The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • LW - Why I hate the "accident vs. misuse" AI x-risk dichotomy (quick thoughts on "structural risk") by David Scott Krueger (formerly: capybaralet)
    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 I hate the "accident vs. misuse" AI x-risk dichotomy (quick thoughts on "structural risk"), published by David Scott Krueger (formerly: capybaralet) on January 30, 2023 on LessWrong.
    I think the large majority of AI x-risk is "structural". Like climate change.
    Here's a good primer on structural risk (note that structural risk is not a synonym for "not caused by out-of-control AI"):
    I am shocked and amazed and dismayed that more people do not seem to view it this way, even among the AI x-safety community. Heck, even Eliezer's stories of doom are steeped in structural risk (race dynamics, teams rationalizing cutting corners on safety when they should know better, etc.)
    I expect irresponsible, reckless, negligent deployment of AI systems without proper accounting of externalities. I consider this the default for any technology with potential for significant externalities, absent regulation.When something bad happens in such a context, calling it "accident risk" absolves those researching, developing, and/or deploying the technology of responsibility. They should have known better. Some of them almost certainly did. Rationalization, oversight, and misaligned incentives were almost certainly at play. Failing to predict the particular failure mode encountered is no excuse. Having "good intentions" is no excuse.So... it must be misuse then, right? Well, no. Calling it "misuse" suggests that those researching, developing, and/or deploying the technology set out with nefarious purposes and the technology achieved precisely what they intended. But ~nobody wants to destroy the world.
    It's just that most people are somewhat selfish and so are willing to trade some x-risk for a large personal benefit.In summary, saying "accident" makes it sounds like an unpredictable effect, instead of painfully obviously risk that was not taken seriously enough. Saying "misuse" makes it sounds like some supervillian or extremist deliberately destroying the world. While some risks may have something more of a flavor or accident or misuse depending on how obvious the risk was, neither of these pictures gives a remotely accurate picture of the nature of the problem. I think this makes it a harmful meme, and ask that others stop making this distinction (without appropriate caveats), and join me in pointing out how it contributes to a confused and misleading discourse when others do.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • LW - Simulacra Levels Summary by Zvi
    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: Simulacra Levels Summary, published by Zvi on January 30, 2023 on LessWrong.
    For More Detail, Previously: Simulacra Levels and Their Interactions, Unifying the Simulacra Definitions, The Four Children of the Seder as the Simulacra Levels.
    A key source of misunderstanding and conflict is failure to distinguish between combinations of the following four cases.
    Sometimes people model and describe the physical world, seeking to convey true information because it is true.
    Other times people are trying to get you to believe what they want you to believe so you will do or say what they want.
    Other times people say things mostly as slogans or symbols to tell you what tribe or faction they belong to, or what type of person they are.
    Then there are times when talk seems to have have gone strangely meta or off the rails entirely. The symbolic representations are mostly of the associations and vibes of other symbols. The whole thing seems more like a stream of words, associations and vibes. It sounds like GPT-4.
    One can refer to these as the simulacra levels as a useful fake framework for understanding this. When looking at talk, one can ask what level or levels a statement or discussion is on, and which ones people care about in context. One can also ask about the level a person, group or civilization most cares about. That is also how they default to understanding new talk.
    This concept has important details that are difficult to understand. The posts linked up top offer discussions of four definitions that all point at the same dynamics. Each is stronger at capturing different elements.
    As a more concise alternative, this post gathers together the most vital information.
    First, the more straightforward definitions from 2020.
    The Lion and Pandemic Definitions
    The lion definition asks what each level means by ‘There is a lion across the river.’
    Level 1: There’s a lion across the river.
    Level 2: I don’t want to go (or have other people go) across the river.
    Level 3: I’m with the popular kids who are too cool to go across the river.
    Level 4: A firm stance against trans-river expansionism focus-grouped well with undecided voters in my constituency.
    The pandemic definition changes the statement to ‘There is a pandemic heading our way from China.’
    Level 1: “There’s a pandemic headed our way from China.”
    Level 2: “I want you to act as if you think there might be a pandemic on our way from China” while hoping to still be interpreted by the listener as meaning “There’s a pandemic headed our way from China.”
    Level 3: “I wish to associate with the group that claims there is a pandemic headed our way from China.”
    Level 4 (old version): “It is advantageous for me to say there is a pandemic headed our way from China.”
    Except no, that’s not quite right. The key issue with the lion and pandemic definitions is treating Level 4 as if it has motivations and does things for logical reasons. One can think strategically about level 4 implications but those operating on Level 4 mostly not only don’t do this, they have lost the ability to do so.
    Therefore, the real situation more like:
    Level 4 (new version): “It feels like the thing for me to say here that there is a pandemic headed our way from China.”
    More abstractly:
    Level 1: Truth. Attempt to accurately share and describe physical reality.
    Level 2: Manipulation of Perception. Lies. Attempt to shape perception of reality, so that others will act on that perception, without regard to whether it is true.
    Level 3: Association. Attempting to change perception of one’s social position and alliances, rather than expecting anyone to act upon beliefs about physical reality. Requires maintaining some plausibility of the underlying physical claim.
    Level 4: Manipulation and Intuition. Occasionally a strategic attempt to manipulate Level 3 dynamics. More cent...
    13 min
  • EA - We're no longer "pausing most new longtermist funding commitments" by Holden Karnofsky
    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: We're no longer "pausing most new longtermist funding commitments", published by Holden Karnofsky on January 30, 2023 on The Effective Altruism Forum.
    In November, I wrote about Open Philanthropy’s soft pause of new longtermist funding commitments:
    We will have to raise our bar for longtermist grantmaking: with more funding opportunities that we’re choosing between, we’ll have to fund a lower percentage of them. This means grants that we would’ve made before might no longer be made, and/or we might want to provide smaller amounts of money to projects we previously would have supported more generously ..
    Open Philanthropy also need[s] to raise its bar in light of general market movements (particularly the fall in META stock) and other factors ... the longtermist community has been growing; our rate of spending has been going up; and we expect both of these trends to continue. This further contributes to the need to raise our bar ...
    It’s a priority for us to think through how much to raise the bar for longtermist grantmaking, and therefore what kinds of giving opportunities to fund. We hope to gain some clarity on this in the next month or so, but right now we’re dealing with major new information and don’t have a lot to say about what it means. It could mean reducing support for a lot of projects, or for relatively few ...
    Because of this, we are pausing most new longtermist funding commitments (that is, commitments within Potential Risks from Advanced Artificial Intelligence, Biosecurity & Pandemic Preparedness, and Effective Altruism Community Growth) until we gain more clarity, which we hope will be within a month or so ...
    It’s not an absolute pause: we will continue to do some longtermist grantmaking, mostly when it is time-sensitive and seems highly likely to end up above our bar (this is especially likely for relatively small grants).
    Since then, we’ve done some work to assess where our new funding bar should be, and we have created enough internal guidance that the pause no longer applies. (The pause stopped applying about a month ago, but it took some additional time to write publicly about it.)
    What did we do to come up with new guidance on where the bar is?
    What we did:
    Ranking past grants: We created1 a rough ranking of nearly all the grants we’d made over the last 18 months; we also included a number of grants now-defunct FTX-associated funders had made.
    The basic idea of the ranking is essentially: “For any two grants, the grant we would make if we could only make one should rank higher.”2 It is based on a combination of rough quantitative impact estimates, grantmaker intuitions, etc.
    With this ranking, we are now able to take any given total Open Phil longtermist budget for the time period in question, and identify which grants would and would not have made the cut under that budget.
    In some cases, we gave separate rankings to separate “tranches” of a grant (e.g., the first $1 million of a grant might be ranked much higher than the next $1 million, because of diminishing returns to adding more funding to a given project).
    Annual spending guidelines: We estimated how much annual spending would exhaust the capital we have available for longtermist grantmaking over the next 20-50 years (the difference between 20 years and 50 years is not huge3 in per-year spending terms), to get some anchors on what our annual budget for longtermist grantmaking should be.
    We planned conservatively here, assuming that longtermist work would end up with 30-50% of all funding available to Open Philanthropy.4 OP is also working on a longer-term project to revisit how we should allocate our resources between longtermist and global health and wellbeing funding; it’s possible that longtermist work will end up with more than 50%, which would leave more room to grow.
    Setting the b...
    10 min
  • EA - An in-progress experiment to test how Laplace’s rule of succession performs in practice. by NunoSempere
    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: An in-progress experiment to test how Laplace’s rule of succession performs in practice., published by NunoSempere on January 30, 2023 on The Effective Altruism Forum.
    Note: Of reduced interest to generalist audiences.
    Summary
    I compiled a dataset of 206 mathematical conjectures together with the years in which they were posited. Then in a few years, I intend to check whether the probabilities implied by Laplace’s rule—which only depends on the number of years passed since a conjecture was created—are about right.
    In a few years, I think this will shed some light on whether Laplace’s rule of succession is useful in practice. For people wanting answers more quickly, I also outline some further work which could be done to obtain results now.
    The dataset I’m using can be seen here (a).
    Probability that a conjecture will be resolved by a given year according to Laplace’s law.
    I estimate the probability that a randomly chosen conjecture will be solved as follows:
    That is, the probability that the conjecture will first be solved in the year n is the probability given by Laplace conditional on it not having been solved any year before.
    For reference, a “pseudo-count” corresponds to either changing the numerator to an integer higher than one, or to making n higher. This can be used to capture some of the structure that a problem manifests. E.g., if we don’t think that the prior probability of a theorem being solved in the first year is around 50%, this can be addressed by adding pseudo-counts.
    Code to do these operations in the programming language R can be found here. A dataset that includes these probabilities can be seen here.
    Expected distribution of the number of resolved conjectures according to Laplace’s rule of succession
    Using the above probabilities, we can, through sampling, estimate the number of conjectures in our database that will be solved in the next 3, 5, or 10 years. The code to do this is in the same R file linked a paragraph ago.
    For three years
    If we calculate the 90% and the 98% confidence intervals, these are respectively (6 to 16) and (4 to 18) problems solved in the next three years.
    For five years
    If we calculate the 90% and the 98% confidence intervals, these are respectively (11 to 24) and (9 to 27) problems solved in the next five years.
    For ten years
    If we calculate the 90% and the 98% confidence intervals, these are respectively (23 to 40) and (20 to 43) problems solved in the next five years.
    Ideas for further work
    Do this experiment for other topics besides mathematical theorems and for other methods besides Laplace’s law
    Although I expect that this experiment restricted to mathematical experiments will already be decently informative, it would also be interesting to look at the performance of Laplace’s law for a range of topics.
    It might also be worth it to look at other approaches. In particular, I’d be interested in seeing the same experiment but for “semi-informative priors”—there is no particular reasons why that approach only has to apply to super-speculative areas like AI. So an experiment could look at experts trying to come up with semi-informative priors for events that are testable in the next few years, and this might shed some light into the general method.
    Checking whether the predictions from Laplace’s law come true
    In three, five, and ten years I’ll check the number of conjectures which have been resolved. If that falls outside the 99% confidence interval, I will become more skeptical of using Laplace’s law for arbitrary domains. I’ll then investigate whether Laplace’s law could be rescued in some way, e.g., by using its time-invariant version, by adding some pseudo-counts, or through some other method.
    With pseudo-counts, the idea would be that there would be a number of unique pseudo-counts which would make Laplace output t...
    6 min
  • EA - Karma overrates some topics; resulting issues and potential solutions by Lizka
    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: Karma overrates some topics; resulting issues and potential solutions, published by Lizka on January 30, 2023 on The Effective Altruism Forum.
    TL;DR: Karma overrates “lowest-common-denominator” posts that interest a large fraction of the community, leading to some issues. We list some potential solutions at the bottom.
    Please see the disclaimer at the bottom of the post.
    Posts that interest everyone — or discussions where everyone has an opinion — tend to get a lot more Forum karma (and attention) than niche posts.
    These posts tend to be
    about the EA community
    accessible to everyone, or on topics where everyone has an opinion
    Why does this happen?
    There are different groups with different niche interests, but an overlapping interest in the EA community:
    When a post about the EA community is published, many people might have opinions, and many people feel that they can vote on the post. Most people upvote, so more people voting usually means that a post will get higher karma.
    Similarly, if the topic of the post is something that doesn’t require particular expertise to have an opinion about, lots of people feel like they can weigh in. You can think of these as “lowest-common-denominator posts.” This is related to bike-shedding.
    This leads to some issues
    This misleads people about what the Forum — and the EA community — cares about
    10 of the 10 highest karma posts from 2022 were community posts, even though less than ⅓ of total karma went to community posts.
    When someone is trying to evaluate the quality of the Forum, they often go to the list of top posts and evaluate those. This seems like a very reasonable thing to do, but it's actually giving a very skewed picture of what happens on the Forum.
    Because discussions about the community seem to be so highly valued by Forum readers, people might accidentally start to value community-oriented topics more themselves, and drift away from real-world issues
    Imagine an author posting about some issue with RCTs that’s relevant to their work — they’ll get a bit of engagement, some appreciation, and maybe some questions. Then they write a quick post about the font on the Forum — suddenly everyone has an opinion and they get loads of karma. Unconsciously, they might view this as an indicator that the community values the second post more than the first. If this happens repeatedly or they see this happening, they might shift towards that view themselves if they defer even a bit to the community’s view.
    Now imagine this happening on the scale of the thousands of people who use the Forum; these small updates add up.
    This directs even more attention to community-oriented, low-barrier topics, and away from niche topics and topics that are more complex, which might be more valuable to discuss
    Karma is used for sorting the Frontpage: higher-rated posts stay on the Frontpage for longer. This is useful, as it tends to hide the most irrelevant posts, and generally boosts higher quality content — more people see the better posts.
    But because posts that hit the middle sections in the Venn diagrams above get more karma, they tend to stick around for longer, which then gets them more karma, etc.
    (We didn't try to make this list of issues as exhaustive as possible.)
    Note that karma is not perfect even within a much more specific topic — pretty random factors can affect a Forum post’s karma, and readers aren’t always great at voting, but that is a separate issue. (We might write a post about it later.)
    Solutions we’re considering or exploring
    Create something like a subforum or separate tab for “community opinion” posts, and filter them out from the Frontpage by default
    Or otherwise move in this direction
    Rename “Top” sorting to more clearly indicate what karma actually measures
    We tend to have a somewhat higher bar for sharing “community” posts ...
    5 min
  • AF - Why I hate the "accident vs. misuse" AI x-risk dichotomy (quick thoughts on "structural risk") by David Scott Krueger
    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 I hate the "accident vs. misuse" AI x-risk dichotomy (quick thoughts on "structural risk"), published by David Scott Krueger on January 30, 2023 on The AI Alignment Forum.
    I think the large majority of AI x-risk is "structural". Like climate change.
    Here's a good primer on structural risk (note that structural risk is not a synonym for "not caused by out-of-control AI"):
    I am shocked and amazed and dismayed that more people do not seem to view it this way, even among the AI x-safety community. Heck, even Eliezer's stories of doom are steeped in structural risk (race dynamics, teams rationalizing cutting corners on safety when they should know better, etc.)
    I expect irresponsible, reckless, negligent deployment of AI systems without proper accounting of externalities. I consider this the default for any technology with potential for significant externalities, absent regulation.When something bad happens in such a context, calling it "accident risk" absolves those researching, developing, and/or deploying the technology of responsibility. They should have known better. Some of them almost certainly did. Rationalization, oversight, and misaligned incentives were almost certainly at play. Failing to predict the particular failure mode encountered is no excuse. Having "good intentions" is no excuse.So... it must be misuse then, right? Well, no. Calling it "misuse" suggests that those researching, developing, and/or deploying the technology set out with nefarious purposes and the technology achieved precisely what they intended. But ~nobody wants to destroy the world.
    It's just that most people are somewhat selfish and so are willing to trade some x-risk for a large personal benefit.In summary, saying "accident" makes it sounds like an unpredictable effect, instead of painfully obviously risk that was not taken seriously enough. Saying "misuse" makes it sounds like some supervillian or extremist deliberately destroying the world. While some risks may have something more of a flavor or accident or misuse depending on how obvious the risk was, neither of these pictures gives a remotely accurate picture of the nature of the problem. I think this makes it a harmful meme, and ask that others stop making this distinction (without appropriate caveats), and join me in pointing out how it contributes to a confused and misleading discourse when others do.
    EtA: Many people have responded that "accident" does not connote "unforseen" or "not negligent", etc., and instead it should simply be interpreted as something like "a result that was not deliberately selected for". While it can be used this way, I basically disagree that this is how it is usually used, see below:EtA: as an additional clarification: my main objection is not to the use of "accident" and "misuse", but rather to their use as a dichotomy. Every use of these terms I can recall seeing in writing (other than those that mention structural risk) supports this dichotomy, and it is often made explicitly.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • AF - Call for submissions: “(In)human Values and Artificial Agency”, ALIFE 2023 by thegearstoascension
    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: Call for submissions: “(In)human Values and Artificial Agency”, ALIFE 2023, published by thegearstoascension on January 30, 2023 on The AI Alignment Forum.
    key points:
    Cash prize of $500 for the best presentation.
    Deadline 3 March, 2023.
    Organized by Simon McGregor (University of Sussex), Rory Greig (DeepMind), Chris Buckley (University of Sussex)
    ALIFE 2023 (the 2023 conference on Artificial Life) will feature a Special Session on “(In)human Values and Artificial Agency”. This session focuses on issues at the intersection of AI Safety and Artificial Life. We invite the submission of research papers, or extended abstracts, that deal with related topics.
    We particularly encourage submissions from researchers in the AI Safety community, who might not otherwise have considered submitting to ALIFE 2023.
    EXAMPLES OF A-LIFE RELATED TOPICS
    Here are a few examples of topics that engage with A-Life concerns:
    Abstracted simulation models of complex emergent phenomena
    Concepts such as embodiment, the extended mind, enactivism, sensorimotor contingency theory, or autopoiesis
    Collective behaviour and emergent behaviour
    Fundamental theories of agency or theories of cognition
    Teleological and goal directed behaviour of artificial agents
    Specific instances of adaptive phenomena in biological, social or robotic systems
    Thermodynamic and statistical-mechanical analyses
    Evolutionary, ecological or cybernetic perspectives
    EXAMPLES OF AI SAFETY RELATED TOPICS
    Here are a few examples of topics that engage with AI Safety concerns:
    Assessment of distinctive risks, failure modes or threat models for artificial adaptive systems
    Fundamental theories of agency, theories of cognition or theories of optimization.
    Embedded Agency, formalizations of agent-environment interactions that account for embeddedness, detecting agents and representations of agents’ goals.
    Selection theorems – how selection pressures and training environments determine agent properties.
    Multi-agent cooperation; inferring / learning human values and aggregating preferences.
    Techniques for aligning AI models to human preferences, such as Reinforcement Learning from Human Feedback (RLHF)
    Goal Misgeneralisation – how agent’s goals generalise to new environments
    Mechanistic interpretability of learned / evolved agents (“digital neuroscience”)
    Improving fairness and reducing harm from machine learning models deployed in the real world.
    Loss of human agency from increasing automation
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • EA - Alignment is mostly about making cognition aimable at all by So8res
    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: Alignment is mostly about making cognition aimable at all, published by So8res on January 30, 2023 on The Effective Altruism Forum.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min
  • LW - What I mean by "alignment is in large part about making cognition aimable at all" by So8res
    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 So8res on January 30, 2023 on LessWrong.
    (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 to the misunderstan...
    4 min
  • EA - Regulatory inquiry into Effective Ventures Foundation UK by HowieL
    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: Regulatory inquiry into Effective Ventures Foundation UK, published by HowieL on January 30, 2023 on The Effective Altruism Forum.
    Earlier today, the Charity Commission for England and Wales announced a statutory inquiry into Effective Ventures Foundation UK (EVF UK). We issued a short statement acknowledging the inquiry. As the recently appointed Interim CEO of EVF UK, I also wanted to say a bit more to the community.
    As you might imagine, writing this post feels a bit tough for a bunch of reasons. It’s a sensitive topic, there are lots of complicated legal issues to consider, and it’s generally a bit weird to write publicly about an agency that’s in the middle of investigating you (it feels a little like talking about someone in the third person without acknowledging that they’re sitting at the dinner table right next to you). One thing I really want to do is avoid speaking for the Charity Commission, which makes me hesitant to speculate too much about why exactly they chose to begin this inquiry, etc.
    But I’m going to do my best to figure out what’s helpful to say here.
    Summary
    Effective Ventures Foundation UK (EVF UK) is the UK charity which, together with Effective Ventures Foundation US (EVF US), acts as host and fiscal sponsor for a wide range of projects including CEA, 80,000 Hours, and others.
    Following the FTX collapse, the Charity Commission for England and Wales has opened an inquiry into EVF UK’s financial situation and its governance and administration by its trustees.
    We were not particularly surprised by the Commission’s decision to open an inquiry. As it notes, some FTX-linked entities (such as the FTX Foundation) and related individuals were major donors to EVF UK, and there was the potential for a variety of conflicts of interest with respect to governance (e.g., two of EVF UK’s board members were affiliated with FTX Foundation). Especially given the scale and profile of the FTX collapse, it makes sense that the Commission wants extra visibility into whether we’re appropriately protecting our property, managing potential conflicts of interest, and generally complying with UK charity law.
    The UK board and executive team are cooperating in full and the Commission has confirmed that the trustees have fulfilled their duties by filing a serious incident report.
    Sharing updates in the midst of an ongoing inquiry is legally complicated, so there will probably be some limitations to the updates we’ll provide before the inquiry is complete. But we’ll do our best to share whatever (if anything) is legally sensible.
    What is Effective Ventures Foundation UK?
    For context, EVF UK is the UK charity which, together with EVF US, acts as host and fiscal sponsor for a wide range of projects, including CEA, 80,000 Hours and others. Everything in this post is only referring to the UK entity since that’s the entity which is regulated by the Charity Commission. (You can read more about the structure and role of EVF UK here.)
    What is a statutory inquiry?
    The Charity Commission is a regulator whose responsibilities include:
    Preventing mismanagement and/or misconduct by charities;
    Promoting compliance with charity law;
    Protecting charities’ property, beneficiaries, and work; and
    Safeguarding the public’s trust and confidence in charities.
    A statutory inquiry is one tool the Commission has for establishing facts or collecting evidence related to these responsibilities. Opening an inquiry gives the Commission information gathering powers; they can require a charity to send relevant information or documents and they can require individuals to answer questions. The Commission also has wide-ranging powers for protecting charities and ensuring compliance. Outcomes can range from taking no action or issuing a warning to appointing an interim manager to manage the charity or even r...
    7 min