The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • AF - What I mean by "alignment is in large part about making cognition aimable at all" by Nate Soares
    Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: What I mean by "alignment is in large part about making cognition aimable at all", published by Nate Soares on January 30, 2023 on The AI Alignment Forum.
    (Epistemic status: attempting to clear up a misunderstanding about points I have attempted to make in the past. This post is not intended as an argument for those points.)
    I have long said that the lion's share of the AI alignment problem seems to me to be about pointing powerful cognition at anything at all, rather than figuring out what to point it at.
    It’s recently come to my attention that some people have misunderstood this point, so I’ll attempt to clarify here.
    In saying the above, I do not mean the following:
    (1) Any practical AI that you're dealing with will necessarily be cleanly internally organized around pursuing a single objective. Managing to put your own objective into this "goal slot" (as opposed to having the goal slot set by random happenstance) is a central difficult challenge. [Reminder: I am not asserting this]
    Instead, I mean something more like the following:
    (2) By default, the first minds humanity makes will be a terrible spaghetti-code mess, with no clearly-factored-out "goal" that the surrounding cognition pursues in a unified way. The mind will be more like a pile of complex, messily interconnected kludges, whose ultimate behavior is sensitive to the particulars of how it reflects and irons out the tensions within itself over time.
    Making the AI even have something vaguely nearing a 'goal slot' that is stable under various operating pressures (such as reflection) during the course of operation, is an undertaking that requires mastery of cognition in its own right—mastery of a sort that we’re exceedingly unlikely to achieve if we just try to figure out how to build a mind, without filtering for approaches that are more legible and aimable.
    Separately and independently, I believe that by the time an AI has fully completed the transition to hard superintelligence, it will have ironed out a bunch of the wrinkles and will be oriented around a particular goal (at least behaviorally, cf. efficiency—though I would also guess that the mental architecture ultimately ends up cleanly-factored (albeit not in a way that creates a single point of failure, goalwise)).
    (But this doesn’t help solve the problem, because by the time the strongly superintelligent AI has ironed itself out into something with a "goal slot", it's not letting you touch it.)
    Furthermore, insofar as the AI is capable of finding actions that force the future into some narrow band, I expect that it will tend to be reasonable to talk about the AI as if it is (more-or-less, most of the time) "pursuing some objective", even in the stage where it's in fact a giant kludgey mess that's sorting itself out over time in ways that are unpredictable to you.
    I can see how my attempts to express these other beliefs could confuse people into thinking that I meant something more like (1) above (“Any practical AI that you're dealing with will necessarily be cleanly internally organized around pursuing a single objective.”), when in fact I mean something more like (2) (“By default, the first minds humanity makes will be a terrible spaghetti-code mess.”).
    In case it helps those who were previously confused: the "diamond maximizer" problem is one example of an attempt to direct researchers’ attention to the challenge of cleanly factoring cognition around something a bit like a 'goal slot'.
    As evidence of a misunderstanding here: people sometimes hear me describe the diamond maximizer problem, and respond to me by proposing training regimes that (for all they know) might make the AI care a little about diamonds in some contexts.
    This misunderstanding of what the diamond maximizer problem was originally meant to be pointing at seems plausibly related t...
    4 min
  • EA - Announcing Interim CEOs of EVF by Owen Cotton-Barratt
    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 Interim CEOs of EVF, published by Owen Cotton-Barratt on January 30, 2023 on The Effective Altruism Forum.
    Overview
    Effective Ventures Foundation USA (“EVF US”) has appointed Zachary Robinson as its Interim CEO as of today. We’re also taking this time to announce that Effective Ventures Foundation (“EVF UK”) appointed Howie Lempel as its Interim CEO back in November.
    EVF UK and EVF US together act as hosts and fiscal sponsors of many key projects in effective altruism, including the Centre for Effective Altruism (CEA), 80,000 Hours (80k), Longview Philanthropy, etc. These charities (EVF UK and EVF US) previously did not have the position of CEO, and project leads (for CEA, 80k, etc.) reported directly to EVF boards for management oversight.
    The FTX situation has given rise to a number of new challenges. In particular, there are financial and legal issues that occur at the level of the charities (EVF UK and EVF US), not at the project level, because the projects are not separate legal entities. Because of this, we’ve chosen to coordinate the legal, financial, and communications strategies of the projects under EVF much more than before.
    In response to the new challenges from FTX, the boards became much more involved in the day-to-day operations of the charities. But it’s not ideal to have boards playing the role of executives, so the boards have now also appointed Interim CEOs to each charity.
    The Interim CEO roles are about handling crises and helping the entities transition to an improved long-term structure. They are naturally time-limited roles; we aren’t sure how they might change, or when it will make sense for Howie and/or Zach to hand the reins off. The announcement of these Interim CEOs doesn’t constitute any change of project leadership; for example, Max Dalton will continue as leader of CEA, the community-building project which is part of EVF.
    Meta remarks
    This post is written on behalf of the boards of EVF. It’s difficult to write or act as a group without either doing things that not everyone is totally behind, or constraining ourselves to the bland. In the case of this post, it is largely the work of the primary authors; other board members might not agree with the more subjective judgements.
    The impetus for getting this post out now was a desire to empower the CEOs to speak on behalf of the entities; there are some time-sensitive updates they expect to share soon.
    Edit: Howie has now shared one
    There’s a lot to say about what FTX’s collapse means for EA and EVF; in many ways we’re still in the early days of wrestling with what this means for us and our communities. This post isn’t about that. However, we know that this is the first major public communication from the EVF boards, and we don’t want the implicature to be that we don’t think this is important or worth discussing.
    We’re hoping to write more soon about why we haven’t said more, how we’re seeing the situation, and how EVF’s choices may have impacted EA discourse about FTX.
    About Howie and Zach
    Howie Lempel has been Interim CEO of EVF UK since mid-November. To take this on, Howie took leave from his role as CEO of 80k, one of the largest EVF projects. In light of Howie’s move, Brenton Mayer was appointed as Interim CEO of 80k in December. Brenton was previously Director of Internal Systems at 80k.
    Before he worked at 80k, Howie went to Yale Law for two years. He left Yale to join Open Philanthropy while it was still being incubated at GiveWell, working as their first Program Officer for Global Catastrophic Risks. Howie has also worked on white collar crime at the Manhattan DA’s office and on U.S. economic policy as a research assistant at the Brookings Institution. In addition to his role as CEO of 80k, he’s also known in the EA community for his personal podcast episode on mental he...
    11 min
  • EA - Time-stamping: An urgent, neglected AI safety measure by Axel Svensson
    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: Time-stamping: An urgent, neglected AI safety measure, published by Axel Svensson on January 30, 2023 on The Effective Altruism Forum.
    TL;DR
    I believe we should use a 5-digit annual budget to create/serve trust-less, cryptographic timestamps of all public content, in order to significantly counteract the growing threat that AI-generated fake content poses to truth-seeking and trust. We should also encourage and help all organizations to do likewise with private content.
    THE PROBLEM & SOLUTION
    As the rate and quality of AI-generated content keeps increasing, it seems inevitable that it will become easier to create fake content and harder to verify/refute it. Remember the very recent past when faking a photo was so hard that simply providing a photo was considered proof? If we do nothing about it, these AI advances might have a devastating impact on people's opportunities to trust both each other and historic material, and might end up having negative value for humanity on net.
    I believe that trust-less time-stamping is an effective, urgent, tractable and cheap method to partly, but significantly so, counteract this lamentable development. Here's why:
    EFFECTIVE
    It is likely that fake creation technology will outpace fake detection technology. If so, we will nominally end up in an indefinite state of having to doubt pretty much all content. However, with trust-less time-stamping, the contest instead becomes between the fake creation technology available at the time of timestamping, and the fake detection technology available at the time of truth-seeking.
    Time-stamping everything today will protect all past and current content against suspicion of interference by all future fake creation technology. As both fake creation and fake detection technology progress, no matter at what relative pace, the value of timestamps will grow over time. Perhaps in a not so distant future, it will become an indispensable historical record.
    URGENT
    Need I say much about the pace of progress for AI technology, or the extent of existing content? The value of timestamping everything today rather than in one month, is some function of the value of the truth of all historical records and other content, and technological development during that time. I suspect there's a multiplication somewhere in that function.
    TRACTABLE
    We already have the cryptographic technology and infrastructure to make trust-less timestamps. We also have large public archives of digital and/or digitized content, including but not limited to the web. Time-stamping all of it might not be trivial, but it's not particularly hard. It can even be done without convincing very many people that it needs to be done. For non-public content, adding timestamping as a feature in backup software should be similarly tractable - here the main struggle will probably be to convince users of the value of timestamping.
    Implementation: Each piece of content is hashed, the hashes put into a merkle tree, and the root of that tree published on several popular, secure, trust-less public ledgers. Proof of timestamp is produced as a list of hashes along the merkle branch from the content up to the root, together with transaction IDs. This technology, including implementations, services and public ledgers already exists. For private content, you might want to be able to prove a timestamp for one piece of content without divulging the existence of another piece of content. To do so, one would add one bottom level in the merkle tree where each content hash is hashed with a pseudo-random value rather than another content hash. This pseudo-random value can be produced from the content hash itself and a salt that is constant within an organization.
    CHEAP
    Timestamping n pieces of content comprising a total of b bytes will incur a one-time cost for processin...
    6 min
  • EA - Proposed improvements to EAG(x) admissions process by Elika
    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: Proposed improvements to EAG(x) admissions process, published by Elika on January 30, 2023 on The Effective Altruism Forum.
    This post is part of an ongoing series: Events in EA: Learnings and Critiques.
    Elika worked (as a contractor) for the Centre for Effective Altruism (CEA) to run EAGxBerkeley. This post is written in a personal capacity, and are not the views of or endorsed by CEA.
    Applications
    Do a better job of auto filling application forms with information from previous applications.
    Applicants with less community experience who might be a good fit for the conference don’t necessarily know community communication norms. The following suggestions aim to provide more context and transparency around the process of application processing, without providing information that would allow applicants to goodhart.
    Make important questions mandatory to reduce the rate of incomplete applications.
    Adding context to questions (e.g. for some questions, if the answer is not at all related to EA it's hard to know if the person misunderstood the question, or just has had very limited interaction with EA)
    Asking applicants to provide specific, concrete answers to questions rather than vague generalities (e.g. “I’d like to learn more about EA” is very vague, it would be more helpful to know what specifically they’d like to learn, and what they already know)
    Tell applicants not to assume the organiser knows who you are (I think it's generally bad for culture and status).
    Admissions
    Make admissions policies clear (ex. this conference is for people in X region), both in public communications and throughout the application process.
    Don’t accept applicants after the deadline has passed. If you do want to accept last-minute applicants, have a consistent and fair criteria for accepting late applications
    Experiment with adding different questions to the application, such as:
    Check yes if you’re okay giving up your spot for a first time attendee if the conference is at capacity. We may still accept you if we think you could help other attendees.
    Checking yes if you’re applying but are uncertain you will register and attend (excluding last minute emergencies) to better forecast how many accepted applicants will actually apply.
    Check yes to “I can attend with <1 week’s notice” for people (e.g. locals) who are willing / able to attend last minute. (H/T Larks)
    "Why do you think it would be valuable to attend this specific conference" Adding more intentionality to the conference could help make applicants more likely to commit, and help organisers decide between applicants.
    Removing the multiple conference check box so you can only apply to one conference at a time to improve intentionality.
    Consider making all tickets (barring needs-based applicants) paid so that people feel a greater sense of commitment towards attending. We will write more about this in a future post.
    Create a “Virtual Only” option for people who:
    Don’t want to attend in-person, but do want to be listed as a resource for others (e.g. see Vael’s comment)
    Initially wanted to attend in-person, but aren’t able to and still want access to the attendee list (e.g. see Lorenzo’s comment)
    Reminders & Comms to Accepted Applicants
    Send more (at least 3) reminders to accepted applicants to register, with reminders about how many people are waitlisted to incentivise people to release their ticket
    Reminder to register: Hey X, please register for this event! If you don’t register by August 18th, we’ll give your spot to someone on the waitlist. Registering helps us ensure that we get estimates to vendors on time, and makes sure that the application process is fair for everyone. Read more here (link to a post explaining the downsides)
    Release accepted but not registered applicants' tickets after a certain date and clearly communicate that ...
    5 min
  • LW - How Likely is Losing a Google Account? by jefftk
    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 Likely is Losing a Google Account?, published by jefftk on January 30, 2023 on LessWrong.
    Let's say you use a Google account as the root of your online identity: Gmail, Fi, "Sign in with Google", Google's password manager, etc. How much should you be worried that you'll suddenly find yourself unable to get into everything? Should you do something else instead?
    To get a sense of how common lockouts are and how they happen I looked through lockout reports on Hacker News by searching for
    [google blocked account] and [google locked out]. I looked at top-level stories and the comments on them for cases where people were entirely locked out of an account; I didn't include cases where people lost access to only some Google services (Payments, AdWords, etc) or where they did get back in on their own. I did count cases where it took making a lot of noise on
    HN or Twitter, though.
    There are two reasons people seem to get locked out:
    Security lockouts: they're not convinced you're you, and are trying to prevent an attacker from getting into your account.
    Policy lockouts: they don't like you. They've flagged your account as abusive, enough that they completely suspend your account.
    I found 32 cases (sheet), going back to 2008. I found 22 security lockouts, 7 policy lockouts, and 3 with too few details to tell. I think this likely majorly undercounts security lockouts relative to policy ones: reading the comments, a lot of the security ones were like "that happened to me too" while the policy ones got mainstream news articles. [1]
    With the security lockouts, in cases where you could tell what had happened the most common reason was that someone had configured a backup method (phone number, recovery email, 2FA) but no longer had access. The second most common were cases where someone hadn't configured any backup methods and Google was considering their login to be suspicious.
    Security lockouts are a tricky situation because failures in either direction are very bad. All of the above are false positives: people who should have been let back into their accounts but weren't. But there are also false negatives: cases where an attacker was let into someone's account.
    There is a question of whether Google should be flagging suspicious logins at all, though. If my account were protected only by a password and someone else got it, maybe Google should just let them in? The problem is that this would mean lots of hacked accounts: it's common for passwords to get revealed through phishing or cross-site password reuse. Using other aspects of your login, like your country, device, activity pattern, etc. as a kind of pseudo-2FA probably does make things better for users overall: if my username and password suddenly appeared from a new device in Russia there really is a good chance that it's someone trying to hack me. Luckily, users have a better option: opting into tighter security by setting up good 2FA. The more ways you can demonstrate that you are actually you, the lower the risk both of hacks and lockouts.
    After going through these, it seems to me that the likelihood of a security lockout is low enough not to worry about if you:
    In addition to memorizing your password, write it down and store it in a safe place. This is also worth doing if there's someone you want to have access to the account immediately if something happens to you (the Inactive Account
    Manager is also good, but it reasonably has a substantial delay).
    Configure backup methods (phone, email).
    Update backup methods promptly when they change. You can see what you have configured at myaccount.google.com/security.
    Ideally, set up security keys for two-factor authentication, and if you do set up three
    (work, home, keychain or phone).
    What about the policy lockouts, though? I think the risk there is also very low: these are ...
    6 min
  • EA - EA novel published on Amazon by timunderwood
    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: EA novel published on Amazon, published by timunderwood on January 29, 2023 on The Effective Altruism Forum.
    Just so everyone knows, the novel I wrote to promote EA has been published on Amazon. This is not a request to purchase it, or try to directly support it in any particular way.
    However if the blurb does sound interesting, go for it :)
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min
  • LW - Against Boltzmann mesaoptimizers by porby
    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: Against Boltzmann mesaoptimizers, published by porby on January 30, 2023 on LessWrong.
    I'm effectively certain that a weakly defined version of "mesaoptimizers"- like a component of a network which learns least squares optimization in some context- is practically reachable by SGD.
    I'm pretty sure there exists some configuration of an extremely large neural network such that it contains an agentic mesaoptimizer capable of influencing the training gradient for at least one step in a surprising way. In other words, gradient hacking.
    I think there's a really, really big gap between these which is where approximately all the relevant and interesting questions live, and I'd like to see more poking aimed at that gap.
    Skipping the inspection of that gap probably throws out useful research paths.
    I think this belongs to a common, and more general, class of issue.
    Boltzmann mesaoptimizers
    Boltzmann brains are not impossible. They're a valid state of matter, and there could be a physical path to that state. You might just have to roll the dice a few times, for a generous definition of few.
    And yet I feel quite confident in claiming that we won't observe a Boltzmann brain in the wild within the next few years (for a similar definition of few).
    The existence of a valid state, and a conceivable path to reach that state, is not enough to justify a claim that that state will be observed with non-negligible probability.
    I suspect that agentic mesaoptimizers capable of intentionally distorting training are more accessible than natural Boltzmann brains by orders of magnitude of orders of magnitude, but it is still not clear to me that the strong version is something that can come into being accidentally in all architectures/training modes that aren't specifically designed to defend against it.
    Minding the gap
    I feel safe in saying that it would be a mistake to act on the assumption that you're a Boltzmann brain. If you know your history is coherent (to the degree that it is) only by pure chance, and your future practically doesn't exist, then in any given moment, your choices are irrelevant. Stuff like 'making decisions' or 'seeking goals' goes out the window.
    Observing that there is the logical possibility of a capable misaligned agentic mesaoptimizer and then treating any technique that admits such a possibility as irrevocably broken would be similarly pathological.
    If we cannot find a path that eliminates the logical possibility by construction, we must work within the system that contains the possibility. It's a little spookier, not having the force of an impossibility proof behind you, but complex systems can be stable. It may be the case that the distance from the natural outcome and the pathological outcome is enormous, and that the path through mindspace doesn't necessarily have the maximal number of traps permitted by your current uncertainty.
    It's worth trying to map out that space.
    Can we create a clear example of an agentic mesaoptimizer in any design at sub-dangerous scales so that we can study it?
    Can we learn anything from the effect of scaling on designs that successfully exhibit agentic mesaoptimizers?
    What designs make agentic mesaoptimizers a more natural outcome?
    What designs makes agentic mesaoptimizers less natural?
    How far does SGD need to walk from a raw goal agnostic simulator to find its way to an agentic mesaoptimizer, for any type of agent?
    How much force does it require to push an originally goal agnostic simulator into agentic mesaoptimization?
    Try repeating all of the above questions for nonagentic mesaoptimizers. Can we learn anything about the distance between agentic and nonagentic mesaoptimizers? Is agency in a mesaoptimizer a natural attractor? Under what conditions?
    Can we construct any example of gradient hacking, no matter how contrived or ephemer...
    8 min
  • LW - Compendium of problems with RLHF by Raphaël S
    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: Compendium of problems with RLHF, published by Raphaël S on January 29, 2023 on LessWrong.
    Epistemic status: This post is a distillation of many comments/posts. I believe that my list of problems is not the best organization of sub-problems. I would like to make it shorter, and simpler, because cool theories are generally simple unified theories, by identifying only 2 or 3 main problems without aggregating problems with different types of gear level mechanisms, but currently I am too confused to be able to do so. Note that this post is not intended to address the potential negative impact of RLHF research on the world, but rather to identify the key technical gaps that need to be addressed for an effective alignment solution. Many thanks to Walter Laurito, Fabien Roger, Ben Hayum, Justis Mills for useful feedbacks.
    RLHF tldr: We need a reward function, we cannot hand-write it, let’s make the AI learn it!
    Problem 0: RLHF is confusing. Human judgment and feedback is so brittle, that even junior alignment researchers like me thought that RLHF is a not-too-bad-solution to the outer alignment problem. I think RLHF confuses a lot of people and distracts people from the core issues. Here is my attempt to become less confused.
    Why RLHF counts as progress?
    Manual reward crafting : “By comparison, we took two hours to write our own reward function (the animation in the above right) to get a robot to backflip, and though it succeeds it’s a lot less elegant than the one trained simply through human feedback”
    RLHF learned to backflip using around 900 individual bits of feedback from the human evaluator.
    From Learning from Human Preferences.
    Without RLHF, approximating the reward function for a “good” backflip was tedious and almost impossible. With RLHF it is possible to obtain a reward function, that when optimized by an RL policy leads to beautiful backflips.
    Value is complex? Value is fragile? But being polite is complex, being polite is fragile, and we can implement this roughly in ChatGPT.
    “RLHF withstands more optimization pressure than supervised fine-tuning: By using RLHF you can obtain reward functions, that withstand more optimization pressure, than traditionally hand-crafted reward functions. Insofar as that's a key metric we care about, it counts as progress.” [Richard’s comment]
    Why RLHF is insufficient?
    Buck highlights two main types problems with using RLHF to create an AGI: oversight issues and the potential for catastrophic outcomes. This partition of the problem space comes from the post Low-Stakes alignment. Although it may be feasible to categorize all problems under this partition, I believe that this categorization is not granular enough and lumps different types of problems together.
    Davidad suggests dividing the use of RLHF into two categories: "Vanilla-RLHF-Plan" and "Broadly-Construed-RLHF" as part of the alignment plan. The "Vanilla-RLHF-plan" refers to the narrow use of RLHF (say as used in ChatGPT) and is not sufficient for alignment, while "Broadly-Construed-RLHF" refers to the more general use of RLHF as a building block and is potentially useful for alignment. The list below outlines the problems with "Vanilla-RLHF" that "Broadly-Construed-RLHF" should aim to address.
    Existing problems with RLHF because of (currently) non-robust ML systems
    Those issues seem to be mostly a result of poor capabilities, and I think those problems may disappear as models grow larger.
    1. Benign Failures: ChatGPT can fail. And it is not clear if this problem will disappear as capabilities scale.
    2. Mode Collapse, i.e. a drastic bias toward particular completions and patterns. Mode collapse is expected when doing RL.
    3. You need regularization. Your model can severely diverge from the model you would have gotten if had gotten feedback in real-time from real humans. You ne...
    18 min
  • EA - Protect Our Future's Crypto Politics by Mohammad Ismam Huda
    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: Protect Our Future's Crypto Politics, published by Mohammad Ismam Huda on January 29, 2023 on The Effective Altruism Forum.
    TLDR; The Effective Altruism-linked political fund, Protect Our Future, has come under scrutiny for perceived cryptocurrency lobbying. Skip to section Irregularities to see the findings of my investigation into the group.
    Background
    The effective altruism movement has become increasingly interested and involved in politics. A major vehicle for this involvement has been Protect Our Future PAC (POF), a political fund active in the recent 2022 US midterm elections. The fund ostensibly sought to lobby for effective altruist cause areas, with a particular emphasis on longtermism and pandemic prevention.
    POF entered the electoral scene with a bang funding the campaign of Effective Altruist Carrick Flynn to the tune of $10 million. The fund proceeded to donate to 27 candidates in total, exclusively in the Democratic Primaries.
    In total POF raised and spent $28 million, nearly all of this coming from Sam Bankman-Fried (SBF) the cryptocurrency tycoon. Despite SBF's financial support, POF has repeatedly denied the fund served SBF's financial interests in cryptocurrency.
    Irregularities
    As a result of my interest in POF, I proceeded to research and investigate the group and have discovered what to me are irregularities.
    I reached out to the President (Michael Sadowksy) and Press Spokesman (Mike Levine) of Protect Our Future, urging them to provide an explanation, but received no response. My inquiry can be viewed here.
    Cryptocurrency Interests Amongst Candidates
    Of the 27 endorsed candidates I have been able to link 16 as having meaningful pro-cryptocurrency sentiment and/or comittee positions that would make them target-worthy by cryptocurrency lobbyists.
    A selection of candidates and their ties are presented here. A full table of all candidates endorsed by POF is available in the appendix.
    POF CandidateCrypto Sympathies/ Conflict of InterestAbigail Spanberger (VA-07)://www.washingtonexaminer.com/policy/economy/sec-official-steps-down-ftx-connectionAdam Hollier (MI-13)Received donations from Web3Forward PAC, a cryptocurrency lobbying group - $412K Gilbert Villegas (IL-03)Expressed pro-cryptocurrency sentiments as an alderman.
    Rep. Spanberger sits on the subcomittee which oversees the CFTC.It is known that SBF and other crypto adherents were lobbying authorities to shift crypto oversight from the SEC to CFTC.
    The presence of cryptocurrency advocates amongst POF's endorsed candidates was also picked up by the press:
    The Bankman-Frieds have insisted the organizations are focused on pandemic preparedness and aren’t about influencing cryptocurrency policy. But endorsed candidates are sometimes major supporters of the crypto sector. Lafazan, for example, said back in June he would take campaign donations in cryptocurrency and said he would file a bill in the Nassau County Legislature to create a cryptocurrency task force. Torres, too, who won support from Guarding Against Pandemics, has also publicly boosted the industry.
    Ray La Raja, a political scientist who studies campaign finance at the University of Massachusetts – Amherst, says it’s likely no coincidence that Protect Our Future’s funds have gone to establishment candidates at a time when Congress is considering cryptocurrency regulations.
    One, Maxwell Frost, a Florida Democrat who has become well known as the first Gen-Z member of Congress, got $8,700 in contributions from the Bankman-Fried brothers and nearly $1 million in help from Protect Our Future, almost all of it after announcing a “crypto-advisory council” for his campaign.
    Opportunity For Tommorrow PAC
    There is also another fund that appears to have close ties to Protect Our Future called Opportunity For Tomorrow. The fund was exclusively donated...
    11 min
  • EA - Biosecurity newsletters you should subscribe to by Sofya Lebedeva
    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: Biosecurity newsletters you should subscribe to, published by Sofya Lebedeva on January 29, 2023 on The Effective Altruism Forum.
    TL;DR
    I searched for other lists of biosecurity newsletters specifically and didn’t find one that suited my needs, so I made one! Please leave a comment with any other newsletters that I missed so that I can add them. I hope you find something useful in this list.
    Newsletters
    Johns Hopkins Center for Health Security (CHS) subscribe here
    Health Security Headlines (also from CHS) subscribe here
    Center for Infectious Disease Research and Policy (CIDRAP) subscribe here
    Global Biodefence subscribe here
    Pandora Report subscribe here
    Nuclear Threat Initiative (NTI) subscribe here
    Bipartisan Commission on Biodefence subscribe here
    The Association for Biosafety and Biosecurity (ABSA) subscribe here (scroll to the bottom)
    Council on Strategic Risks (CSR) subscribe here
    Biomedical Advanced Research and Development Authority (BARDA) subscribe here
    Administration for Strategic Preparedness and Response (ASPR) do not recommend subscribing as it is very narrowly biosafety oriented and is sometimes poorly referenced.
    Comments
    I sourced a lot of these from recommendations by Caitlin Walker, as well as from looking through various posts by Chris Bakerlee and Tessa Alexanian. Please don’t hesitate to point out any links that are broken, comment about the relative quality of the above newsletters, or comment with any newsletters that I have missed. Thank you!
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    2 min