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: Counterbalancing “It's time for EA leadership to pull the short-timelines fire alarm”, published by Matthew Barnett on April 9, 2022 on LessWrong.

Recently, a post claimed,

it seems very possible (>30%) that we are now in the crunch-time section of a short-timelines world, and that we have 3-7 years until Moore's law and organizational prioritization put these systems at extremely dangerous levels of capability.

We (Tamay Besiroglu and I) think this claim is strongly overstated, and disagree with the suggestion that “It's time for EA leadership to pull the short-timelines fire alarm.” This post received a fair amount of attention, and we are concerned about a view of the type expounded in the post causing EA leadership to try something hasty and ill-considered.

To counterbalance this view, we express our disagreement with the post. To substantiate and make concrete our disagreement, we are offering to bet up to $1000 against the idea that we are in the “crunch-time section of a short-timelines”.

In particular, we are willing to bet at at 1:1 odds that no more than one of the following events will occur by 2026-01-01, or alternatively, 3:1 odds (in our favor) that no more than one of the following events will occur by 2030-01-01.

A model/ensemble of models achieves >80% on all tasks in the MMLU benchmark

A credible estimate reveals that an AI lab deployed EITHER >10^30 FLOPs OR hardware that would cost $1bn if purchased through competitive cloud computing vendors at the time on a training run to develop a single ML model (excluding autonomous driving efforts)

A model/ensemble of models will achieve >90% on the MATH dataset using a no-calculator rule

A model/ensemble of models achieves >80% top-1 strict accuracy on competition-level problems on the APPS benchmark

A gold medal for the IMO Grand Challenge (conditional on it being clear that the questions were not in the training set)

A robot that can, from beginning to end, reliably wash dishes, take them out of an ordinary dishwasher and stack them into a cabinet, without breaking any dishes, and at a comparable speed to humans (<120% the average time)

Tesla’s full-self-driving capability makes fewer than one major mistake per 100,000 miles

Since we recognize that betting incentives can be weak over long time-horizons, we are also offering the option of employing Tamay’s recently described betting procedure in which we would enter a series of repeated 2-year contracts until the resolution date.

Specific criteria for bet resolution

For each task listed above, we offer the following concrete resolution criteria.

TaskFull detailsA model/ensemble of models achieves >80% on all tasks in the MMLU benchmarkA public document reveals that a model or ensemble of models had an average accuracy score of more than 80.0% on all tasks listed in the Measuring Massive Multitask Language Understanding benchmark by Dan Hendrycks et al.A credible estimate reveals that an AI lab deployed EITHER >10^30 FLOPs OR hardware that would cost $1bn if purchased through competitive cloud computing vendors at the time on a training run to develop a single ML model (excluding autonomous driving efforts)If there is a dispute, our counterparty will offer an example that they believe may trigger this condition to resolve positively. Then, we will use the method described in this post to estimate the number of FLOPs used for a given training run.

We will then find the competitive rates by taking the minimum of the most recent prices found on Google Cloud, Microsoft Azure, and Amazon AWS, or the equivalent services corresponding to each of those companies if their names change in the future. We will determine the rate corresponding to the processing unit that would be capable of training the model in the paper. If the training run was completed than 2 years...

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: Four categories of effective altruism critiques, published by jtm on April 9, 2022 on The Effective Altruism Forum.

When discussing critiques of effective altruism, I often find that lots of different kinds of critiques are lumped together, which can lead to people talking past each other.

Here's an attempt at categorising different things one might focus on – I'm not sure the categories are exhaustive and I also welcome suggestions for refined categorisations.

Four categories of critiques

Normative assumptions & moral values

This includes the philosophical underpinnings of effective altruism, including (but not limited to) claims about what is valuable and what people ought to do.

Empirical assumptions & theories of change

This includes beliefs about empirical questions like "How likely is human extinction this century?" or "What is the most cost-effective way of improving human health?" It also includes worldviews or theories of change, like "Positive societal change is more likely to be gradual than radical" or "Positive societal change is more likely to be driven by elites than by large movements."

Institutions & organisations in the effective altruism social movement

This includes both institutions in the broad sense of the word, such as how hiring practices or how strategic decisions for the social movement are made. It also includes questions pertaining to specific organisations, such as how transparent a particular funder is or how much an organisation achieved in the past year.

Social norms & practices in the effective altruism community

This includes things about the community aspect of effective altruism, i.e. the ways that people in the movement interact not just as colleagues, but also as friends, partners, housemates, and subscribers to a shared ideology. Topics to explore could include how people interact with those outside the movement, the effects of talking about money in a certain way, or the causes and effects of demographic and intellectual homogeneity.

Examples in each category

I also include below some examples of existing work in each category, though each list is thoroughly non-exhaustive. Also, I'm sure that some of the examples could fit into more than one category; I tried placing them in the category they seemed to represent the most.For several of these examples, I haven't read them in their entirety; some I've only barely skimmed. For each, I'm sure there are things I don't agree with, just as there is something in every one of these that I do agree with.

You'll note that these aren't sweeping critiques of effective altruism or longtermism as a whole, they are rather critical assessments of specific aspects (although those would also be interesting). I personally find that while both "we should use evidence and reason to do the most good" (effective altruism) and "positively influencing the longterm future is a key moral priority of our time" (longtermism) are rather unobjectionable as general claims, the way they are practised could be better, hence the value of critiques.

I would love to see more examples – please post your favourites in the comments!

Normative assumptions & moral values

Xuan: AI Alignment, Philosophical Pluralism, and the Relevance of Non-Western Philosophy

Unruh: Constraining longtermism (Working paper, likely available upon request)

Mogensen: The only ethical argument for positive 𝛿?

Russell: On two arguments for Fanaticism

Joe_Carlsmith: On infinite ethics

Empirical assumptions & theories of change

Garfinkel: How sure are we about this AI stuff?

Srinivasan: Stop the Robot Apocalypse

Denise_Melchin: Why I am probably not a longtermist

weeatquince: The case of the missing cause prioritisation research

Singer: The Hinge of History

Weyl: Hear This Idea interview (second half)

Brian Tomasik: Why Charities Usually Don't...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Takeaways From 3 Years Working In Machine Learning, published by George on April 8, 2022 on LessWrong.

Disclaimer: Feeling so-and-so about posting this on LW, but given how many people here work in ML or adjacent fields I might as well.

After 3 years working on automl at Mindsdb, I quit; I won’t be working on anything ml-related in a professional capacity, at least for a short while.

I am in the uncanny valley of finally understanding what I don’t know, and maybe even having a feel for what nobody knows.

I might as well write a summary of my takeaways, in hopes of them being useful to someone, or maybe just as a ritualistic gesture of moving on, a conceptual-space funeral.

Please don’t take this as an “expert summary”, there are dozens of thousands of people more suited than myself for doing that. Instead, think of it as a piece of outsider art, takeaways from someone that took an unusually deep dive into the zeitgeist without becoming part of “the community”.

i - What Is The Role Of Research?

The role of research into machine learning is, half a decade after first pondering it, still a mystery to me.

Most scientific sub-fields (the real ones) can claim they are a dual process of theory-building and data-gathering. Even if the standard model ends up being replaced, the data leading to its replacement will include the same set of experimental observations that lead to it being built. Even if the way we conceptualize the structure of DNA and the idea of genetics changes to better fit many-tissued eukaryotes, the observations these new concepts will have to fit shall remain unchanged.

In more theoretical areas, such as those revolving around the terms mathematics and computer science, the gains are at an almost purely-conceptual level, but they hold fast because the concepts are so foundational they seem unlikely to be replaced. Some alien species living inside deep space that we could barely understand as “life”, or some space-faring empire or languageless primate tribes that we could barely classify as “humans” might build impressive conceptual machinery without the alonzo-church thesis or even without concepts such as function, number, discreteness or continuity. But as long as those things hold true, a lot of old ideas in mathematics and computer science remain at most inefficiently phrased, but broadly valid.

But machine learning is at an odd intersection. Even if we grant it has as much theoretical rigor as, say, physics, it still lacks the timeless experimental observations. This is not for lack of any virtue in participants, it’s just that the object of study is a moving target, rather than a concrete reality. At most one can make timeless claims like:

Given it’s possible to get on of ImageNet

Given dataset and using whole-dataset leave-one-out CV we can achieve x/y/z state of the art values for certain accuracy functions.

Given environment that allows a programmable agent to interact with it, we can reach x/y/z points on a cycles/punishment/time/observations & reward/knowledge/understanding matrix.

But these are not the kind of claims theory revolves itself around, the easily verifiable and timeless takeaways from machine learning are theoretically uninteresting. At most one can say they place a lower bound on the performance of digital hardware on certain human-specific tasks.

Nor is it backed by strong theoretical guarantees, at least not without reaching “increased milk production if you assume cows are perfect spheres in a void” type absurdity. There are “small” theoretical guarantees that can help us experiment more broadly (e.g. proving differentiability within certain bounds). There are “impressive” theoretical guarantees that hold under idealized conditions which can at most point towards potential paths to experiment.

You might assume that this leads t...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Elicit: Language Models as Research Assistants, published by Andreas Stuhlmüller on April 9, 2022 on The AI Alignment Forum.

Ought is an applied machine learning lab. We’re building Elicit, the AI research assistant. Our mission is to automate and scale open-ended reasoning. To get there, we train language models by supervising reasoning processes, not outcomes. This is better for reasoning capabilities in the short run and better for alignment in the long run.

In this post, we review the progress we’ve made over the last year and lay out our plan for Elicit. We'd love to get feedback on how to make Elicit more useful for LW and to get thoughts on our plans more generally. To make this easier, we've erred on the side of giving detail even where we know it will change.

Progress in 2021:

We built Elicit to support researchers because high-quality research is a bottleneck to important progress and because researchers care about good reasoning processes.

We identified some building blocks of research (e.g. search, summarization, classification), operationalized them as language model tasks, and connected them in the Elicit literature review workflow.

On the infrastructure side, we built a streaming task execution engine for running compositions of language model tasks This engine is supporting the literature review workflow in production.

About 1,500 people use Elicit every month.

Roadmap for 2022+:

We expand literature review to digest the full text of papers, extract evidence, judge methodological robustness, and help researchers do deeper evaluations by decomposing questions like “What are the assumptions behind this experimental result?”

After literature review, we add other research workflows, e.g. evaluating project directions, decomposing research questions, and augmented reading.

To support these workflows, we refine the primitive tasks through verifier models and human feedback, and expand our infrastructure for running complex task pipelines, quickly adding new tasks, and efficiently gathering human data.

Over time, Elicit becomes a general-purpose reasoning assistant, transforming any task involving evidence, arguments, plans and decisions.

How we think about success

Our mission is to automate and scale open-ended reasoning. If we can improve the world’s ability to reason, we’ll unlock positive impact across many domains including AI governance & alignment, psychological well-being, economic development, and climate change.

As AI advances, the raw cognitive capabilities of the world will increase. The goal of our work is to channel this growth toward good reasoning. We want AI to be more helpful for qualitative research, long-term forecasting, planning, and decision-making than for persuasion, keeping people engaged, and military robotics.

Good reasoning is as much about process as it is about outcomes. In fact, outcomes are unavailable if we’re reasoning about the long term. So we’re generally not training machine learning models end-to-end using outcome data, but building Elicit compositionally and inspired by human processes.

In the short term, supervising process is necessary for AI to help with tasks where it’s difficult to evaluate the work from results alone. In the long term, process-based systems can avoid alignment risks introduced by end-to-end training.

Success for us looks like this:

Elicit radically increases the amount of good reasoning in the world.

For experts, Elicit pushes the frontier forward.

For non-experts, Elicit makes good reasoning more affordable. People who don’t have the tools, expertise, time, or mental energy to make well-reasoned decisions on their own can do so with Elicit.

Elicit is a scalable ML system based on human-understandable task decompositions, with supervision of process, not outcomes. This expands our collective und...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: AMA Conjecture, A New Alignment Startup, published by Adam Shimi on April 9, 2022 on The AI Alignment Forum.

Conjecture is a new alignment startup founded by Connor Leahy, Sid Black and Gabriel Alfour, which aims to scale alignment research. We have VC backing from, among others, Nat Friedman, Daniel Gross, Patrick and John Collison, Arthur Breitman, Andrej Karpathy, and Sam Bankman-Fried. Our founders and early staff are mostly EleutherAI alumni and previously independent researchers like Adam Shimi. We are located in London.

As described in our announcement post, we are running an AMA this week-end, from Today (Saturday 9th April) to Sunday 10th of April. We will answer any question asked before the end of Sunday Anywhere on Earth. We might answer later questions, but no guarantees.

If you asked question on our announcement post, we would prefer that you repost them here if possible. Thanks!

Looking forward to your questions!

Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: A tough career decision, published by PabloAMC on April 9, 2022 on The Effective Altruism Forum.

In this post, I summarize a tough career decision I have had to take over the last few weeks.

Setting the stage

I am a few months from finishing my Ph.D. in quantum algorithms. During these 4 years, I have become quite involved in Effective Altruism: I attended two EA Globals, facilitated a couple of virtual intro fellowships, and helped organize the EA group in Madrid. Given my background, I have also felt closer to AI Safety than any other cause area. As such, I have also been involved in the AI Safety community, by participating in two AI Safety Camps, and as a facilitator in some intro fellowships. I even did a summer internship in AI Safety with José Hernández Orallo last summer which led to a rather lucky AAAI publication.

My Ph.D. has also gone well. While it started a bit dubitative and was unable to get anything published for the first two years, at that point I got my first two publications and over the last two, I have been able to do well. It is not a superstar Ph.D. but I believe I have learned enough to make contributions to the field that actually get used, which is harder than it looks. In fact, I feel happy that thanks to my last article, one rather serious quantum startup contacted me to collaborate, and this led to another quite high-quality paper.

The options

Since I am finishing my Ph.D., I had to plan my next step. The first obvious choice was to apply for funding to do AI Safety. I cold emailed Victor Veitch, who I found through the super useful Future of Life AI Safety Community, and he was happy to take me as long as I could work more or less independently.

The reason why I opted for applying with Victor was that my research style is more about knowing well the tools I am using, not to the level of a pure mathematician, but to the level where techniques are known and can be used. Additionally, I think causality is cool, and being able to apply it in large language models is rather remarkable. I am also a big fan of Ryan Carey, who works in causality and is one of the people in the community who has helped me the most. I am really grateful to him. Apart from the Future of Life postdoc program, I also applied to the EA Long Term Future fund, and Open Philanthropy, with this proposal. Out of this, the EA Long Term Future fund accepted funding me, most likely on the basis of a career change, while FLI declined based on the proposal and probably an interview that while I prepared was not able to use to explain well why I think this perspective could be useful. This came a bit of a disappointing result, to be honest.

Open Philanthropy, on the other hand, is behind schedule, and I don't know their answer yet.

The alternative was to pursue, at least for now, a more standard research topic. I applied to IBM, Google, Amazon, and three startups: Zapata, PsiQuantum, and Xanadu. With the latter, I have been working, so it was not really an application. I never heard back from any of the big companies, but got offers from the three startups. To be fair I also applied to a couple of non-EA AI postdocs with the hope of getting them, but they were a very long shot. For a bit of context though, PsiQuantum is a very serious contender in building a photonic quantum computer and error correction and has really deep pockets, while Xanadu is probably a bit behind them, but it's also quite good and has a bit more focus on ML.

The situation and conditioning factors.

Perhaps the main conditioning factor on all of this was the fact that my girlfriend is really a very important part of my life, and responsible for my happiness. A friend of mine called this the unsolvable two-body problem 😛. In particular, I think that we both would like to have a smooth life. She is currentl...

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: A method of writing content easily with little anxiety, published by jessicata on April 8, 2022 on LessWrong.

Hello, I have today been informed about a method for writing content that is relatively easy and painless. The idea is that by default a human will when reading predict what word is going to come next. Studies show that humans can predict what letter is going to come next and get pretty good entropy scores. That means that even the process of reading involves generating plausible next words.

And so it is possible to similarly do the same when writing. When writing a sequence of words, after the end there will be a feeling that some word will come next. It is unexpected and abrupt for the sentence to immediately end, so there is a feeling of continuation, that something will come after. One can while writing simply look at the space right to the right of the text that has already been written and there will be a feeling there about what word will come next.

Do not worry if this is difficult and slow at first. A practice exercise is to write a few words and then take some time to get a feeling of what word is going to come next. With meditation it is possible to notice experiential phenomena that are in one's experience and sometimes have a spatial location. Well, maybe someone who knows how to find these phenomena is going to be able to find some sense of "next word" ness when focusing on the end of a sentence fragment. That will tell them what word they think is going to come next.

It is possible to simply, when noticing what word is going to come next, type that word. That is in fact what I am doing right now and what I have been doing for this entire post. It is rather easy and painless to do. It is not anxiety-provoking because it is a meditative, instruction-following exercise where I simply get a sense of what word will come next and type it. I am not responsible for the output of my text predictive model. I am not to blame. I am not unworthy.

This shortcuts various anxiety loops that are possible when writing in a more traditional manner. At that point, while there are many sentence continuations that are invisibly generated, all are filtered out because it would feel somehow bad to have written that sentence. Maybe the sentence would be cringeworthy or stylistically bad or offensive or embarrassing or I-don't-know-what, it might just be off, and someone with enough of an aesthetic taste might simply be repulsed by the idea of actually writing out that sentence. There is no filter this way, because the sense of what word is going to come next is always present even when reading text, and so it is not specific to writing and doesn't indicate that the person having such a predict is, themselves, responsible for that text.

I know this is a somewhat silly way of thinking about moral responsibility, but it is emotionally compelling enough that it can increase writing output by a lot. And output is not always good, quality matters too. It would be to be susceptible to Goodhart's law to say that writing more words is always better. However, writer's block is a pain, it is really bad if it isn't possible to write anything at all, or to only be able to write very short sentences and paragraphs. That way it is possible to go months without making any progress on a writing project.

I prefer to have something written even if it is low-quality. That way, it is possible to edit or re-try later. That will increase the level of quality. And I know that that seems like a confident statement, that it will increase the quality, not that it might. The difference is that right now I do not care about seeming overconfident, since I am simply reporting on what word I consider most likely, and not worrying that the word I think is most likely to come next is not actually the m...

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: Open Philanthropy is hiring more people than ever this year, published by Anya Hunt on April 8, 2022 on The Effective Altruism Forum.

At Open Philanthropy, our mission is to help others as much as we can with the resources available to us.

We gave away over $400 million in 2021. We aim to double that number this year, and triple it by 2025.

As a result, we hope to grow our staff to roughly 80 people this year, and to keep growing at a rapid pace for years to come. That means we’re hiring over 30 people this year, and probably well over 100 in the next four years. (It took us nearly 7 years to get to our current size, so this is a major shift for us!)

With this post, we want to answer some common questions about working at Open Phil, and to make the following points:

If you’ve ever wanted to work for Open Phil, you’ll have more chances to apply this year than ever before.

If you know someone else who may be a good fit, you should refer them to us (and maybe send them this post). If you refer someone, and we hire them, we’ll pay you $5000.

You can see our open positions here, and make a referral here.

What growth means for Open Phil

We are the largest funder in several cause areas, but we still have ample room to grow. A few examples of ways we’ve worked towards increasing our giving over the last year:

We added three new focus areas — South Asian air quality, global aid advocacy, and EA movement building (focused on global health and wellbeing).

We hired new program staff to grow many existing areas, including farm animal welfare, biosecurity, global health R&D, and longtermist EA movement building.

We also hired several new staff who focus on cause prioritization, to help us figure out how to expand to new areas.

We launched a $150 million regranting challenge to support highly effective grantmakers who are doing better work than we could in their respective specialties.

Our program officer for biosecurity published a list of specific biosecurity projects on which we’d like to see people coordinate.

Note that many of these items represent ways to scale our giving in entirely new categories. By 2025, we’ll likely be working on things we haven’t even imagined yet. (We’re hoping to hire now partly so that our new staff can help us figure out what those things should be.)

Open roles

Right now, we have several open roles in Operations and Farm Animal Welfare:

A new Director of Operations to help us scale our operations and ensure we’re able to execute on our most important goals.

Three new roles to help our Farm Animal Welfare program expand its giving, focused on research & modeling, generalist grantmaking, and science.

Recruiters to help us hit these growth targets!

If none of these excite you, we’ll be posting lots of additional roles over the course of the year. Many of those will probably end up on the Forum — you can subscribe to the “Open Philanthropy” tag to see those.

You can always apply, and you can pitch us on roles that don’t exist yet

Needs often emerge before we have time to write job descriptions and run a search to fill them. If you don’t currently see your ideal job open, we’d encourage you to fill out our general application (it’s quick!) and let us know what you’d like to do. We are currently reading these ~weekly and will reach out if we’d like to talk.

What we’re looking for

We can’t generalize too much here — we try to make our job descriptions very specific. But at a high level, we’re currently interested in:

People who have experience in management. Our need for managers is a major bottleneck to scaling up our work. If this is a strength of yours (especially for managing large operations functions — we’re currently hiring for a Director of Operations and a manager for our business operations team ops), we strongly recommend you apply.

People who a...

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: Convincing All Capability Researchers, published by Logan Riggs on April 8, 2022 on LessWrong.

This post is heavily based off this excellent comment (the author's name is not relevant)

Give the world's thousand most respected AI researchers $1M each to spend 3 months working on AI alignment, with an extra $100M if by the end they can propose a solution alignment researchers can't shoot down. I promise you that other than like 20 industry researchers who are paid silly amounts, every one of them would take the million. They probably won't make any progress, but from then on when others ask them whether they think alignment is a real unsolved problem, they will be way more likely to say yes. That only costs you a billion dollars! I literally think I could get someone reading this the money to do this (at least at an initially moderate scale) - all it needs is a competent person to step up.

Usually people object to paying people large sums of money to work on alignment because they don't expect them to produce any good work (mostly because it's very hard to specify alignment, see below). This is a feature, not a bug.

Being able to say "the 1000 smartest people working in AI couldn't make headway in Alignment in 3 months even when they were paid $1 million and a solution would be awarded $100 million" is very good for persuading existing researchers that this is a very hard problem.

Would this really go a long way to convincing those at major AI labs that alignment is hard? We could actually ask people working at these places if there were no progress made in those 3 months if it would change their minds.

Another problem is looooong timelines. They can agree it's hard, but their timelines may be too far to matter to work on it now. I could think of a couple counter-arguments that may work (but it would better to actually test them with real people)

1. To predict AGI (or just AI that has the capability of destroying the world), you could argue "It needs these 10 capabilities, and I predict it will get each capabilities at these different time periods". If someone actually did that and was historically correct, even within 5-10 years, I'd be impressed and trust their opinion. Have you successfully predicted AI capabilities of the past 10 years?

possibly mention most recent advances (Transformers, some RL, Neural Nets in 2010's)

2. If there's a 10% chance that it happens during you or your kids lifetime, then that's extremely important. For example, if I thought there was a 10% chance that I died taking the train home tonight, I wouldn't take the train. Even with a 0.1%, I wouldn't.

Though, generally having a set of people working on apologetics for both US & Chinese AI Researchers is potentially extremely high-impact. If you're interested and might be able to donate , DM me for a call.

Next Steps to seriously consider this proposal

Specifying "Solving Alignment"

Part of the problem of alignment is we don't know the correct framework to specify it. The quoted text suggests the criteria for a solution as "a solution alignment researchers can't shoot down.", which side-steps this issue; however, specifying the problem in as fine-grained detail would be extremely useful for communicating to these researchers.

One failure mode would be them taking the money, not get work done, and then argue the problem wasn't specified enough to make any meaningful progress which limits how persuasive this stunt could be. Documents like ELK are more useful specifications that captures the problem to various degrees, and I wish we had more problems like that.

Listing 1000 AI researcher intellectuals

The initial idea is to hire 1000 best AI researchers to work on the problem, not because we expect them to solve it, but by all of them failing

here are a few different proxy's we can use, such as c...

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: The Bearable Brightness of Wellbeing: The life of an HLI researcher, published by JoelMcGuire on April 8, 2022 on The Effective Altruism Forum.

0. What is this?

This is a "write about your job" piece that has a secondary aim of giving potential hires a sense of what working at HLI as a researcher is like.

1. Who am I?

I quit an economics PhD programme at a state school in the USA to join the Happier Lives Institute (HLI) as a researcher a little more than two years ago. It was one of the better decisions of my life. After joining HLI, I gained much-needed research experience by working on important questions; I began to engage deeply with effective altruism (EA) and quelled the fear of not finding a way to positively impact the world.

I had a Master's degree before starting the PhD. But I’d rushed through both degrees in 3½ years. This left a blank space where practical research (or any work) experience would typically go. In my PhD, I had some vague hope of how I’d pivot my research into an EA-aligned area, but this was a hope, not a plan. I didn’t know how I’d do research that mattered.

Thankfully, HLI offered me an excellent opportunity to work on worthwhile topics. I escaped the PhD programme and my gnawing doubt by joining HLI as an empirical researcher (and their first employee!). If you’re concerned about your lack of elite pedigree (like I was), then trying to start at a young EA organisation seems like a plausible path towards acquiring a much-coveted EA job.

2. Why do I do this job?

I find it thrilling to work on the edge of a research field and potentially expand its borders. I love the intoxication of discovery, the flow of coding, and the challenge of designing a beautiful visualisation. But attempting pioneering research is tough. The fields I work in are riddled with methodological issues that have no established solutions. The way is undiscovered. The map is mostly blank.

Studying wellbeing feels like chasing the holy grail. Wellbeing is good, and if we can measure it better, we can do good better. Plus, I get a dopamine hit every time someone asks me what I do, and I get to say I’m a happiness researcher. Never mind that they almost inevitably ask about how one acquires personal happiness (which I think is common sense) when I’m in the business of global happiness.

HLI, even though it’s growing, is still small. That means I have a sizeable say in setting the research agenda. This makes me feel like I have skin in the game. It can be a lot of pressure, but it’s deeply fulfilling to feel like I’m doing the most good I can reasonably do.

3. Some details on how I divide my time

Below is a chart that shows how I split my time between distinct tasks in my role as a researcher. As you can see, I spend almost 70% of my time doing research. I think this means something organizationally is working.

Figure 1: The distribution of Joel’s time across tasks

Research

About 75% of my research time is dedicated to my primary work of analysing the cost-effectiveness analyses (CEAs) of different interventions. I spend the rest doing general research tasks like reviewing studies or giving feedback on ideas.

For my work on CEAs, I spend about 50% of the time writing or refining R scripts, 35% writing and revising reports, 15% adding data to spreadsheets, and 5% searching for and reading studies.

Meetings and other tasks

The nature of my work varies throughout the year. Last summer, I spent a lower bound of 29 hours organising, hiring and managing HLI’s inaugural cohort of Summer Research Fellows. I was pretty involved in hiring the second HLI empirical researcher, the excellent Samuel Dupret. At the end of the calendar year/beginning of the new one, we typically spend more time reviewing and outlining our strategy. When reports are published, I spend some time replying to...