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: School Daze, published by Zvi on January 14, 2022 on LessWrong.

In our educational system, the people are represented by two separate yet equally important groups. The children who have committed no crime, and the adults who prosecute the offenders. These are their stories.

We need to close the schools to slow the spread, they said. But what profit us to close the schools, if most everyone is getting Omicron anyway, the schools aren’t less safe than what kids would do anyway, and the kids were never in real danger?

We need to do remote learning if we close the schools, they said. But what does ‘remote learning’ accomplish in its current form, other than to ensure children are properly punished for the crime of being children? Teaching them that society wants them to suffer and that their success depends on obeying arbitrary rules while looking like they are looking into a computer screen all day and ‘paying attention’ without secretly doing anything useful or fun? Teaching them how to get around such rules?

We need to keep the schools open, they said. No one is learning anything with this ‘remote learning,’ they (correctly) said. But what profit us to keep the schools open if no one is doing any learning there either, no matter what you think of a normal school day, because the rules and obsessions regarding Covid-19, combined with absences, render the entire operation non-functional? What profit all these precautions, tests, isolations and rules, if they transparently don’t identify the sick and keep them out of the building, and in many cases create situations that look quite a bit like superspreader events?

No profits. Only pain.

Because all these options have one key thing in common.

They all fail to Play to Win the Game.

Or at least, they fail it for any good game. For any game where winning is preferable.

The Game is up to you. You define what winning means.

The Game can be learning reading, writing and arithmetic, probability and statistics, economics, history, science, art and music, or anything else you care to teach. The Game can be to cultivate their creativity, or give them socialization. The Game can be our children not suffering and not being depressed and suicidal. The Game can be the living of life.

If you’re cynical but not too cynical, The Game can be allowing parents to go to work and society to function. It can be teaching children how to sit still and control themselves, and how to guess the teacher’s password. The Game can be to avoid having the wrong marks on a child’s permanent record. It can be for students to signal how smart, conscientious, rules abiding or rules abiding they are, or to certify their class membership. It can be a place for them to make business connections.

Remote learning is widely acknowledged to have failed at these goals, and is very unpopular. Sending children to schools where their time is chaotic, stressful and wasted, and very little is taught, will doubtless do the same.

The Game can also be ‘keep children from getting Covid-19, even if they are not themselves in danger’ whether or not one agrees that this game is worth playing. Some people seemed to think this is the ballgame, that children’s lives should be about this for years on end, and now that many no longer think this they are turning around and saying that school is ‘the safest place for children to be.’

This is obvious nonsense for a given child, and only somewhat less obviously nonsense for children in general, if taken literally. It’s also a rather scary statement about such people’s range of potential places ‘for children to be.’ Why can’t they be?

The actual claim is, as far as I can tell, some combination of ‘in practice many kids end up taking more risks otherwise and on net it’s worse’ and ‘we want the schools open so we need to call them safe,’ or...

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: 2020 Review Article, published by Vaniver on January 14, 2022 on LessWrong.

A common thing in academia is to write ‘review articles’ that attempt to summarize a whole field quickly, allowing researchers to see what’s out there (while referring them to the actual articles for all of the details). This is my attempt to do something similar for the 2020 Review, focusing on posts that had sufficiently many votes (as all nominated posts was a few too many).

I ended up clustering the posts into seven categories: rationality, gears, economics, history, current events, communication, and alignment.

Rationality

The site doesn't have a tagline anymore, but interest in rationality remains Less Wrong's defining feature.

There were a handful of posts on rationality 'directly'. Anna Salamon looked at two sorts of puzzles: reality-masking and reality-revealing, or those which are about controlling yourself (and others) or about understanding non-agentic reality. Listing out examples (both internal and external) helped explain cognitive biases more simply. Kaj Sotala elaborated on the Felt Sense, a core component of Gendlin’s Focusing. CFAR released its participant handbook. Jacob Falkovich wrote about the treacherous path to rationality, focusing on various obstacles in the way of developing more rationality.

Personal productivity is a perennial topic on LW. alkjash identified a belief that ‘pain is the unit of effort’, where caring is measured by suffering, and identifies an alternative, superior view. Lynette Bye gave five specific high-variance tips for productivity, and then later argued prioritization is a huge driver of productivity, and explained five ways to prioritize better. AllAmericanBreakfast elaborated on what it means to give something a Good Try. adamShimi wrote about how habits shape identity. Ben Kuhn repeated Byrne Hobart’s claim that focus drives productivity, and argued that attention is your scarcest resource, and then talked about tools for keeping focused. alkjash pointed out some ways success can have downsides, and how to mitigate those downsides.

orthonormal discussed the impact of zero points, and thus the importance of choosing yours. Jacob Falkovich argued against victim mentality.

There was some progress on the project of 'slowly digest some maybe-woo things'. Kaj Sotala gives a non-mystical explanation of “no-self”, detailing some 'early insights' into what it means, as part of his sequence on multiagent models of mind. Ouroboros grapples with Valentine’s Kensho. I write a post about how Circling (the social practice) focuses on updating based on experience in a way that makes it deeply empirical.

Gears

John Wentworth wrote a sequence, Gears Which Turn the World, which had six nominated posts. The first post discussed constraints, and how technology primarily acts by changing the constraints on behavior. Later posts then looked at different types of constraints, and examples where that constraint is the tight constraint / scarce resource: coordination, interfaces, and transportation. He argued that money cannot substitute for expertise on how to use money, twice.

Other posts contained thoughts on how to develop better models and gearsy intuitions. While our intuitive sense of dimensionality is low-dimensional space, much of our decision-making and planning happens in high-dimensional space, where we benefit from applying heuristics trained on high-dimensional optimization and geometry. Ideas from statistical mechanics apply in many situations of uncertainty. Oliver Habryka and Eli Tyre described how to Fermi Model. Maxwell Peterson used animations to demonstrate how quickly the central limit theorem applies for some distributions. Mark Xu talked about why the first sample is the most informative when estimating a uncertain quantity.

Scott Alexander w...

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 Should Run More EA Contests, published by Akash on January 12, 2022 on The Effective Altruism Forum.

TLDR

I think we should have more EA contests. EA contests can generate direct value, help people develop/test their aptitudes, incentivize people to develop inside views, and help organizations find potential hires.

Also, I recently ran a small contest at a retreat (skip to the "Minimum Viable Contest" section to see how that went).

What is an EA Contest?

An “EA Contest” is essentially a call for proposals that includes a monetary reward to winners. In other words, a contest includes a contest prompt (e.g., work on X problem) and a prize (e.g., the winners get $Y).

Examples include the Creative Writing Contest, the Philosophy Contest, and the Eliciting Latent Knowledge Contest.

Why Are EA Contests Valuable?

First, I think contests can provide direct value. For example, it seems plausible that winning entries to the Creative Writing Contest could be used to generate EAs or increase the motivation of existing EAs.

Second, I think contests help people test their aptitudes. I have met many EAs who think “ah, I have no idea if I’m a good fit for X, and I wish there were easy ways to test it.” Contests can give these people a straightforward, low-friction way to test out their fit for a particular aptitude.

For example, the Eliciting Latent Knowledge Contest provides a clear way to test your fit for theoretical alignment research. This reduces friction considerably. To spell it out:

Ron: How can I test my fit for theoretical alignment research?

Hermione: Well, theoretical alignment is a pretty messy and confusing area at the moment. I suggest you read a bunch of posts in the area and try to find holes in the arguments. But also be aware that many of the posts are unhelpful and inaccurate, so part of your job is to figure out which posts actually make compelling arguments.

Ron: Oh, uh, okay. filled with confusion

Harry: Actually, Ron, another option is to try the Eliciting Latent Knowledge Contest! There’s a defined problem that you can work on, all of the content is in one Google Doc, and it seems like you could get a “taste” for this work by investing about 30 hours.

Ron: Oh, wow. That seems less nebulous– I’ll give it a shot!

Third, I think contests improve the critical thinking of the EA community. Contests often incentivize people to engage with primary sources, form their own ideas/opinions, and refine their thoughts on a topic. Contests can also be used to help people develop inside views on important topics.

For example, let’s imagine a hypothetical contest that involves red-teaming Open Philanthropy’s EA/LT survey. There could be direct benefits to this kind of contest (e.g., helping us better understand how to interpret the results; finding ways that OP can improve the survey design).

But I think the other benefit– which might be even more important– is that this contest would get more people to read the Open Phil survey (in detail) & generate their own interpretations. This illustrates a broader point: contests can incentivize people to defer less and form inside views more.

Fourth, I think contests help EA orgs identify untapped talent & potential hires. The Creative Writing Contest, for example, cast a spotlight on several talented EA writers. If an EA org is looking for creative writers, there’s now a pool to draw from. The Eliciting Latent Knowledge contest may also be used to help the Alignment Research Center find potential hires (indeed, this might be the primary goal of the contest!).

What could go wrong?

First, this could waste EA time. This is especially concerning if there’s a contest that looks appealing but actually produces little value. This is a waste of money and time (I’m especially concerned about the latter. A “bad contest” could suck up hun...

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: Covid 1/13/22: Endgame, published by Zvi on January 13, 2022 on LessWrong.

Important note, fleshed out a bit more in Omicron Post #14: From various sources, I have become convinced that rapid tests taken from nose swabs are likely to often be several days slower at detecting infections than rapid tests that use throat swabs.

So if you are testing with the goal of actually learning whether you have Covid or are infectious, that’s what you should do. Otherwise, you’re mostly testing in order to have conducted a test.

Your call.

We’re on the way down in the Northeast – we have the wastewater in Boston, which is already down 50%, we have a clear peak in New York City, and several states are clearly stabilized or better. Within a week, I expect the peak to be hit on average throughout the country.

That doesn’t mean things suddenly end, of course. Assuming no worries about hospitals, the highest value to vigilance actually comes right after the peak, if you think you’ve avoided infection until now. You’ve survived roughly half the risk at that point, making it more reasonable to try and avoid infection entirely, yet risk is still very close to its maximum on a per-day basis.

Yet it is time to start once again making plans for The After Times, which will be with us sooner than it might seem. Right now, the schools are a mess, everything is facing staff shortages and it all feels a little scary, but (death counts and a few hospital conditions aside) this is as bad as it’s going to get, and within a few weeks things will be improving on all fronts.

Executive Summary

If you do a rapid test and want to know the real answer, you need to do a throat swab.

Cases peaking in the Northeast now, rest of country within two weeks.

Schools are a complete disaster, but that’s another entire post.

Let’s run the numbers.

The Numbers

Predictions

Prediction from last week: 6mm cases (+71%) and 9,700 deaths (+10%).

Results: 4.9mm cases (+37%) and 11,856 deaths (+34%).

Prediction for next week: 5.6mm cases (+13%) and 15,000 deaths (+30%).

That’s a clear miss on cases. My presumption is that we were missing more cases than I realized, and for that and/or other reasons we peaked earlier than I expected and a lot of the Northeast is on its way back down shortly. The rest of the country will follow within a few weeks. On net I still expect an increase because I don’t expect big drops in measured cases in the places that have peaked, not quite yet.

On deaths, the wonkiness continues. The West came back to its early-December rate, despite the Omicron wave having arrived late. Other areas also so increases that seem a little early slash a little big, I definitely expected things to go better on this front. I still don’t expect things to get that much worse, and presume that the declines the previous two weeks involved holiday wonkiness, and the number now is nowhere near as scary as it looks without that context.

I still do expect a rise next week, and it’s possible it could be substantially bigger than this – the ‘real’ bump should be bigger than the one I’m predicting, the uncertainty is both the real size and the measured size of the increase.

Deaths

Cases

Almost a full doubling out West, but not in other regions and the Northeast has likely stabilized. A lot of cases are being missed right now and that should continue for a bit.

Vaccinations

Omicron version of the Pfizer vaccine will be ‘ready’ in March, but not sure if anyone will even want them.

Pfizer will produce the doses to be ready in case countries want the shots, but Bourla noted that it was unclear if a vaccine targeting variants was necessary or how exactly it would be used.

Moderna boosters now also down to five months from second dose. No confusion here.

Vaccine Effectiveness

Fourth doses, how would they even work?

Moderna to report d...

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: Omicron Post #14, published by Zvi on January 13, 2022 on LessWrong.

Important note on testing that I’m including at top of both posts today.

From various sources, I have become convinced that rapid tests taken from nose swabs are likely to often be several days slower at detecting infections than rapid tests that use throat swabs.

Here’s Washington Post on that, this quote should tell you what you need to know.

On one side are well-regardedexperts who argue that swabbing your throat in addition to your nose may increase the chances that a home test can detect omicron. On the other side are similarly well-regarded experts, including those at the Food and Drug Administration, who urge Americans to avoid experimenting with kits that were developed and tested using only nasal swabs.

This ‘disagreement’ over ‘experimentation’ made me that much more confident in the conclusion I’d already reached.

That is echoed by a lot of sources both personal and general, including in this heartfelt report of Bob Watcher, whose son has Covid-19 and reports back on the experience, including five days later follow-up where he’s still positive and acts confused about what to do because CDC guidelines are confusing a situation that is otherwise extremely not confusing.

Thus, if you want to know if you have infectious Covid-19, you need to use throat swabs. It’s still not 100% sensitivity or anything, but it’s a lot better, whereas nose swabs are useful but in no way all that reassuring.

If you want to satisfy a testing requirement that is of course another matter. These are very different goals.

This is a whole week’s roundup, because I got an opportunity to impact developments more directly and needed to jump on that with my non-day-job time instead for a few days. I hope to be able to talk about it soon, but can’t right now.

Travel

It’s a very good question. Why the hell are we still doing this?

The Centers for Disease Control and Prevention on Monday advised Americans to avoid travel to Canada, citing “very high” levels of the coronavirus.

Canada was placed under a Level 4 travel health notice — the highest category.“Because of the current situation in Canada, even fully vaccinated travelers may be at risk for getting and spreading Covid-19 variants,” the C.D.C. said.

The United States has higher rates of Covid-19 than Canada right now, yet we do this thing. I simply can’t take seriously, as someone who might want a physically better and healthier world, anyone who restricts travel in such ways.

The CDC resumed briefings this week. This is the kind of thing they think we need to know.

Spread

CDC Nowcast has Omicron at 98.3%, which seems reasonable to me.

New update from Bob Wachter on San Francisco.

Two big notes are that there’s about an even mix of vaccinated and unvaccinated Covid patients outside of the ICU (this is SF, so a large majority of people are vaccinated) and then 11 of 12 in the ICU are unvaccinated. And then there’s the background rate, which is pretty astounding.

Trevor Bedford estimates that about one third of London infections are being detected. I continue to think that is low, especially given the peak already happened, but it’s not outside the range of possible answers.

Anecdata from Michael Mina.

A key question for modeling how things go is whether people are modifying their behavior. Some people definitely are, but others definitely aren’t, including when they know they have Omicron. One commenter noted that their friends treated it like a cold and mostly ignored it (as opposed to what one should do with a cold, which is stay home) and then there’s this.

Thread defending the UK’s case declines as genuine. I agree, there’s no reason to think the declines aren’t real.

A thread from 9 January on progress in Scotland and rest of UK, pointing out things have peaked bu...

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: Question Gravity, published by KingSupernova on January 13, 2022 on LessWrong.

Earlier today, I was in the shower. I had just finished a bottle of shampoo, and I tried balancing it on the thin railing of the shower door. I expected it to fall down, but instead it stayed there. From the side, it certainly looked like it was sticking out more than halfway over the edge. (Picture) Yet it stubbornly stayed put.

On a whim, I filled it with water and put it back. It fell over.

Wait, what? Filling it up shouldn't change the location of its center of gravity, so why would that make it fall over? I started performing more experiments. I rotated the bottle to see if there was some asymmetry in its construction. I thought maybe the water sloshing around could cause its center of gravity to move over the edge, so I I tried holding it there for several seconds before letting go, in order to let the water settle. I noticed the door had some wiggle room, and the weight of the bottle could be causing the door to shift to a different angle, so I tried holding it at different angles. Nothing caused any different behavior. I put it on the ledge empty and slowly filled it with water. When it got to around 1/20 full, it fell over. I found a second bottle and performed all the same tests on it; same results.

At this point I was thoroughly confused. I figured some assumption of mine must have been wrong, so I tried inspecting them. Does the center of gravity of an object really have to be over the edge in order for it to fall? I couldn't think of any situation where that wasn't true. (Generalizing from a non-exhaustive list of arbitrary examples is obviously not a great approach, but I couldn't do much better while in the shower.) Does filling it with water really not change its center of gravity? Oh wait; it does! The center of gravity doesn't move it laterally, but it does move it vertically. That assumption was wrong. (I probably wouldn't have made this mistake if I had been thinking of it as the center of mass rather than the center of gravity. Framing is important.) Unfortunately, that epiphany didn't lead to anything useful; I couldn't see how moving the center of mass vertically could affect whether it was balanced or not.

If you haven't figured it out yet, take a moment to think about it. You have all the information you need. Also, let me know in a comment if and how easily you figured it out before reading the answer below. I'm curious how well I stack up.

Answer: Cohesion between the water on the bottle and the water on the glass of the shower door.

This came to me after I noticed that when the bottle was almost, but not entirely, empty, it would seem to hesitate for a moment before falling. I didn't come to this immediately though. While I did notice the delay at the time and mentally flag it as "weird", I failed to consider how it might fit into the picture, and unconsciously dismissed it as irrelevant without giving it proper consideration. Only later did I have an independent realization that the bottle might be sticking to the door. That realization came from my unconscious mind; I didn't arrive at it though a deliberative process. Once I'd had it, it was trivial to grab a towel and confirm that when dry, the bottle would fall over even when empty.

There were several failures here, but the most important one was not going back far enough in the chain of assumptions I was making. I realized I needed to go through the process "Question Assumptions", but ended up only inspecting the irrelevant ones. I was unable to notice the assumption that actually needed questioning; that gravity was the only relevant force at play.

Properly questioning assumptions; not just going through the motions, but actually locating and inspecting all the foundational pillars of your thought process,...

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: Political Office for Beginners?, published by Asgård on January 12, 2022 on LessWrong.

If an American citizen was looking to run for office (from local, state, all the way to federal), what would be the recommended steps to take?Specific questions that come to mind: How would a Millennial or Gen Z'er deal with existing social media accounts? How would a Gen X'er or Boomer create and navigate social media accounts and advertising? Where and how would the first $10,000 do the greatest good? First $100,000? Etc. Are there political grants/party grants/etc. available in the United States for less-funded campaigns? When should a citizen start campaigning for an election? Where and how would the first 100 hours do the greatest good? First 1,000? Etc.

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: Uncontroversially good legislation, published by braces on January 10, 2022 on LessWrong.

Say you’re fortuitously seated next to a senator, state assembly member, or city supervisor at a wedding. You’ve got to talk about something, so you figure it wouldn’t be rude to mention some legislative issue you’re passionate about.

What would the legislation be?

I’m curious about almost definitely good laws that Republicans, Democrats, and Independents could all get behind. This could be at the local, state, or federal level although it seems as though there might be more obvious reforms in smaller governments. (I searched around for this here and on EA forums and didn’t find any, but would love to know about any similar lists.)

Here’s a few I could think of. These are debatable and surely reflect my own biases and limited knowledge.

Price transparency: Airlines are required by the Full Fare Advertising rule to show the all-in cost whenever they advertise prices. Why not the same for hotels and food delivery, where the price jumps at the last page from taxes and fees? This would better allow businesses to compete on price rather than deception.

Let us buy glasses: We can’t buy glasses or contact lenses if our eye prescription is over 1-2 years old. This means that every 1-2 years, glasses-wearers need to pay $200 to optometrists for the slip of paper (and stinging eyeballs). Seems like it’s probably a racket and the benefit from detecting the odd eye cancer is outweighed by the costs, although see the debate here. (Edit: it's actually not that hard to get around this, at least for glasses. Thanks Dustin.)

Tax filing in the US should be more automated: This one is well-known (see Planet Money episode) and seems like another case of common sense vs. lobbyists. (Speaking of Planet Money, another candidate is repealing the Jones Act.)

What are some 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: New year, new research agenda post, published by Charlie Steiner on January 12, 2022 on The AI Alignment Forum.

Thanks to Steve Byrnes, Adam Shimi, John Wentworth, and Peter Barnett for feedback.

In a nutshell, my plan A is to understand what we want from superintelligent AI really, really well. So well that we can write down a way of modeling humans that illuminates human preferences including higher-order preferences about how we want to be modeled, and do this in a principled rather than ad-hoc way. Achieving this understanding is highly ambitious, in a way that is mostly but not entirely parallel to "ambitious value learning."

If we understand value learning before we build superintelligent AI, there's a straightforward path to achieving a good future without paying a costly alignment tax - by alignment tax I mean all those things that slow down aligned AI being developed and "ready for liftoff," that collectively create selection pressure against safety. This problem is more tractable than many people think.

I think of this plan as an instance of the more general plan "solve value learning first." Some other tenable plans are "try to put humans in control," "get the AI to do good things prosaically," and "tinker with the AI until it value learns" - plus intermediate points between these.

What do I think the future is like?

I expect superintelligent AI in the short to medium term, centralized around a small number of points of development.

By short to medium term, I mean I'd put my 50% confidence interval between 2031 and 2049. I don't think we need compute to be many orders of magnitude cheaper, and I don't think we need two or more paradigm shifts on the order of neural nets overtaking support vector machines. The timeline is urgent, but not to the point that we should start ditching things like blue-sky research or gradual coalition-building.

By centralized, I mean it's possible to make big changes by having good solutions implemented in a small number of systems. Coordination may be important, but it isn't an inherent part of solving the problem.

All that said, that's just what I think is going to happen, not what's required for "solve value learning first" to be a good idea. Value learning research is still valuable in decentralized scenarios, unless we go so far as to avoid powerful agential AI long-term. Because it's more on the blue-sky end of the spectrum, longer timelines actually favor solving value learning over more atheoretic approaches, while if timelines are very short I'd advocate for "try to put humans in control" and hope for the best.

If we fail to understand value learning before we build superintelligent AI, I'm worried about some combination of groups committed to building aligned AI being less competitive because we can't learn human values efficiently, and practical-minded alignment schemes having bad behavior in edge cases because of simplifying assumptions about humans.

A basic example: if humans are assumed not to be manipulable, then an AI that thoroughly maximizes what humans (are modeled to) want will be incredibly manipulative. From the AI's perspective, the humans love being deceived, because why else would they rate it so highly? And in fact it's a tricky technical problem to avoid manipulation without sophisticated value learning, because the notion of "manipulation" is so intertwined with human meta-preferences - labeling things as manipulation means un-endorsing some of our revealed preferences. Similar hidden gotchas can pop up in other attempts to cut corners on human modeling, and at some point it just becomes faster to solve value learning than to deal with each gotcha individually.

What's the broad plan?

The basic strategy can be summed up as "be Geoff Hinton" (as in godfather of deep learning Geoffrey Hinton). Know an im...

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: Value extrapolation partially resolves symbol grounding, published by Stuart Armstrong on January 12, 2022 on The AI Alignment Forum.

Take the following AI, trained on videos of happy humans:

Since we know about AI wireheading, we know that there are at least two ways the AI could interpret its reward function[1]: either we want it to make more happy humans (or more humans happy); call this R1. Or we want it to make more videos of happy humans; call this R2.

We would want the AI to learn to maximise R1, of course. But even without that, if it generates R1 as a candidate and applies a suitable diminishing return to all its reward functions, then we will have a positive outcome - the AI may fill the universe with videos of happy humans, but it will also act to make us happy.

Thus solving value extrapolation will solve symbol grounding, at least in part.

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