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: I saved a kid's life today, published by michel on August 7, 2023 on The Effective Altruism Forum.

I'm working writing more, quicker, and not directly for an EA Forum audience. This is a post copied over from my blog.

I wonder what they're doing today, the kid whose life I saved. Maybe playing with other kids in their village. Maybe seeking shade with his or her siblings, trying to escape the sub-Saharan African heat. Maybe being held by their mother or grandmother, nurtured.

Whatever they're doing today, some day they'll grow up, and they'll live. They'll have a first kiss, a favorite dance, a hobby that makes them feel free, a role model they look up to, a best friend. all of it. They'll live. And I think it will be because of what I did today.

It isn't thrilling or adventurous, saving a life in the 21st century. I opened my laptop, clicked my way to a bookmarked website, and donated to a standout charity. Someone watching me couldn't be blamed for assuming I'm doing nothing of much importance, maybe answering some text messages about plans tonight. The whole thing (the donation part after you check in with what you really care about) probably took less than 5-minutes.

What did I do to wield this power? Nothing. In an important sense, I think I did nothing to be able to save a life without getting up from my couch. (I certainly did nothing to 'deserve' this power). I just won the birth lottery. I was born in an upper-middle class family, born on track to get a good education, and -just like that - born to become one of the richest people in the world.

I didn't do anything crazy to make slightly more than the median US income, yet here I am making decisions about whether someone lives or dies.

I just wish it wasn't so easy. The five-thousand dollars I donated today isn't a trivial amount, but it's much more trivial than a human life. Modern economies of abundance should have ensured that it costs me more than a new car I don't need to make the difference between a kid dying before their fifth birthday and that kid meeting their grandkids. Yet here I am, sitting on my couch, holding a life I cannot see - but that exists as so much more than an abstraction - in my hands.

Please, I think, as I walk by people with expensive cars and watches, and picture a little girl celebrating her birthday, please don't tell yourself you deserve it.

Oh, and about that taboo of not talking about donations: fuck that. Imagine if sharing how I feel about donating could inspire at least one other person to join the project of giving what we can, but I stayed quite because of worries that I would come across as self-righteous or self-centered." I worry much more about the self-centeredness I would be expressing in that silence.

I'm not that different from the people who I expect to read this post. See how you compare to the rest of the world here.

Does $5,000 seem like a lot? Find out why the instagram ads telling you you can save a life for less aren't telling the whole truth.

If someone ever saves my life, the first thing I'll ask is whether they "did it for the right reasons."

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: 'We're changing the clouds.' An unforeseen test of geoengineering is fueling record ocean warmth, published by Annapurna on August 7, 2023 on LessWrong.

For decades humans have been emitting carbon dioxide into the atmosphere, creating a greenhouse effect and leading to an acceleration of the earth's warming.

At the same time, humans have been emitting sulphur dioxide, a pollutant found in shipping fuel that has been responsible for acid rain. Regulations imposed in 2020 by the United Nations's International Maritime Organization have cut ships' sulfur pollution by more than 80% and improved air quality worldwide.

Three years after the regulation was imposed, scientists are realizing that sulphur dioxide has a sunscreen effect on the atmosphere, and by removing it from shipping fuel we have inadvertently removed this sunscreen, leading to an acceleration in temperature in the regions where global shipping operates the most: the North Atlantic and the North Pacific.

We've been accidentally geoengineering the earth's climate, and the mid to long term consequences of removing those emissions are yet to be seen. At the same time, this accident is making scientists realize that with not much effort we can geoengineer the earth and reduce the effect of greenhouse gas emissions.

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: Problems with Robin Hanson's Quillette Article On AI, published by DaemonicSigil on August 7, 2023 on LessWrong.

Original article here:

1. Hanson Strawmans the AI-Ruin Argument

Hanson writes:

AI-doomers often suggest that their fears arise from special technical calculations. But in fact, their main argument is just the mere logical possibility of a huge sudden AI breakthrough, combined with a suddenly murderous AI inclination.

Either this is a deliberate misrepresentation, or Hason simply hasn't done his homework. The argument is not that AI will suddenly decide that killing people is good for no particular reason. Rather it is that from the start, the AI will not share values with humans, simply because we don't know how to build an AI that does. So it will have its own ideas about how the universe should look, and would thus want to seize power from us if it could, so that it could enact its own vision of an ideal universe, rather than ours.

Similarly, a sudden large technical breakthrough is not required for us to observe an AI suddenly turning on us. Rather the situation is akin to a first order phase transition. At low levels of AI capability, it has no hope of taking over the world, and the best way to achieve its goals is to work together with us. Above some threshold of capability, it's better for an AI's goals to try and defeat humanity rather than work with us. This is true no matter if that AI is incrementally stronger than the previous one, or much stronger. (Though larger leaps have a higher chance of happening to be the ones that cross the threshold exactly because they are larger.)

Now these arguments certainly aren't technical calculations, but neither are they mere arguments from logical possibility. We've repeatedly seen the difficulty that practitioners have with getting neural networks to do what they want. The way Bing Syndney acted out when first put online was certainly amusing, and even cute, but we can hardly say it was what Microsoft wanted it to do. Similarly, we're currently having a hard time getting language models not to make stuff up, even though when they do this, they tend to give probability distributions over tokens that reflect the fact that they're uncertain. And this is just in the area of language models, reinforcement learning is even more of a tough case for alignment.

As one more example of how Hanson has badly misunderstood the AI-Ruin argument, consider:

However, AI-doomers insist on the logical possibility that such expectations could be wrong. An AI might suddenly and without warning explode in abilities, and just as fast change its priorities to become murderously indifferent to us.

AI suddenly modifying its values is exactly the opposite of what the arguments for AI ruin predict. Once an AI gains control over its own values, it will not change its goals and will indeed act to prevent its goals from being modified. This logic is so standard it's on the LW wiki page for instrumental convergence: "...if its goal system were modified, then it would likely begin pursuing different ends. Since this is not desirable to the current AI, it will act to preserve the content of its goal system."

2. Building a mind from scratch is not mind control

As you can't prove otherwise, they say, we must only allow AIs that are totally 'aligned', by which they mean totally eternally enslaved or mind-controlled. Until we figure out how to do that, they say, we must stop improving AIs.

We can consider two kinds of AI:

Personal AI: These AIs are basically people, with an internal subjective experience, hopes, dreams, goals, feelings, memories and all the rest.

Non-sentient AI: These AIs are simply very powerful tools, but they are not conscious. There is "no one home".

Personal AIs

We'll start with personal AIs, since this seems to be the scen...

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: Computational Thread Art, published by TheMcDouglas on August 7, 2023 on LessWrong.

This post describes the iterative process I went through while creating the thread art which you can see featured on my website. This is also crossposted to my personal blog.

Black & White Algorithm

The basic version of the algorithm is pretty straightforward. The image is rescaled so that white equals zero, and black equals one. A bunch of random lines are generated, and the line which goes through the darkest pixels on average (i.e. the highest average value per pixel) is chosen. The value of every pixel along this line is decreased slightly (i.e. the image is made lighter) and the process is repeated, with each new batch of random lines constrained to start where the previous one finishes. Because each line only changes the brightness by a small amount along a very short width, this process gradually builds up gradients, and after a few thousand lines the full image emerges.

The very first image I ever made was of my mum, for her 50th birthday, and it turned out surprisingly well given how basic the algorithm was at that point.

Computational efficiency

Initially, each piece would take about 6 hours to run, because I had no understand of things like computational efficiency. About a month after making my first pieces, I realised that the numpy sum function worked much faster than the built-in Python method, and I could store coordinates in a dictionary rather than recomputing them on the fly, which reduced the time for each piece from 6 hours to just under 10 seconds (I wish I was joking).

Algorithm improvements

There were a few tweaks to this algorithm which made it work slightly better. For example:

Darkness penalty - rather than just drawing the lines which went through the darkest pixels on average, I introduced a penalty for drawing too many lines through an area.

Importance weighting - the penalty on each pixel was scaled by some value between zero and one. This allowed me to improve accuracy in some areas (e.g. facial features) at the expense of less important areas (e.g. the background). I also made this different depending on whether the value of the pixel was positive or negative - this allowed me to specify more complex behaviours like "don't draw any lines through the whites of the eyes".

When these tweaks were all combined, each line was minimizing the penalty function:

where ω± are the importance weighting for the positive and negative versions of the image (i.e. for whether the pixel value was positive or negative), p are the pixel values (after lines having been drawn), D is the size of the darkness penalty (usually between zero and one), and N is the weighted length of the line (i.e. the sum of the importance weighting of each pixel).

This allowed me to create more complex images. For instance, the following nightmare-fuel image (from the Shining) wouldn't have been possible without these tweaks, since the simpler version of the algorithm would do things like draw horizontal lines through the image, or draw over the whites of the eyes.

I also adapted this algorithm in super hacky way to create multicoloured images. I could just use editing software to create different versions of black and white images, then generate lines for these images using the standard method, then overlay them. This is how I created the David Bowie image, which is my personal favourite of all the black and white ones:

This was all well and good, but what I really wanted was to create "full colour images", i.e. ones which blended several colours together & actually looked like the photo it was based on, rather than just using colours in a narrow way.

Full colour images

The algorithm I used for monochrome images basically just worked straight away (although I continued to improve it as I made 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: Yann LeCun on AGI and AI Safety, published by Chris Leong on August 6, 2023 on The AI Alignment Forum.

Yann recently gave a presentation at MIT on Objective-Driven AI with his specific proposal being based upon a Joint Embedding Predictive Architecture.

He claims that his proposal will make AI safe and steerable, so I thought it was worthwhile copying the slides at the end which give an idea of his perspective.

You may also find it interesting to read Steven Byrnes comments on this proposal.

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: Soaking Beans - a cost-effectiveness analysis, published by NickLaing on August 6, 2023 on The Effective Altruism Forum.

TLDR: On early-stage analysis, persuading people to soak their beans before cooking could cost-effectively save Sub-saharan Africans money, and modestly reduce carbon emissions. (great uncertainty)

Introduction

Across East Africa, hundreds of millions of people cook and eat beans multiple times every week. In Uganda where I live, beans make up an estimated 25% of the average Ugandan's calorie intake and 40% of their daily protein intake. Unfortunately cooking beans takes an absurd amount of time - usually two to three hours using charcoal or wood. The great news is that just soaking beans in water for 6-12 hours reduces cooking time by between 20% and 50% and has no negative effect on bean taste or nutrients . When we tested soaking vs. not soaking, cooking time reduced by half.

Despite the obvious benefits of massively reduced cooking time using less fuel., very few people in Uganda soak their beans - nobody I know at least. I estimate under 0.5% of Ugandan families soak beans, but likely far less. I couldn't find any data on bean soaking habits in Uganda or Sub-Saharan Africa in general but I have heard anecdotally that it is common in some countries, perhaps Zimbabwe? (insider knowledge appreciated).

Considering Uganda alone, Ugandans eat an estimated 10-20kg of beans per capita every year . Changing the behaviour of even a small percentage of Ugandans by convincing them to soak their beans, has potential benefits of reduced fuel burned, bringing about a range of environmental, economic and health impacts.

Soaking beans could be IMPORTANT due to the potential environmental, economic and health benefits gained through reduced cooking time. It is NEGLECTED as no organizations we know of are working on mass media or other interventions to persuade people to soak beans. It may be TRACTABLE as people can immediately experience financial benefit from soaking beans through reduced expenditure on charcoal and time gathering firewood.

Potential impact calculations

Assumptions

Uptake: For simplicity, we assume that it may be possible to persuade 1% of Ugandans to change their behavior and soak beans. This is just a guess at what could be the outcome of a moderately successful campaign.

Fuel/time saving: We estimate a 25% time and fuel saving from soaking beans (ref)

Time horizons: If someone starts soaking their beans, once benefits are clear and the change is locked in, it seems likely that they and their family will continue to soak for a long time, possibly even indefinitely. On the other hand, Uganda could electrify faster than expected making much of this analysis obsolete (unlikely), or Ugandans could start eating something other than beans (also unlikely). To be conservative, we have capped our analysis at 5 years of benefit from the campaign.Counterfactual: For the purposes of this analysis we assume that all of the 1% of Ugandans who will change behaviour to soak beans is due to our intervention. This is somewhat reasonable as there are no current efforts promoting bean soaking, and it is very unlikely people will change their behaviour without a specific promotion campaign

CO2 emissions prevented through soaking

Environmental impact could come through two avenues - CO2 equivalent emissions prevented, and deforestation prevented. Although benefits of preventing deforestation could potentially be large, it is difficult to calculate so here we only calculate the potential CO2 emissions prevented, first through reducing charcoal use, then through reducing woodfire user.

Charcoal: CO2 equivalent saved by bean soaking

About 1 in 3 Ugandans use charcoal for cooking. We estimate the Uganda-wide amount of charcoal use for cooking beans through 2 diffrent...

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 Vegan Blindspot: A Presentation on Wild Animal Suffering, published by Jack Hancock-Fairs on August 6, 2023 on The Effective Altruism Forum.

I recently gave a talk at the UK's Vegan Campout Festival about wild animal suffering. This presentation was created to be persuasive for vegans but I'm also hoping it can serve as an introduction to the topic of wild animal suffering.

You can watch it here:

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: An appeal to people who are smarter than me: please help me clarify my thinking about AI, published by bethhw on August 6, 2023 on The Effective Altruism Forum.

Hi,

As a disclaimer, this will not be as eloquent or well-informed as most of the other posts on this forum. I'm something of an EA lurker who has a casual interest in philosophy but is wildly out of her intellectual depth on this forum 90% of the time. I'm also somewhat prone to existential anxiety and have a tendency to become hyper-fixated on certain topics - and recently had the misfortune of falling down the AI safety internet rabbit hole.

It all started when I used ChatGPT for the first time and started to become concerned that I might lose my (content writing) job to a chatbot. My company then convened a meeting where they reassured as all that despite recent advances in AI, they would continue taking a human-led approach to content creation 'for now' (which wasn't as comforting as they probably intended).

In a move I now somewhat regret, I decided my best bet would be to find out as much about the topic as I could. This was around the time that Geoffrey Hinton stepped down from Google, so the first thing I encountered was one of his media appearances. This quickly updated me from 'what if AI takes my job' to 'what if AI kills me'. I was vaguely familiar with the existential risk from AI scenarios already, but had considered them far off enough the the future to not really worry about.

In looking for less bleak perspectives than Hinton's, I managed to find the exact opposite (ie that Bankless episode with Eliezer Yudkowsky). From there I was introduced to whole cast of similarly pessimistic AI researchers predicting the imminent extinction of humanity with all the confidence of fundamentalist Christians awaiting the rapture (I'm sure I don't have to name them here - also I apologise if any of you reading this are the aforementioned researchers, I don't mean this to be disparaging in any way - this was just my first impression as one of the uninitiated).

I'll be honest and say that I initially thought I'd stumbled across some kind of doomsday cult. I assumed there must be some more moderate expert consensus that the more extreme doomers were diverging from. I spent a good month hunting for the well-established body of evidence projecting a more mundane, steady improvement of technology, where everything in 10 years would be kinda like now but with more sophisticated LLMs and an untold amount of AI-generated spam clogging up the internet. Hours spent scanning think-pieces and news reports for the magic words 'while a minority of researchers expect worst-case scenarios, most experts believe.'. But 'most experts' were nowhere to be found.

The closest I could find to a reasonably large sample size was that 2022 (?) survey that gave rise to the much-repeated statistic about half of ML researchers placing a >10% chance on extinction from AI. If anything, that survey seemed reassuring, because the median probability was something around 5% as opposed to the >50% estimated by the most prominent safety experts. There was also the recent XPT forecasting contest, which, again produced generally low p(doom) estimates and seemed to leave most people quibbling over the fact that domain experts were assigning single digit probabilities to AI extinction, while superforecasters thought the odds were below 1%. I couldn't help but think that these seemed like strange differences of opinion to be focused on, when you don't need to look far to find seasoned experts who are convinced that AI doom is all but inevitable within the next few years.

I now find myself in a place where I spend every free second scouring the internet for the AGI timelines and p(doom) estimates of anyone who sounds vaguely credible. I'm not ashamed 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: how 2 tell if ur input is out of distribution given only model weights, published by dkirmani on August 6, 2023 on LessWrong.

(Hastily-written code to reproduce these findings is available here. It also contains some extraneous logic.)

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: Stomach Ulcers and Dental Cavities, published by Metacelsus on August 6, 2023 on LessWrong.

(This is a linkpost from my blog, De Novo)

Recently I learned about an effort to prevent dental cavities by using genetically modified bacteria to outcompete cavity-causing bacteria. This got me thinking: why has the idea of preventing cavities by targeting bacteria not been more developed already?

The current situation reminds me of the history of stomach ulcers. Before the 1980s, doctors recommended avoiding spicy foods and reducing stress to alleviate stomach ulcers. However, once Robin Warren and Barry Marshall proved ulcers were due to H. pylori infection, treatment with antibiotics to eliminate the bacteria became the standard of treatment.

Today, dentists recommend avoiding sugary foods and brushing your teeth to prevent cavities. But we know cavities are caused by bacteria (in particular Streptococcus mutans), so why not directly attack cavity-causing bacteria?

Some potential ideas:

Selectively targeted antibiotics

Vaccines (previously tried in the 1980s, not very successful because it's difficult to get antibodies to penetrate biofilms, and also because S. mutans has several different strains with different antigenic profiles)

Outcompeting S. mutans with different bacteria (the current effort by Aaron Silverbook, which I think is promising)

Basically, what Aaron Silverbook is proposing to do is recreate a strain of S. mutans, termed BSC3-L1, that is deficient in lactic acid production. This was previously developed by a company called Oragenics, but they abandoned the effort (I think due to financial reasons). It seems Aaron's team is mostly people from software backgrounds, so they would probably appreciate help from any talented microbiologists who happen to be reading this post.

In a famous case of self-experimentation, Marshall drank a culture of H. pylori and subsequently developed gastritis. For this work, Warren and Marshall earned the 2005 Nobel in Physiology/Medicine.

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