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 Librarian: we want to help answer your EA questions!, published by calebp on January 17, 2022 on The Effective Altruism Forum.

TLDR;

We want to help answer your questions about EA!. ‘Dumb’ questions or questions that you would usually be embarrassed to ask are especially encouraged.

You can ask questions using the forum question feature with the EA Librarian tag, using this form or via the discussion channel in certain fellowship groups.

Feel encouraged to ask questions without reading the rest of this post!

Background

The community health team at CEA is excited to launch the EA Librarian on the EA forum. This is part of the team’s work on community epistemic health.

One way in which the EA movement could fall short of its potential is if epistemic norms degrade significantly within the community. It’s hard to get a sense of exactly how large an issue this but there is at least some evidence to suggest that this may be happening already (e.g. deferral problems previously identified by the EA leaders’ forum).

CEA’s work in this area is quite exploratory and we are looking to pilot programmes like this one in an effort to retain and improve the epistemic norms of the EA community. The team expects to try out different projects and iterate over time. If you have ideas or feedback, please have a low bar for sending them through! (You can reach Nicole, the community health team manager, at nicole[dot]ross[at]centreforeffectivealtruism.org)

I’d like to thank Nicole Ross, Will Saunter and Hannah Erlebach for providing feedback on earlier drafts of this post, all errors are my own.

Overview

As the name suggests, the EA librarian will act as a librarian for the EA community. Users can submit questions related to EA via the forum, privately using this form and our team of librarians will do their best to research and answer your questions, pointing you towards existing resources where useful. We think that this will not only be useful for people new to EA but also for more experienced community members. This is the second iteration of this programme after running a similar service on slack for a remote fellowship.

We think that this service might add value in the following ways, and are excited to experiment to see if this seems correct:

Increase awareness of existing resources in the EA community.

The EA community (and other related groups) have produced a large amount of high-quality literature. EAs exploring unfamiliar areas can often save a substantial amount of time by asking people broadly familiar with EA literature relative to an independent investigation.

Celebrate and reinforce a norm of investigating confusions (particularly asking ‘dumb’ questions).

My impression is that the perceived bar to asking questions online for new members is much higher than the bar ‘should’ be.

Feedback from the previous iteration of the programme suggested that, new members can be overly concerned about wasting other people’s time and reputational costs of asking basic questions.

I also think that asking questions is one of the quickest ways to advance your understanding of EA concepts so lowering the bar or nudging people to ask questions could be very valuable.

Provide clear demonstrations of working through questions using good reasoning processes to help users investigate similar questions independently in the future.

One way in which people develop their reasoning is through conversation with others. Various staff members at EA organisations have mentioned that conversations with people with good judgment were helpful in improving their own judgment.

I think that outside of a handful of organisations, it can be difficult to find a high concentration of people with good judgment, who can explain their reasoning processes well enough for others to understand and imitate them.

We ho...

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 Application is not the Applicant, published by JJ Hepburn on January 17, 2022 on The Effective Altruism Forum.

Remember when dealing with any application process as either an applicant or a reviewer that the application is not the applicant. It’s like something to do with maps and territories. Applications for jobs, schools, funding, grants, internships, events, fellowships or even the Bahamas all have this same problem.

For rejections in particular it’s important to remember that it is not you that has been rejected, but your application. We hope that application processes are filtering for the right things but there are many application processes that are a skill in itself to be mastered.

When we were hiring for a position last year we gave everyone that didn’t make it to the interview stage an opportunity to talk with us and get feedback on their application. There were definitely people who if their applications better represented who they actually are, would have easily been among the top few candidates. Of course the opposite is true as well. But we are quite familiar with seeing applicants turn out to be not as good when interviewed.

Accounting for some of this, I try to word rejections around the application, not the applicant. “Your application was not.” rather than “You were not.”

The incentives are for high precision not high recall. It is acceptable to reject a good candidate (false negative) but not to accept a bad candidate (false positive). So there are a lot of good candidates that get rejected. Reviewers need to be accurate about those they accept but not those they reject, so a rejection has more potential to be a bad signal of the applicants fit. It might be a good signal that the application is bad though.

Review processes need to be efficient more than they are effective. An organisations growth in reviewing capacity almost always lags behind it's growth in applicants. It's never enough and with false negatives being acceptable, reviewers are often looking for reasons to reject, more than they are to accept. Particularly in early stages.

When applying for things you should always assume the reviewer is incredibly lazy. And, at least with respect to any one application, they kind of are. I have often rejected people because they haven't answered a question properly enough for me to be able to judge them in any way. Reviewers don’t know anything about you other than what is written on the page and are unlikely to go looking something up if it’s not clear. Don’t assume you will get a chance to explain things in more detail in an interview.

I have seen plenty of instances of either people or projects being wildly different or misrepresented by what is in an application. It is far more common for the application to be underwhelming than an exaggeration. If you get a rejection for something, go back to your application and see what was actually rejected.

People will spend years studying hard in a particular field only to start getting really selective about wasting their time. You’ve achieved mastery in your field but this won’t count for much if you don’t have any skills in application processes. There are some really low hanging fruits that would massively improve a lot of peoples applications.

Hot tip: If you want to learn how to write a good application or resume use resources for how to review applications, resumes or interviews. Looking at the process from the reviewers perspective quickly allows you to see things you would have never noticed from resources for how to write a good application. As someone pointed out to me recently it is like a generative adversarial network (GAN)

When it comes to grants and funding it’s really important to remember that when someone tells you their project didn’t get funded, their application may not describe...

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 are some artworks relevant to EA?, published by Lizka on January 17, 2022 on The Effective Altruism Forum.

[Question motivated by personal interest: I’m asking for myself, not for my employer.]

I’m interested in a wide variety of media; the two most important factors here seem to be relevance to EA and the inherent coolness of the artwork. We could use this an opportunity to crowdsource examples and kick off a conversation about art and EA if there’s one that “wants to happen.”

Some things to consider:

Creative data visualizations

E.g. New infographic based on "The Precipice". any feedback?

Animations or videos

E.g. Kurzgesagt stuff like Worst Nuclear Accidents in History or The Egg - A Short Story

Comics or graphic novels

E.g. Saturday Morning Breakfast Cereal: EV, xkcd: Bigger Problem, or xkcd: Masks

AI-generated art

(you could make the case that it’s a useful illustration of growing capabilities of AI)

E.g. DALL·E: Creating Images from Text, or this AI-generated art on Twitter

Symbolic art

E.g. Proposed Longtermist Flag (see the comments)

Especially cool scientific art

E.g. penicillin art, or pretty scientific illustrations like this and this (although they might be less relevant)

Relevant photography

E.g. Peaceful protester/armed police pictures, or these photos of chickens (potentially less relevant)

Fiction or other writing

We’ve had a variety of places to collect instances of creative writing, but I see no reason to exclude it if you think of something

Musical works (although there’s already a question on the topic)

Board and video games?

E.g. the paperclips clicker game (warning: don’t click if you’re easily addicted to something and don’t have a few hours)

Random EA-related art

EA Art: Neural Style Transfer Portraits

Inspirational art

Visual art that inspires compassion towards the very neglected

Insightful art

...including parodies and satires

If you don’t think the relevance of your answer(s) is obvious, you might consider explaining it.

If you’ve made some art you think is relevant, I’d really love to see it. I realize that it’s quite intimidating to post art (at least to me), but even if you don’t want to share publicly, I’d really appreciate any private sharing you’re comfortable with. (But I do encourage you to share it on the Forum if you can!)

Some relevant tags and posts:

Please use art to convey EA!

Effective altruism art and fiction (tag)

What music do you find most inspires you to use your resources (effectively) to help others?

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: FLI launches Worldbuilding Contest with $100,000 in prizes, published by ggilgallon on January 17, 2022 on The Effective Altruism Forum.

The Future of Life Institute (FLI) invites individuals and teams to compete for a prize purse worth $100,000+ by designing visions of a plausible, aspirational future including artificial general intelligence.

This post gives an overview of the contest and our reasons for running it. For full details on how to enter, visit worldbuild.ai.

What is Worldbuilding?

Worldbuilding is the art and science of constructing a coherent and relatively detailed fictitious world. It is frequently practised by creative writers and scriptwriters, providing the context and backdrop for stories that take place in future, fantasy or alternative realities.

Overview of the Worldbuilding Contest

This contest challenges entrants to use worldbuilding to explore possible futures for our own world.

Worldbuilding in this context is not prediction so builds need not reflect the most probable scenarios but they must be a) plausible, b) aspirational and c) consistent with a set of ground rules:

The year is 2045.

AGI has existed for at least 5 years.

Technology is advancing rapidly and AI is transforming the world sector by sector.

The US, the EU and China have managed a steady, if uneasy, power equilibrium.

India, Africa and South America are quickly on the ride as major players.

Despite ongoing challenges, there have been no major wars or other global catastrophes.

The world is not dystopian and the future is looking bright.

For definitions of plausible and aspirational, scroll to "Your Mission" here.

How to Enter

Applications comprise of four parts:

A timeline from 2022 to 2045. For each year, you must you specify at least two events that occurred (e.g. “X invented”) and provide one data point (e.g. “GDP rises by 25%”). Participants are encouraged to fill all the data points on the timeline but entries will still be accepted and judged (though at a disadvantage) if fewer than 23 years are provided. Submissions will be disqualified if fewer than 10 years are provided.

Two “day in the life” short stories of between 750 and 1000 words. These should recount a day in the life of an individual somewhere in the world in 2045. The stories can focus on the same individual or two different individuals.

Answers to the following prompts. Each answer must be fewer than 250 words.

AGI has existed for years but the world is not dystopian and humans are still alive. Given the risks of very high-powered AI systems, how has your world ensured that AGI has remained safe and controlled, at least so far?

The dynamics of an AI-populated world may depend a lot on how AI capability is distributed. In your world, is there one AI system that is substantially more powerful than all others, or a few such systems? Or are there many top-tier AI systems of comparable capability? Or something else

How has your world avoided major AI/AGI arms races and wars?

In the US, the EU, and China, how and where is national decision-making power held, and how has the advent of advanced AI changed that, if at all? (max 500 words)

Is the global distribution of wealth as measured by national or international gini coefficients more or less unequal than 2022’s, and by how much? How did it get that way?

What is a major problem that AI has solved in your world and how did it do so?

What is a new social institution that has played an important role in the development of your world?

What is a new non-AI technology that has played an important role in the development of your world?

What changes to the way countries govern the development and/or deployment and/or use of emerging technologies (including AI), if any, played an important role in the development of your world?

Pick a sector of your choice (education, tr...

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: Truthful LMs as a warm-up for aligned AGI, published by Jacob Hilton on January 17, 2022 on The AI Alignment Forum.

This post is heavily informed by prior work, most notably that of Owain Evans, Owen Cotton-Barratt and others (Truthful AI), Beth Barnes (Risks from AI persuasion), Paul Christiano (unpublished) and Dario Amodei (unpublished), but was written by me and is not necessarily endorsed by those people. I am also very grateful to Paul Christiano, Leo Gao, Beth Barnes, William Saunders, Owain Evans, Owen Cotton-Barratt, Holly Mandel and Daniel Ziegler for invaluable feedback.

In this post I propose to work on building competitive, truthful language models or truthful LMs for short. These are AI systems that are:

Useful for a wide range of language-based tasks

Competitive with the best contemporaneous systems at those tasks

Truthful in the sense of rarely stating negligent falsehoods in deployment

Such systems will likely be fine-tuned from large language models such as GPT-3, hence the name.

WebGPT is an early attempt in this direction. The purpose of this post is to explain some of the motivation for building WebGPT, and to seek feedback on this direction.

Truthful LMs are intended as a warm-up for aligned AGI. This term is used in a specific way in this post to refer to an empirical ML research direction with the following properties:

Practical. The goal of the direction is plausibly achievable over the timescale of a few years.

Valuable. The direction naturally leads to research projects that look helpful for AGI alignment.

Mirrors aligned AGI. The goal is structurally similar to aligned AGI on a wide variety of axes.

The remainder of the post discusses:

The motivation for warm-ups (more)

Why truthful LMs serve as a good warm-up (more)

The motivation for focusing on negligent falsehoods specifically (more)

A medium-term vision for truthful LMs (more)

How working on truthful LMs compares to similar alternatives (more)

Common objections to working on truthful LMs (more)

Warm-ups for aligned AGI

There are currently a number of different empirical ML research projects aimed at helping with AGI alignment. A common strategy for selecting such projects is to select a research goal that naturally leads to helpful progress, such as summarizing books or rarely describing injury in fiction. Often, work on the project is output-driven, taking a no-holds-barred approach to achieving the selected goal, which has a number of advantages that aren't discussed here. On the other hand, goal selection is usually method-driven, tailored to test a particular method, such as recursive decomposition or adversarial training.

The idea of a warm-up for aligned AGI, as defined above, is to take the output-driven approach one step further. Instead of selecting projects individually, we attempt to choose a more ambitious research goal that naturally leads to helpful projects. Because it is harder to predict the course of research over multiple projects, we also try to make the goal structurally similar to aligned AGI, to make it more likely that unforeseen and auxiliary projects will also be valuable.

Whether this output-driven approach to project selection is preferable to the method-driven approach depends on more specific details that will be discussed later. But it is worth discussing first the advantages of each approach in broad strokes:

Momentum versus focus. The output-driven approach involves having a consistent high-level goal, which allows different projects to more directly build upon and learn from one another. On the other hand, the method-driven approach involves more frequent re-evaluation of goals, posing less of a risk of being distracted from the even higher-level goal of aligned AGI.

Testing assumptions versus testing methods. The output-driven approach makes it easie...

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: Different way classifiers can be diverse, published by Stuart Armstrong on January 17, 2022 on The AI Alignment Forum.

With thanks to Lee Sharkey and Michael Cohen for the conversations that lead to these ideas.

In a previous post, I talked about how we could train classifiers on the same classification problem - a set of lions vs a set of huskies - but using different approaches to classify.

What we want is something we can informally call a 'basis' - a collection of classifiers that are as independent of each other as possible, but that you can combine to generate any way of dividing those two image sets. For example, we might have a colour classifier (white vs yellow-brown), a terrain classifier (snow vs dirt), a background plant classifier, various classifiers on the animals themselves, and so on. Then, if we've done our job well, when we find any not-too-complex classifier Cn, we can say that it's something like '50 colour, 60% nose shape and −10% plant[1]'.

We shouldn't put too much weight on that analogy, but we do want our classifiers to be independent, each classifier distinct from anything you can construct with the all others.

Here are four ways we might achieve this this.

Randomised initial seeds

An easy way of getting an ensemble of classifiers is to have bunch of neural nets (or other classification methods), initialise them with different initial weights, and train them on the same sets. And/or we could train them on different subsets of the lion and husky sets.

The advantage of this method is that it's simple and easy to do - as long as we can train one classifier, we can train them all. The disadvantage is that we're relying on luck and local minima to do the job for us. In practice, I expect these methods to all converge to "white vs yellow-brown" or similar. Even if there are local minima in the classification, there's no guarantee that we'll find them all, or even any. And there's no guarantee that the local minima are very independent - 99.9% colour and 0.01% nose shape might be a local minima, but it's barely different from a colour classifier.

So theoretically, this isn't sound; in practice, it's easy to implement and play around with, so might lead to interesting insights.

Distinct internal structure

Another approach would be to insist the classifiers internal structures are distinct. For example, we could train two neural net classifiers, C1 with weights →w1 and C2 with →w2. They could be trained to minimise their individual classification losses and regularisations, while ensuring that →w1 and →w2 are distinct; so a term like −→v1−→v2 would be added to the loss function.

This approach has the advantage of forcing the classifier to explore a larger space, and is not restricted to finding local minima. But it's still theoretically unsatisfactory, and there's no guarantee that the classifiers will really be distinct: C1 and C2 may still end up as colour classifiers, classifying the same colour in two very different ways.

Distinct relative to another set

In the previous methods, we have defined independence relative to the classifiers themselves, not to their results. But imagine now that we had another unlabelled set of images U, consisting of, say, lots of varied animal images.

We can now get a theoretical definition of independence: C1 and C2 are independent if they give similar results on the lion-vs-husky problem, but are distinct on U.

We might imagine measuring this difference directly on U: then knowing the classification that C1 gives on any element of U, tells us nothing about what C2 would give. Or we could use U is a more semi-supervised way: from these images, we might extract features and concepts like background, fur, animal, tree, sky, etc. Then we could require that C1 and C2 classify huskies and lions using only those features;...

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 needs a hiring agency and Nonlinear will fund you to start one, published by Kat Woods on January 17, 2022 on The Effective Altruism Forum.

If your dream job is to work in longtermism, be your own boss, and talk to EAs all day, then you might be the perfect fit for starting an EA recruitment agency through Nonlinear’s incubation program.

To find out more about the idea itself and why it’s high impact, read below.

For more details about how to tell if you’re a good fit, what support we provide, and how to apply, see the second half of the article.

If you think this is an important charity to have in the EA space, please like and share this Request for Founders so that the right people see it.

Deadline: February 1st, 11:59pm EST

Fill out this form to apply

Why it’s high impact to start an EA headhunting agency

The idea is for you to start an organization that helps hire employees for longtermist orgs. This will help the world in a few ways:

Save orgs’ time. Hiring takes a lot of time. It often takes over 200 hours to hire a new employee. Fortunately, a lot of this can be outsourced. A typical approach would be for you to talk to the org about what they’re looking for. You’d then write the job ad, advertise it, do the first couple rounds of interviews and test tasks, then send the client the top ones for the final evaluation. This can shorten the amount of time they spend on hiring from multiple full-time weeks to an afternoon, freeing up their time for direct work only they can do.

Hire better candidates. Since you’ll be a specialist in hiring, you’ll have a bigger EA network and know all the best practices for recruiting, so you’ll be able to get better candidates for the positions. This is huge. Team quality is one of the largest sources of differences in an organization’s effectiveness. This increases the impact of the organization until they hire a new person, which is often years later. Furthermore, having better colleagues has positive externalities on the rest of the team by decreasing conflict, creating a culture of excellence, and generally helping the team work more smoothly. This is an example of passive impact, where you make an upfront investment, then it continues having an impact without ongoing effort.

Save applicants’ time. Instead of separately having to apply to each new job, people can apply to multiple jobs at once, even expressing general interest in certain sorts of jobs. This also increases their odds of getting an EA job per time invested, which can help with feelings of rejection in the movement.

Make jobs. Lowering the barrier to hiring will increase the number of jobs orgs hire for. This will be especially true for hiring personal assistants (PAs) since a large barrier to hiring them is the process being so difficult.

What roles will the startup hire for?

Nonlinear’s research team has identified hiring PAs as one of the best use cases of a hiring agency in EA. This is for a few reasons.

Firstly, as 80,000 Hours describes in their career guide, PAs can be a high-impact role. Ben Todd says it better than me here:

“Consider: if you can save that researcher one hour spent on activities besides research, then that researcher can spend one more hour researching. So, by saving that researcher time, you can convert your time into their time. Suddenly, one of your hours becomes one more hour spent by the best researcher, working in the best field!”

Put another way, imagine you could add another Paul Christiano, Eliezer Yudkowsky, or Chris Olah to the alignment community. With a PA hiring agency, you could essentially do that by getting the top researchers PAs.

To use an extremely oversimplified example, say a PA increases the output of a researcher by 10% on average. That means if you hire ten PAs for the top ten researchers, that’s the equivalent...

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: Teaching Street Crossing, published by jefftk on January 16, 2022 on LessWrong.

When Lily was about four we started practicing crossing streets. There were two small streets on the way to the

farther park, and Lily was very excited about the idea of crossing them by herself. After talking through the process (look both ways, then if you don't see any cars coming across slowly and steadily while continuing to look both ways) I would stand with her at the corner and say "cross when you think it safe", watching for cars myself so I could stop her if she stepped out when she shouldn't.

Unfortunately, she took looking both ways to be about pointing her head, not about actually looking. She would wiggle her head as fast as possible, trying to maximize how many times she could look left and right, much faster than she could actually tell what she was seeing. (Lily: "when I first started learning how to cross streets I would turn my head like a maniac.") This was not useful practice, and after trying and failing to show her how to look I decided to wait until she was older. After a while I stopped bringing up the idea, she stopped bringing it up, and this was on hold for a couple years.

This summer, when Lily was seven and Anna was five, we tried again. This time it went much better: they would both actually look for cars, and would see them when they were coming. Pretty quickly we got to where we would do this with most streets: I would ask whether they wanted to practice, and they would cross on their own with me checking that they were doing it right. On streets with parked cars, I taught them to slow down and look again after they got to the "car line". This is the point about 6 feet from the curb where sightlines are no longer blocked by parked cars.

Over the next few weeks with lots of walks together they were both, but especially Lily, reliably doing it right. I started giving them permission when they had gone ahead of me, to where I could see them and tell whether they were looking properly, but couldn't see everything they could see. By September (7y5m) Lily was almost ready to do it on her own, and in October she was ready for a set of small streets near our house. We used walkie-talkies, and she would check in with us over the radio before crossing and after to let me know it went ok. Lily could walk places herself, and could bring Anna if they both wanted that.

The biggest street they wanted to cross but couldn't was a busy one between our house and the closest playground. I was expecting it was going to be a year or two yet, but then in November the city put in several serious "speed humps". Traffic slowed down dramatically, enough that I thought Lily could do it. I did the same thing as before, watching her many times to check her work, but by the end of the month she had demonstrated she could do it safely.

(Anna overheard me dictating this post, and asked whether we could do more practice so that she could get to crossing streets on her own. Looking forward to running through this a second time!)

I asked Lily what she saw as the main benefit of being able to cross streets, and she said "I'm able to go to friends houses without having a grown-up available." I asked if she had anything she wanted to tell parents or other kids, and she said "it's tricky at first but over time you get better at it and you get more responsibility."

In many ways, this was the same process we used in teaching them to stop at the curb: slowly increase responsibility while continuing to verify they're doing it safely. It does require concentrated parental time and attention, and slows down walks, but now they can go and visit friends entirely on their own which frees up parental time later. I think the payback period is under a year if you think of it that way, but that's not countin...

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: Challenges with Breaking into MIRI-Style Research, published by Chris Leong on January 17, 2022 on The AI Alignment Forum.

Trying to break into MIRI-style research seems to be much, much harder than trying to break into ML-style safety research. This is worrying if you believe this research to be important. I'll examine two kinds of causes: those which come from MIRI-style research being a niche area and those which go beyond this:

Challenges beyond MIRI-style research being a niche area:

MIRI doesn’t seem to be running internships or running their AI safety for computer scientists workshops

If you try to break into ML-style Safety and fail, you can always be reuse at least part of what you've learned to obtain a highly-compensated role in industry. Agent foundations knowledge is highly niche and unlikely to be used elsewhere.

You can park in a standard industry job for a while in order to earn career capital for ML-style safety. Not so for MIRI-style research.

MIRI publishes a lot less material these days. I support this decision I support as infohazards deserve to be taken seriously, but it also makes it harder to contribute.

There are well-crafted materials for learning a lot of the prerequisites for ML-style safety.

There seems to be a natural pathway of studying a masters then pursuing a PhD to break into ML-style safety. There are a large number of scholarships available and many countries offer loans or income support.

The above opportunities mean that there are more ways to gauge fit for ML-style safety research.

There's no equivalent to submitting a paper. If a paper passes review, then it gains a certain level of credibility. There are upvotes, but this signaling mechanism is more distorted by popularity or accessibility. Further, unlike writing an academic paper, writing alignment forum posts won't provide credibility outside of the field.

Challenges that come from being a niche area

I think this probably should be a niche area. It would be a bit strange if foundations work were the majority of the research. Nonetheless, it's worth highlighting some of the implications:

General AI safety programs and support - ie. AI Safety Fundamentals Course, AI Safety Support, AI Safety Camp, Alignment Newsletter, ect. are naturally going to strongly focus on ML-style research and might not even have the capability to vet MIRI-style research.

It is much harder to find people with similar interests to collaborate with or mentor you. Compare to how easy it is to meet a bunch of people interested in ML-style research by attending EA meetups or EAGx.

If you want to feel part of the AI safety community and join in the conversations are having, you will have to spend time learning about ML-style research. While is likely valuable to broaden your scope as this can lead to cross-pollination, it also sucks up time when you could be learning about MIRI-style research.

Further Thoughts

I think it's worth thinking about what this looks like overall. If you want to try breaking into MIRI-style research, the most likely path looks like saving up 3-12 months runway. 3 months might be possible if you've been consistently working on things in your free time and you've already read a lot of the material that you need to read + made substantial research progress. That said, even if you're able to produce material to prove yourself in 3 months, you'd probably need an extra month or two to obtain funding and you always need more runway than the minimum possible time. It would be possible to apply for an LTFF grant to support this research, but it's probably easier to build up the credibility for ML-style research.

Further, if you fail, then you haven't learned skills nor gained credibility that would assist you for any other paths.I suspect that these considerations not only significantly 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: Nudging My Way Out Of The Intellectual Mosh Pit, published by Elizabeth on January 15, 2022 on LessWrong.

As part of my ongoing effort to improve my cost/benefit ratio on social media, I’m nudging myself away from intellectual mosh pit platforms like Facebook and Twitter, and towards blog posts, articles, videos essays, etc. Really longform consumption (e.g. books) remains about the same, my limitations on that are mostly my insistence on fact-checking nonfiction and very narrow tastes in fiction, this post is about changing what I reach for when I’m bored in a line. Towards that goal I have made a few changes, which I list here roughly in ascending order of how much effort they were:

Put all of my screens in greyscale. If you only try one thing on this list, make it this one. It takes five seconds to test once you have instructions, and the relief for me was immediate and has lasted days so far. Every time I have to turn color on to look at graphs I resent it.

Instructions for OS X.

Instructions for iPhone and older Android (untested).

Instructions for modern Android.

Windows users: you’re on your own I only use Windows for games.

Discovered the existence of Focus Mode for android, which allows you to use the internet but disables notifications.

All of the Focus Mode instructions require several clicks into a deep menu. You can access it more easily by enabling the relevant fast-access card, using the instructions for older Android phones above.

Moved all my short-OODA-loop apps off my phone home in favor of long-OODA-loop apps.

The newly defavored apps include obvious candidates like Twitter, but also all messaging apps and Chrome itself.

The new stars include my article aggregator, as well as very long-form content like Kindle, Audible, and PodCatcher.

The home screen continues to hold non-content-consumption apps I want to access quickly, like Maps, Calendar, etc.

Resumed use of a read-it-later tool, which lets me save cool articles I see on social media to be consumed when I’m in a better head space (I can’t switch between reading microblogs and regular blogs quickly – different headspace ).

Switched to an RSS reader that lets me read things out of order without marking earlier articles as Read.

Combined my RSS feeds, email newsletters, and saved articles in a single service (that lets me read in any order) so I can choose from all my essay-length options at once. This was a stupid amount of effort and yak shaving and it’s not pretty but I got it working. Most of this post will be about that.

How to Combine All Article-Length Content

The following instructions work with both Inoreader and Feedly. I eventually chose Inoreader but it was a close call and your mileage may vary.

Import your existing RSS feeds to Inoreader.

Feedly instructions.

You have two options for saving arbitrary content to your new aggregator.

Inoreader has a built-in feature to do this with a Chrome plugin, but there’s no way to see saved articles and RSS articles in the same list. So if that’s important for you.

Sign up for Pocket, the leading read-it-later app, and configure Pocket to put your saved articles on an RSS Feed, then add that RSS feed to Inoreader per normal.

Install the save to Pocket extension in your browser to make it easy to add to the feed as you come across things you want to read at some point (note: not available for Android, so I have to manually copy the URL and open the Pocket app).

All articles in the saved-to-Pocket feed will show in Inoreader as having the same author (“my content feed”) and they won’t have the body of the article, only the title and header image. I really care about having all of my articles in the exact same place, but if you don’t, just using save-to-Inoreader will save you several steps.

You also have two options for newsletters.

Set up forwar...