides flower only once every 120 years:
    That's really puzzling. Wouldn't a species with a shorter cycle, hence a higher growth rate, take over the forest exponentially fast? Why would bamboos wait for so long before blooming?
    The extension of flowering cycles to ridiculous lengths probably happened in two stages. The first stage is about bamboos fighting their predators, the second is about bamboos fighting each other. Then, there is a speculative bonus third stage about bamboos destroying every form of life that has the misfortune to be nearby.
    Imagine you're a bamboo. You just released a batch a freshly-baked seeds, only to discover that all the animals in the forest are eating them. You could respond by making more seeds, but then the population of predators would just become bigger and still eat everything. That's when it becomes useful to increase the gap between two flowering events. Instead of producing n seeds every year, you produce 3n seeds every three years. This way, all the predators with a lifespan below 3 years will starve between two flowerings, so their population remains low. When the day comes to release your seeds, there are so many of them that the small predator population can only eat a little part of it, until they are simply not hungry any more. And then, they should leave enough seeds to start a new generation of bamboos.
    The predators might reproduce like crazy after the feast, but it's ok: your descendant will take so long to flower that most of their descendants will have died of starvation by then.
    (Now you can stop imagining you're a bamboo.)
    This destructive but effective phenomenon is known as "predator satiation". And when we say satiation, we are not kidding. Here we love beautiful stories of cute little animals, so here are some from the archives: "The seed grew in clusters and resembled oats, and all the animals and fowls got rolling fat from eating this seed" (Hughes 1951), "I have known of village cattle gorging themselves on the fruit to such an extent as to die subsequently from the effects of overeating" (Thom 1935) and so on.
    Alr...
    11 min
  • EA - Announcing the NYU Mind, Ethics, and Policy Program by Sofia Fogel
    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: Announcing the NYU Mind, Ethics, and Policy Program, published by Sofia Fogel on September 19, 2022 on The Effective Altruism Forum.
    We are thrilled to share that the NYU Mind, Ethics, and Policy (MEP) Program has now launched!
    MEP will conduct and support foundational research about the nature and intrinsic value of nonhuman minds, including biological and artificial minds. In particular, this program aims to advance understanding of the consciousness, sentience, sapience, moral status, legal status, and political status of nonhumans – with special focus on invertebrates and AIs – via research and outreach. The team will include Jeff Sebo as director, me (Sofia) as coordinator, and Ned Block, Samuel Bowman, David Chalmers, Becca Franks, Joshua Lewis, S. Matthew Liao, Claudia Passos-Ferreira, Luke Roelofs, Katrina Wyman, and others as faculty affiliates.
    MEP has a variety of projects underway; you can find information about two of them below.
    David Chalmers Public Talk
    On October 13 2022 at 5:00pm ET, MEP will host a public talk by David Chalmers (University Professor of Philosophy and Neural Science and Co-Director of the Center for Mind, Brain, and Consciousness, NYU; author, The Conscious Mind, Constructing the World, and Reality+: Virtual Worlds and the Problems of Philosophy) on whether large language models are sentient. This talk, which will be co-sponsored by the NYU Center for Bioethics, the NYU Center for Mind, Brain, and Consciousness, and the NYU Minds, Brains, and Machines Initiative, will be free and open to the public, in person and online. You can register to attend here.
    Early-Career Award and Workshop
    MEP is announcing an Early-Career Award and Workshop on Animal and AI Consciousness. PhD students and early-career faculty (PhD 2017 or later) in any field are invited to submit current or recent (published 2021 or later) work on this topic. Selected authors will receive a $500 award and an all-expenses paid trip to the Association of Scientific Studies of Consciousness Conference at NYU in June 2023. They will also be invited to speak at a Workshop on Animal and AI Consciousness associated with this conference. You can find more information about this award and workshop, including how to submit, here.
    Finally, if you have interest in receiving updates from MEP, you can visit our website or sign up for our email list.
    If you have interest in participating in our work, please contact Jeff or me to discuss.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • LW - Why doesn't China (or didn't anyone) encourage/mandate elastomeric respirators to control COVID? by Wei Dai
    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 doesn't China (or didn't anyone) encourage/mandate elastomeric respirators to control COVID?, published by Wei Dai on September 17, 2022 on LessWrong.
    (This post was inspired in part by this comment by Florin, which I think makes some good related points.)
    Elastomeric respirators are reusable respirators made with some elastic material such as silicone, typically with N95 or better filters attached. They are low cost (there's currently one from a reputable manufacturer on sale for $2.99[1], which may be a loss leader but even the regular price is just $10.99) and are easier to correctly use than disposable N95s (since they form a soft seal against the face and are therefore less prone to leaks), and of course are much more effective than the cloth or surgical masks that most people still wear, even now (if they wear masks at all, e.g. in Chengdu, China).
    It seems easy to imagine that if everyone or most people wore such a respirator, COVID would be much easier to control and we could depend less on other measures, including draconian, extremely high cost ones like lockdowns. Yet every locality on Earth throughout the pandemic, including present day China, would apparently prefer to not mandate or even encourage using such respirators and instead incur the high costs of other control measures. I'm tempted to pass this off as typical irrationality of humans, but feel like I should check with the LW hivemind before I do. Is there some reasonable explanation that I'm missing?
    (At one point I saw A Clunky Mask May Be the Answer to Airborne Disease and N95 Waste in the NY Times, which made my heart jump a little, but it was only talking about use of these respirators by medical personnel, not the general population.)
    ETA: Elastomeric respirators do have some downsides, however many of them can be mitigated through better design. Here is an improved design which is currently on sale for $89.99, but can probably be made much cheaper if manufactured at scale.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min