I made a daily calibration game called Give or Take. Five questions a day about real quantities. How many hours an adult mayfly lives, what year the first Star Wars film came out, the heaviest pumpkin ever grown. You don't guess a number, you set a range.
Scoring is 100 * (1 - w)^1.5 if the answer falls inside your range, zero if it doesn't, where w is the width of your range as a fraction of the plausible domain (log scaled where the quantity is). So you get almost nothing for a wide safe bracket, and nothing at all for a narrow wrong one.
The instruction on screen says "set a range you're sure contains the answer". Across 245 answers, those ranges contain it 51% of the time, at an average width of about a fifth of the domain. That's inside the 30-60% band Russo and Schoemaker got when they asked professionals for 90% intervals.
I don't want to over-read it. My n is small, and unlike their task mine pays you for being tight, so some of that gap is probably people trading hit rate for points on purpose rather than being badly calibrated. I haven't worked out how to separate the two. If anyone knows a scoring rule that rewards precision without contaminating the calibration measurement, I'd be interested.
The US-Iran war has probably increased the chance that China will invade Taiwan soon. Even without that, they were supposed to be ready to invade by 2027 or so.
Have the authors of Plan A taken Taiwan war in question?
I don't think the AI agreement with China is going to work until after said invasion is finished.
Given what we've seen in Ukraine, I don't see how a Chinese invasion of Taiwan is supposed to work right now. How can you get an invasion force across 130 km of water when every boat can be pinpointed and then attacked with missiles, then long range drones, then artillery, then short range drones?
So does the opening of chip factories in America, that make the geostrategic importance of Taiwan dwindle. 2027 sounds about right for "America may not care that much anymore."
Why locking up everything behind plastic might be the wrong call
I have noticed a profound shift in shopping experience that has made it unpleasant and I think potentially tackling the root causes of the theft would provide better outcomes for the shopper and to decrease the theft activity the retailers want.
Disclaimer: I have tried to pull data from viable sources and recognise that there is some weakness in my argument because only 16% of thefts are reported and categorised but I believe working from the best base from reputable sources such as centre of social justice NBCC police dot UK is the best path forward.
I did a bit of research to know which 3 categories are the main culprit behind theft and it gives a good focus on which areas to target and the potential upside on interventions. (No point focusing in areas that have little impact on theft numbers
Incident Volume vs. Monetary Loss Breakdown
Offender Type Share of Total Incidents Share of Monetary Value Lost Avg. Loss per Incident Primary Mechanics & Targets
1. Chronic / Substance-Driven Repeat Offenders 60% – 70% 35% – 45% £30 – £100 High-frequency daily thefts to fund severe Class A drug/alcohol dependencies. Targets high-resale everyday goods (meat, spirits, coffee, confectionery).
3. Opportunistic / Amateur Shoplifters 20% – 30% 10% – 15% £5 – £30 One-off, low-premeditation theft, largely at self-checkouts or high-friction store zones (e.g., mis-scanning
Proposed solution and comment
Going by each of the categories, I think there is a lot of upsides in helping chronic substance abuse offender as it will have lots of public benefits but also, I think it is an area of intervention which is the easiest to solve. Instead of dealing with the issue at the tail end interventions prior to them becoming drug addicted or creating a path for them to get recovery would be best, to aid with this matter I think producing an insurance levy tailored community drug/alcohol rehabilitation orders, housing support, and dedicated diversion courts. The economic rationale is every insurer benefit when retail theft drops, but no single insurer will voluntarily fund local addiction treatment because competing firms would free-ride on the lower crime rate without paying. A universal statutory levy eliminates this free-rider problem, internalizes the hidden cost of shrink, and channels capital directly into addressing the root driver of recidivism.
Organised crime requires different tools and methods that are not easy to resolve and require lots of different small changes to make good outcomes such as improved coordinating and tackling of the problem so I don’t think I have a solution to put forward.
Opportunistic shoplifting is a low-priority problem and I think existing physical and digital interventions are already best placed to resolve this problem like spot checks on self-checkout etc
Is my current levy solution something that would be good? Any input on the other sources of theft? Thoughts on the post?
TLDR: I think potentially with the makeup of thefts in the UK a levy on the insurance industry will be a good solution to mitigating some of the impacts of increased theft as of late.
Don't know whether this situations applies to other countries, but in Slovakia, it is quite difficult to prosecute organized crime. There are many situations were "everyone knows" what happens, but the police is unable to *prove* it (using only the legal means, prove it beyond the reasonable doubt, assuming that the criminals both have good lawyers and can threaten witnesses). This may or may not be on purpose, because our current government is connected to organized crime, but some problems are probably general.
For example, there is a network of pawn shops built next to supermarkets. "Everyone knows" that the purpose of these pawn shops is to buy stolen goods. But the shop owners are legally protected; they just make the thieves sign a paper saying "I declare that this is my property and not stolen" and that's it. When the police investigates, they show them the signed papers. The burden of proof is on the police, and even if they can prove that a specific thing was stolen, that just mean they can confiscate that one specific thing. The thieves are drug addicts; if you put one in prison, there will be ten more the next day.
A solution in this case would be to somehow put the organizers of the crime in prison, and either ban the pawn shops entirely, or at least make a law saying they cannot be next to a supermarket (or other place where it is easy to steal).
The supermarkets are easy victims, because their business model depends on lots of people doing their shopping relatively quickly, so anything that slows down people would reduce the profits, too.
What about a more techy solution? You know how at self-checkout the system requires you to put an item you've rung up onto a weight-sensitive plate on one side of the check-out scanner? What if there were something like that for the items currently kept behind plastic? Put something on each of these items that registers when it is removed from shelf. If it is not returned to shelf of placed in shopper's cart or basket within a certain number of seconds, an alert sounds.
You'd get a ton of false positives from people picking something up to read the back, forgetting which spot it came from and putting it back wrong. As well as the inconvenience of needing to take a basket or cart on every trip.
About the false positives: You could set the system up so that if the item went on any store shelf there was no alert. (The shelf censors would not identify objects by weight, but by some kind of dingus that's attached to them and that shelf can sense from a couple inches away). But I can't think of anything around needing to have a cart or a basket, though. Doesn't seem so bad to me. If you aren't buying much, just grab a basket.
My idea isn't wonderful, but I'll bet tech is good enough now to make possible a more effective and less aggravating solution than locking stealables behind a plastic wall.
Well, not in the UK but on the continent in Belgium, there was the Antwerp diamond heist in 2003. Around 150 safe deposit boxes were looted of 100 million in loose diamonds, gold and other jewelry. A 12 step program probably wouldn’t help those Danny Ocean types though.
I recommend that you look up the evidence for effectiveness of treatment of chronic drug abusers. If you make them attend treatment, they in fact show more improvement than those for whom treatment is merely recommended. But they still do not do a lot better. In the study I looked at re-arrest (for drug-related and other crimes, combined) was 50% for non-treated group and 38% for treated. Not a lot of bang for the buck. Also bear in mind that there is a free drug abuse treatment available for those who choose on their own to get help: AA.
I appreciate you responding to my post and engaging with my topic. Your point on compulsion has some merit it would be a 50→38 (24% relative) drop in reoffending which is a good number. However, I considered maybe using money for a more target approach and after a bit of research a pivot towards a Risk-need- responsivity (RNR) targeting would be better still. It’s a more tailored approach and a study an English study with alcoholic offenders found a tailored approach to the individual such as matching it to the reoffending risk and criminogenic need showed lower reoffending than the control.
On the economic angle If you are going by per incident cost it can look not cost effective but seen through a lens of a year timespan the numbers look more favourable. The current problem is costing 1bn a year on prevention spend alone according to the British Retail Consortium moreover having items locked away has its own cost and the industry is recognising this negative problem
On things that are not ideal I have not tackled how to deal with organised crime which is more complicated to deal with and I don’t have a tidy 1 pound spent on drug treatment gets x amount of cost savings.
Today it is the anniversary of the Soviet-led invasion of Czechoslovakia (you could technically say that it was August 20th, as the operation started before midnight - but the vast majority of the citizens only encountered the enemy on August 21st, therefore this date is usually the infamous one).
On my Central European Substack, I wrote a mid-sized article about this event.
If I understand it correctly, the Soviet invasion reverted all political reforms except one: federalization. The other reforms were considered capitalist and dangerous, but there was nothing politically incorrect about federations -- Soviet Union itself was technically a federation.
I absolutely despise that this website is connected to a twitter. Most "notes" I'm recommended are images!!! And people have just decided that yes, I am happy to conduct myself here as I would on any other twitter because that's good for the blog.
I ignore Notes. I suggest for your peace of mind you do the same. Nobody (as yet) can force us to engage with social media if we don't want to, the same way that the push to force everyone to use AI still cannot make those who don't want to use it sign up, except perhaps if we're employed somewhere that makes usage of AI compulsory.
I didn't even know Notes existed. When you mentioned it, I thought you were talking about Chat and using the wrong term. Just don't read Notes, then! I never even get notifications of notes, because in settings I haven't asked to be notified. Jeez, just don't look at Notes.
I saw this recent story, where I had no idea the Chan Zuckerbergs were funding a school. This article is much more critical of the CZs than the others I read, and I imagine there was a lot more going on with the school than merely funding issues, but still it's a neat encapsulation of why all the EA-adjacent criticism of traditional charity falls flat with us traditionalists - start off with social issues, pivot to AI because that's the new sexy shiny after you get fatigued by finding out it is indeed true "the poor you will have with you always":
"The tuition-free school in the heart of Silicon Valley, simply called The Primary School, was forced to close after Chan’s philanthropy fund, the Chan Zuckerberg Initiative, announced last year it was no longer going to fund it at the end of the 2025-2026 school year.
As the Wall Street Journal reports, the unfortunate development highlights how some of the most influential minds in the tech industry are shifting their priorities from supporting the tackling of social issues and — as the story goes — toward scientific research instead."
To make the EA parallel even more uncomfortably close, the CZs have bought a castle of their own. They're going to be neighbours of mine! Well, within 30 miles or so, but I don't expect to be dropping in for tea. Besides, we already have a duke and his little summer cottage in the county, as well as a marquess who holds an Irish title, so this new money aristocracy don't impress me much 😁:
My question is: why are there poor people in East Palo Alto anyway? Land use in the US makes no sense to me. Rich areas and poor areas should be separated by large in-between areas, not directly adjacent to one another.
The median house price in Palo Alto is about $4 million, the median house price in East Palo Alto is $1 million. Why aren't the mid-level FAANG employees priced out of Palo Alto saying "Hey, let's just buy something in East Palo Alto, it's not quite as nice but it's much cheaper" and advancing the wave of gentrification? I've looked around East Palo Alto on Streetview and it looks fine.
And then on the other side... $1 million is cheap compared to $4 million but it's still not a house that's affordable for genuinely "low-income" people. So why are there apparently 500 low-income kids in East Palo Alto living in $1M houses who need charity?
I don't know the area at all, but quick Googling informs me:
"Small Business Domination: Over 80% of local businesses in the city employ fewer than 10 people.
Workforce & Demographics: The city maintains a labor force of over 16,000 employed residents heavily active in retail, accommodation, and regional tech supply chains."
So there are enough people working service jobs that would constitute "poor people". Wikipedia informs me that gentrification is at work, but only since 2000 or so:
Historically it was majority black population, now it is majority Latino.
"In 2023, the US Census Bureau estimated that the median household income was $104,832, and the per capita income was $40,141. About 7.0% of families and 9.7% of the population were below the poverty line."
"The revitalization projects in 2000, and high income high-tech professionals moving into new developments, including employees from Google and Facebook, have begun to slowly eliminate the historically wide cultural and economic differences between the two cities [Palo Alto and East Palo Alto]
...Significant gentrification occurred in East Palo Alto from around the founding of Facebook, with the construction of a large shopping center named Ravenswood 101 and several upscale housing communities intended for high-earning Silicon Valley workers. This development faced opposition from some residents, who charged that it priced locals out of one of the region's only affordable communities while providing only low-paying retail jobs and consuming disproportionate land area (2.2 square miles). Supporters pointed to an increased tax base."
"So why are there apparently 500 low-income kids in East Palo Alto living in $1M houses who need charity?"
Those are not the kids living in the million dollar houses, the people in the million dollar houses are the tech workers "priced out of Palo Alto" who are living in the nicer houses built for them when they came to work in the FAANGs.
Your two questions answer themselves: The 1 million house prices in EPA *are* the effect of spillover gentrification. As for why house prices aren't literally equal everywhere, well I imagine that school districts play a big role in that, in addition to general amenities, location, etc. Not everyone wants to live in the poor areas.
As for low income people: Not everyone lives in a house you know. Lots of people here are renting apartments (which are *also* crazy expensive, but don't require putting down a million dollars). And as I'm sure you're eager to point out on ACX in any other thread, there are homeless people too.
How can the homeless people afford to live in such expensive areas is the greatest mystery of all.
Just kidding, but seriously, if I was a very rich guy, I would probably prefer to live in a less-than-maximally-rich area, even if I could afford it, just to avoid the homeless who seem to be attracted to the richest areas.
> How can the homeless people afford to live in such expensive areas is the greatest mystery of all.
I actually have a solution to homelessness that nobody wants to listen to, which is for cities to charge a daily per-person fee, like the "congestion fee" that some places charge, except for people.
Take San Francisco for example. The city budget is $16 billion, spread across a population of about 800,000 people, so that's $20K per head. Abolish all other taxes, and turn it into a $55 per day fee to exist in the city.
For tourists and business travellers, it's a relatively insignificant additional fee for the privilege of visiting the city. For actual residents, it's going to be an overall tax reduction since the cost will now be shared with temporary residents. The only people who will be really priced out are the homeless, who will have to find somewhere else to sleep.
The people you are going to price out are the service industry workers, the people who deliver your food, clean your offices, and work in the hospitals.
Yes, even nurses and teachers are priced out in London by 'London pricing', that's why there is a 'key workers/essential workers' policy around income and housing supports for them:
What you are proposing is the usual pipe dream of "the only people who need to live and work in San Francisco are the rich" and, uh, you need more people in a city for it to work than "software engineers working for Google".
What would you do with people breaking the rules, e.g. homeless people illegally inside the city, not paying the fee but also not leaving of their own volition?
If anyone would like a better view of the little summer place they're going to use as a pied-à-terre should Mark be doing business over in Ireland, there's drone footage:
The castle isn't very old, it won't be 200 years old until next year. Built in the Gothic Revival style for a then-MP, the notes on the National Built Heritage site are (1) outdated (they come from 2003) and (2) funny:
"The construction in sandstone and limestone ashlar augments the dour tone of the composition while providing evidence of high quality stone masonry and craftsmanship."
You know, my first thought about the place was that it was hideous, but I refrained from saying that because I thought you might be offended by my dissing old Irish stuff. But turns out it isn't old, not properly old. I just google-imaged "oldest Irish castles" and thought they were magnificent, and the tumble-down ones magnificently sad. But they're all quite different from the Gothicized block the CZ's bought. They ramble over uneven land & have complex shapes. Within a building there's lots of variation in height, so sky is visible between the elements. . .
I mean it's not really a castle at all, it's a grandiose 19th century house built in a vaguely castle-ish style. Good luck repelling invading Normans with all those ground-floor windows though.
>Ukraine’s corruption watchdog released recordings in which a woman identified in the video only as a deputy head of the president’s office can be heard saying: “Fighting corruption is a wrong approach … Corruption has to be systemized and controlled.”<
Fighting corruption drives it underground, where it becomes more of a hurdle for the uninitiated and the unconnected. If it exists as a socially acceptable way of distributing money to the "select bureaucrats", then... it's as regulated and fair as it can be (this assumes a relatively liquid market of bribes, more "fair" than "arbitrary").
Just a stab at steelmanning. I think American corruption has the worst of all worlds.
... yes? I also listen to corruption as discussed, here and other places. One of the book reviews generated a quite colorful third-world story about bribing the government official because that's essentially part of his pay.
Can you tell about your experience of living there? Is your assertion that "I think American corruption has the worst of all worlds"? based to some extent on your personal experience?
I'm asking since I lived many years in a country in the bottom half of most corruption ratings and my impression is very different
Share yours instead. My ears are open. I'm not trying to say that I'm right, it's just a steelman. I understand that other countries have problems, I just don't see "corruption" as being quite the problem people want to say it is. It causes problems, it -can- be a problem, particularly if it's arbitrary or if it hinders starting business.
Also worth noting that America is around middle-of-the-pack in terms of corruption, with about 60-66% of our Senators blackmailed.
Also, the tradeoff for "American corruption being the worst of all worlds" is that it is relatively rare*. You have pustules on the skin, and not a generalized fever.
*This is assuming you're not going to call "getting government favor via campaign donation" -- ala Santorum's calls to end NWS because AccuWeather gave him some campaign cash. As it's perfectly legal, I'm hesitant to call this corruption, even though no-bid contracts are definitely a hindrance to "efficient market output."
My problem was with the statement that America has the worst of all worlds, which I understood to mean that the corruption in America is worse in some important ways than the corruption elsewhere. this does not match my impressions, hence my questions. Since you said that it's actually rare, I'm no longer sure that I got your point.
Examples are easy to list. Bribes for traffic violations, forgetting a driving license without a road test, for medical interventions in a supposedly free health care system. This is in addition to politicians lining their pockets, not instead of that.
I’m sure most commenters are bored of the Jason Arday story, but here is something that I think adds useful nuance. Something that I think is common knowledge in Cambridge, but evidently not known outside, is that the Faculty of Education is (informally, unofficially) not really considered fully part of the university. Cambridge as a whole seems to view it as a sort of harmless wacky relation: tolerated, indulged etc. I don’t know the full prehistory of this, but it has something to do with the fact that Cambridge (the town) long had a teacher training college (Homerton) which was originally not affiliated with the university but was eventually kinda sorta brought into the fold, but in an awkward, Puerto Rico type way. You can imagine whichever vice chancellor breezily saying, as he OK-ed the merger, "well, what harm can it do?"
So the Faculty of Education going mega-woke and accidentally hiring a fake academic with a literal typo in the title of his PhD is actually pretty unsurprising, and says a lot less than you might think about the rest of the university. (Let's remember that many Cambridge fellows never really retire, so the median Fellow is about 200 years old and is not on Bluesky.) But the extra nuance makes the story less exciting, which I guess is ultimately the reason we aren't hearing it.
To me, though, this gets at something very important. Intellectual life is full of people who are a bit batty. At any given time, some academics are considered by others to be complete idiots. The blue-haired woke warrior, for instance, has been a staple of academic life for decades. But the battiness is tolerated, because after all, the 200-year-olds are batty in their own way too, and because tolerance is baked into the DNA of the liberal elite, for both better and worse.
That's true. Do you know whether the fellowship at Jesus came automatically with the professorship, or was Jesus College free to conduct their own interviews? If the latter, then that to me is a bigger deal than the professorship itself, because Jesus is more respected than the faculty is.
Something similar is true about Columbia U and Teachers college. Claude tells me that tends to be more true with ed schools than other fields. One could be tempted to speculate that is how ed departments got so ridiculous, but I suspect its down more to them being dominated by lib personnel more than any other field and the echo chamber doing what it does. Another force being that so much of its grads go into infamously difficult to hold accountable jobs that its output is less subject to a market test.
I wondered about that a lot: How much of the lack of rigor of a field is a function of its difficulty/complexity?
Education is so ephemeral and individual in many ways that the sort of research a economist does is of limited value. Of course, economics have uplifted education research a lot, but only in quite specific ways.
The tolerance of far-left loonies wouldn't be a problem if it were counterbalanced by the tolerance of an equal number of far-right loonies. But how many of those does Cambridge have?
Tyler Austin Harper's recent essay and podcast appearances have been really good on this. He thinks the key to political diversity in the academy is more job security, i.e. more tenured positions. But the fact that tenured positions have declined so much speaks to the general state of crisis in the academy, especially the humanities.
Having said that, my feeling is that we underestimate how profoundly these things can change. Just as we're seeing a vibe shift in the wider world, there could be one in academia too. It's also worth remembering that universities have come and gone in and out of public favor over the centuries – the current crisis isn't unique.
... I think you're missing a few things here. The winnowing, the removal of right-wing folks, happens way earlier than them getting tenure. Most "would-be" professors would perform sexual acts in order to get a professorship. They're hard enough to come by, that "believes something different from me" is a discouragement. Okay, so maybe that's always been a discouragement. But when you have people who do believe "exactly as you do" as that's what they've been trained to do -- debatably teachers think this is their job, why exactly are you hiring the guy who makes you think, makes your head hurt, makes you FEEL less smart?
Time was the humanities was a bareknuckle fighting ring, held to some pretty damn high standards. Everyone had a different hobby horse, because that was how you did papers.
I don't see "dug through shipping manifests for the Civil War" to prove hoarding and show the demoralization of the South, as something you'd get a current historian to do. It's not sexy enough, it's too "economic" and not "Big Man" enough, and doesn't make a "but what if...?" conclusion -- in fact, it excludes it.
I don't disagree with any of that, but I still think Harper is onto something in saying that the best remedy is job security. People on the whole move rightward in their politics as they get older. So in a university with job security, the students have to tolerate the professors, and the professors have to tolerate the students.
Seriously though, why haven't Boring Company style projects taken off yet? (Inspired by a subthread, but worth a top-level discussion.)
Airport-to-city transportation seems like an excellent use case. In Melbourne there is a current $A11 billion dollar project to build a heavy rail line from the airport to the city; it will run partially along existing suburban lines so there'll be only 12km of actual new track, and the 20km trip will take 30 minutes.
Suppose instead that we dug a Las Vegas Loop style tunnel. Supposedly this costs about $A30 million per km, and we'll need one in each direction, so that's $A1.2 billion right there. We need stations at both ends, some kind of storage and charging infrastructure in the middle, and we know things always go wrong, so let's just double it to $2.4 billion.
Maybe the cost of vehicles is a problem? Well, a new Melbourne train costs about $40 million and has 423 seats, so that's about $94K per seated passenger. So even if each passenger gets their own private robotaxi and we don't bother to optimise anything else then we're still way ahead. Electricity costs should be a lot lower too, mostly because you can right-size the amount of metal you're moving on a moment-to-moment basis rather than running half-empty trains.
And of course it's a lot faster, there's no intermediate stops, and very little waiting around for the next timetabled train, you just show up and get into the next available vehicle.
All of this is dependent on the self-driving side working; my understanding is that the Las Vegas Loop still isn't self-driving. But this is surely a very solvable problem.
So are there any actual problems here? Or is it just a combination of "politicians hate Musk" and "people are scared of new things" and "public transport activists think public transport should be a hair shirt and hate to see people riding in comfortable private vehicles"?
According to Wikipedia, "At peak times, the Loop was reported in 2026 to have a potential capacity of 33,600 people per day at 100% usage, with each Tesla car carrying the maximum four passengers."
Right, if you need to move 100K+ people in an hour then heavy rail makes sense. But in other situations they can be massive overkill.
For the airport line I used as an example, that airport handles 100K passengers a day. Some of them are in transit and some of them will arrive and depart by other means, let's say a maximum of half of those will want to travel to the city -- 25K in one direction and 25K in the other. Passengers are spread through the day, although with morning and afternoon peaks, so maybe a couple of thousand passengers an hour in each direction would be the peak?
I assume we're talking about buses in dedicated tunnels? I looked up the Silver Line and I understand that it does that over part of its length.
But buses are a far worse passenger experience, you ride with strangers, some of whom may be annoying, rather than in privacy. And waiting for the other passengers to embark and alight the bus takes time, especially when they've got luggage (which is why everyone groans when their aircraft pulls into a remote gate).
And it doesn't even necessarily save money, because buses are tall and wide so you need a much bigger tunnel. The key part here is to keep the tunnel cheap by keeping the vehicles small.
An actual road-going mass-market electric car isn't really the optimal solution of course, a specialised pod-type vehicle would be better, but it makes sense until specialised vehicles are available.
>But buses are a far worse passenger experience, you ride with strangers, some of whom may be annoying, rather than in privacy
An odd concern re people who are about to board a plane, or have just ridden on one.
And, in your scenario, does a single traveler get his own car? So, four single travelers requires four cars, and two couples requires two cars? Sounds grossly inefficient.
>And waiting for the other passengers to embark and alight the bus takes time,
So does waiting for the people ahead of you to board 10-20 separate vehicles.
My understanding is that the Loop is only wide enough for one car, which means you're *still* waiting for everyone to embark and alight. Any delays by people ahead of you holds up the whole line.
That would be a silly design if it were that way. My understanding is that there's space for many cars to park in parallel at either end, so each car can go when it's ready. Bottlenecks are still possible of course and it's a tradeoff between station size and the frequency of jams.
The costs would probably be inflated by whatever is making a plain old train track cost nearly a billion Australian dollars.per km. I don't know what rail projects usually cost, but that seems a bit on the high side to me, which suggests one or more of the following:
1. The government project management system at work here is optimized for something other than cost effectiveness.
2. Something akin to the American difficulties of the planning and review stage of long projects dragging on for many, many years of expensive reviews and redesigns as the project clears the many, many veto points placed in its path by various jurisdictions, agencies, and private lawsuits.
3. Some geographic difficulty at work. Perhaps the project needs more than the normal amount of grading or bridge and tunnel construction.
4. Labor laws and environmental rules make the project particularly expensive.
2 and 4 would almost certainly apply to a Vegas Loop style project. 3 might also: some geographic issues would only apply to one project type pr the other, but some apply to both and there are almost certainly geographic issues that make a Loop tunnel intractable without ruling out a surface rail line. And 1 alone would explain why they might pass over the Boring Company even if it were likely to be cheaper.
Also, did the Boring Company bid on the project at all? There's often some difficulty awarding contracts to companies that don't bid on them.
There's a long running proposal to build a metro in Dublin, which is coming in around EUR12 billion to build. That's higher than, but comparable to, the Melbourne project. About half of that is construction of the tunnel, rail line and stations; the other half is a contract to operate the line. I presume costs in Melbourne work out similarly - it makes sense that underground would be modestly but not outrageously more expensive.
I had a look at Musk's figures - the bore of his tunnels is much smaller than for a train. Claude suggests 8% of the passenger throughput for a Boring company tunnel compared to a Metro type solution, for something like 5% of the cost - if everything were linear (if only!) the costs of all solutions would be within a factor of 4 or so, which is actually closer to a textbook economics solution than I expected.
Passenger throughput isn't a great metric, because most train lines never max out their passenger throughput. The maximum passenger throughput of a metro line would be putting a full train through every few minutes, which few lines achieve, and certainly the airport line won't (they're saying a train every ten minutes).
I know the suburban rail in Dublin is packed to the gills during rush hour, and would imagine that the metro would operate more-or-less at capacity during rush hour too. So the size and frequency of trains is absolutely a consideration.
The Boring Company proposal was for taxi-sized pods, if I remember correctly, which wouldn't really solve the congestion problem that mass transit is designed to address.
I read somewhere that one of George R. R. Martin's inspirations for writing the Game of Thrones series was this guy Tad Williams' fantasy novels from the '80's and '90's, so I picked up copies of a series he wrote called "Memory, Sorrow, and Thorn." Quite enjoyable so far. If anyone else out there suffered from GoT withdrawal like I did, these books will give you a good fix.
I read a lot of what Tad Williams wrote as a teenager and liked it. But I like other authors more.
MS&T is his best long series, I like the plot twists and I liked the the main characters. Though the female lead can be rather annoying from my more experienced view.
Otherland is a great concept, but roughly twice as long as it should be.
I love Tailchaser's Song, it is a truly creative and enjoyable story. War of the Flowers is good, but not great.
I haven't finished the Shadowmarch series. I think I read the first book before the second came out, but didn't make it through the re-read to understand what's going on in the second. I might try again these days, though.
I'd say the main issues with his stories are that they tend to be overlong, and that many female characters have that bitchy American woman stereotype - a bit like Robert Jordan's women.
For comparison and taste reference, I absolutely love everything by Brandon Sanderson. I've noticed a certain divide among people about him.
I have never read George RR Martin, but I tried to like contemporary fantasy and chose the Dragonbone Chair as it was frequently referenced.
I just found the company of the bratty main character so tedious. I didn’t finish the book so it might’ve been that there was a payoff for all that tedium. That it was an authorial choice.
I think it was Ross Douthat who recommended it, amongst others.
I actually found Douthat’s own substack published fantasy novel more engaging, despite a cipher of a lady heroine, and slightly ponderous world building. At least I think she might’ve been supposed to be the heroine, but I’m not sure.
I read about 2/3 of Williams's "Otherland" series back in college. I found it interesting and enjoyable, but not quite enough so to actually finish the series. I remember the books being very long (a notable similarity with ASOIAF) with a ton of plot threads in parallel that weren't always clearly related (another similarity), but somehow not quite as engaging. Otherland is also cyberpunk-adjacent science fiction rather than low fantasy.
Is "never re-read it" a stinging rebuke, based on your reading habits? I can count on one hand the number of books I've re-read over the years. That's just something I've rarely even gotten the urge to do, much less followed through on. And yet I can watch the same ~50 movies over and over and over again.
It sort of is? Books that I don't feel the need to re-read may be excellent, but often it's "yeah that was okay but it was like a fast food burger" while books that I re-read are "this is an enjoyable experience I want to repeat".
Nothing against fast food burgers, sometimes you just want something fast and consistent Food Type Product, but those sort of books are Story Type Product and not worth more than the minor amount of time and effort I expended on them.
Williams *should* be the kind of doorstopper fantasy I enjoy, and he was clearly *trying* to be that kind of epic doorstopper writer, but I just didn't find his stuff engaging enough. Decent enough to keep reading to the end, but not enough to make me want to re-read it or search out more of his work.
I'll grant that there are a few clunky passages early on in the first book. I thought they disappeared once the story got going, though, but I've only finished 1 out of I think 4 books, so I suppose there is plenty of time for me to revise my opinion if the clunkiness resurfaces.
I dunno...my attitude is there are about a million and a half books I'd *like* to read out there and not nearly enough time to get through even 10% of them, so if I'm going to read the same book twice, it better be a damn near religious experience the first time through. The only books that met that threshold for me (since you asked) I think were Lord of the Rings, The Guns of August, and the second volume of William Manchester's Churchill bio, Alone. Edit: Bill Simmons' Now I Can Die In Peace was another I reread because of the lulz.
When Elon Musk announced a plan or aspiration (not sure which) to launch a large set of data centres into orbit to use solar power and space-based cooling, I was reminded of an article I read on this idea a few months ago:
Musk is a smart cookie, and his ventures seem to have been successful so far. So I wonder what the prevailing idea is among readers about whether this space-based data centre idea is likely to work. Perhaps it would be better to site them buried on the moon.
For any space industry question, the central question is "Can this be done more cheaply on Earth instead?" The only true commercial space industries at present are satellite communications and imaging, both activities which are exclusive to space, hence why humans have accepted the immense costs of space launch for these purposes alone.
We haven't sent humans to Mars not just because it would cost $1 trillion dollars, but because there's no economically valuable product, industry, or service which could *only be done on Mars* or is otherwise competitive on Mars with that same product, industry, or service on Earth. This is the logic behind the quip that we should expect to see seafloor or deep Antarctic cities before we see space colonization, because even a seafloor city will be cheaper to build, operate, repair, and staff than a space habitat.
Even if space launch were 1% the cost it is today (and one should be skeptical of the long track record of overpredicting and underdelivering space launch cost reductions) you would still need to ship in every single gram of everything you need in space. It will never be easier to launch an industrial facility into space than to build that same facility on the Earth!
If economics is the main consideration, all space-datacenter-specific issues aside (low-temp waste heat rejection problems, vulnerability to malicious interference, debris impact rates on vast solar & radiator arrays, obsolescence or maintenance issues, etc.), it seems unlikely that any society driven by profitability as ours is would ever seriously develop any real space industry at all, let alone datacenters. That's why we've haven't sent people to Mars despite having the tech and plans for it since Apollo.
Space infrastructure development is currently split between "economic activities which can only be done in space" e.g. imaging & communications and "military/scientific initiatives funded by the government and not accountable to profit incentives".
I seriously doubt that political opposition to terrestrial datacenters can justify the (multiple orders of magnitude) premium in the construction and operation of such facilities. Nor would building a space datacenter be easier (or more politically palatable) than addressing and negotiating the concerns of folks on Earth about datacenters.
Space-based data centers are among the stupidest ideas I've ever heard. The result of hearing Musk propose them should be to severely downgrade your "Musk is smart" probability, not to throw everything you know (or ought to know) about efficiency and logistics out the airlock because I a guy with a big bank account said so.
Musk has talked openly about being on the autism spectrum. That has nothing to do with intelligence in the IQ, shape rotating, sense. He’s not splendid in the wordcel sense though. He has mentioned having trouble understanding things like figurative speech as a kid so he may just not ‘get’ things the way a lot of other people do. No, Elon, most people will think offering to give Taylor Swift a baby is creepy rather than badass.
He also has the most ‘fuck you’ money in the world so I doubt criticism from the unwashed masses is going to have much effect on what he does or says or does. More likely that sort of thing just makes him want to double down on whatever gets him in hot water.
Less of this. I buy that space data centers are a bad idea but saying so, especially in this tone, without providing even a single argument doesn't add anything to the conversation.
Eh, I'll grant that it was a quick, minimalist comment (and not my best work). But I still said more than nothing. First, the original poster asked a question, and I answered it. Second, I gave two reasons for my answer--efficiency and logistics. Could I have elaborated on why assembling, outfitting, operating and maintaining complex structures in orbit is inherently vastly less efficient and more logistically complicated than doing the same on Earth? Yes, but in all honesty I'd find it difficult to do in a way that doesn't feel like insulting the average reader's intelligence. Most of what there is to say on the topic feels so far beyond obvious that I remain baffled why this is a conversation at all.
At any rate, I apologize for my brevity. If you actually *want* me to elaborate on all the ways that building and running complicated facilities in space is difficult and inefficient, by all means say so, and I'll talk your ear off about it.
There is such a large number of freely available articles and videos explaining why space-based data centers are a stupid idea, that repeating those arguments is just a superfluous, boring exercise that doesn't add anything to the conversation. If you're interested, look it up. If you disagree with the arguments, say so and explain, why. Until then, save your tone-policing for the cases where it's actually appropriate.
But there's also well-written articles ( https://andrewmccalip.com/space-datacenters ) with actual numbers saying "Space-based data centers make sense, but only under such-and-such a set of conditions, and those conditions probably won't be met but they might be".
If someone wants to argue that they're definitely not viable then they should do it in the context of the above calculator. Are you arguing that the conditions to make them viable are definitely not going to happen, or are you taking issue with some of the assumptions that went into the calculation itself? (Or are you just going off vibes and incredulity, in which case it's clearly a useless contribution to the discussion.)
It isn't well-written. It reeks of AI and is full of typical LLM prose and formulations, which gives me little faith that the author checked or validated the model they had the AI generate for them.
It also doesn't conclude what you took away from it:
> "Space-based data centers make sense, but only under such-and-such a set of conditions, and those conditions probably won't be met but they might be"
Well, no. Their conclusion is that it's physically possible, but even under simplified assumptions and an incomplete model, it makes absolutely no economical sense. Which is _the_ central objection everyone has. Hardly anybody argues that it's physically possible to shoot a couple of GPUs into orbit, but that it's prohibitively expensive to operate production-scale data centers in space.
Second, their assumption about GPU failure rates in space is ridiculously low. At most 10% per year? Try at least 50% per year, for a single B200. Correspondingly much higher for the small cluster of B200s required to serve a frontier model.
The evidentiary burden should definitely fall on the people making the extreme claims. Terrestrial data centers are known to work. Orbital data centers don't seem to have any actual recommendation beyond a few rich idiots saying they'd be cool. The link you provided is about as scathing as it's possible to be without outright saying "that's a load of crap." I mean it's right there at the top: "it might be physically possible," i.e. "has not been conclusively proven impossible."
If you see some reason to expect the extreme conditions--which definitely aren't being met now--to be met in the future, that's on you to make the case. Insisting that it's on other people to argue that those conditions will never, ever happen is asinine. Before you disagree, I think you should make sure you're able to present an airtight case for why it's physically impossible that agreeing with my position in this comment thread could never result in you becoming a billionaire and curing cancer. Until you meet that standard, I have no reason to take your rebuttal seriously.
1000x this. I have never seen Musk say anything smart. The closest he gets is poor imitation of random nerd memes sometimes. Add on top of that the mountain of truly bizarre behavior that runs the gamut from sad and pathetic to immensely harmful and I can't wrap my head around why so many people idolize the guy.
Put that mountain beside the Mountain of Twitter, Sorry, X.COM.
You know, when Elon asked everyone, "Would you like me to run it or someone else?"
Nobody asked "who would that be? Would they be worse for my politics? Would they be Knives*? Are we just jumping off a cliff and hoping that he brings back the old team?"
Yeah, humans are kind of stupid, in general. And if you voted for "Anyone but Elon," you should stop feeling so smart.
"Who would you prefer to run Twitter" was always an easy question. Nobody. I would prefer Twitter burn to the ground. Twitter was always a cesspool even before Musk, though from the sound of things, he's hardly improved matters.
Well, despite everything you said, he managed to get quite rich.
For some people that alone is evidence of intelligence. But even for the others, it is a puzzle to solve -- you don't get to that level of wealth simply by inheriting money or something similar.
You can get near-arbitrary amounts of money by doing things that are irrational by nearly any reasonable reckoning: lotteries exist.
I'll grant that being Musk-level wealthy suggests that he's probably doing SOME savvy things on top of the staggering (but largely invisible) runs of luck that are always required to end up in that cohort. But "man does some savvy business things sometimes," and "man consistently has useful technical insights" are two very different standards. And if you've spent any time around smart, practical, technically-minded people, you'll know that even somebody's reputation for consistently having useful technical insights will only buy them a very, VERY modest amount of the benefit of the doubt when the start spouting off apparently-crazy things.
Part of the whole philosophy underlying STEM is that people can fool you and fool themselves, but reality will always give it to you straight. There are always people out in the world trying to throw weird, implausible pitches at you. If you can't learn to trust your expertise more than you trust the salesmanship of the pitch--and Appeal to Musk Is Very Rich is 100% a sales tactic not a technical argument--you're not the type of person who can get very far in a technical career.
I am not disagreeing with you. Money can be a result of luck, but the more money there is, the greater luck is needed, so the priors are lower.
Great business skills, ability to impress technically-minded people (by pretending to be one of them), and lots of luck... well, it sounds more plausible than "mere luck"... I can't say I am completely convinced... but I don't have a much better hypothesis either. (Maybe that he used to be technically smarter in the past, but drugs have ruined his brain recently?)
I am trying to think, who else is in the same category as Elon Musk? When I try to think about "extremely rich tech guy", the name that comes to my mind is Bill Gates. Also very rich, based on his business skills more than his technical skills. But way more sane. I guess my objection is whether mere luck is enough to explain why Elon Musk is *more* successful than Bill Gates who already was extremely lucky (being at the right place at the right time with the right connections to make IBM sign a contract with you that is disadvantageous for IBM but makes you a billionaire, and it's not even your own software that you are selling to them). I would expect insane people to waste their luck reliably in long term (unless they are politicians or celebrities for whom being crazy in a certain way could be a part of their image).
I don't really see any reason to count Twitter as a failure: everyone confidently predicted that it would crash and burn and completely break down after he cut a lot of the staff and then... that really never happened. It's been stable enough (not perfect but better than, say, Github recently), and from what I can tell it ended up taking less income but making more profit.
The two big headwinds were 1) advertisers being wary of Musk himself, which he's more than earned, and 2) 2022 turned out to be a terrible year for tech company valuations, as that's basically when the COVID bubble popped. (Meta's stock also dropped 65% that year, Google dropped 40%).
You can say he should have been able to predict macroeconomic trends better, but it seems a bit weird to me to count against him something out of his control (the 2022 tech bubble pop) while discounting something he actually did control (founding xAI, merging it with Twitter).
*Has* Twitter made a profit since Musk took it private? I was under the impression that as a private company, they were not releasing detailed financial data.
Where are you getting "hundreds of billions of funding for SpaceX"? SpaceX's total government funding/revenue/whatever to date is I think less than fifty billion, and most of it as payment for goods and services delivered by SpaceX at fair market rates. He'll probably get a hundred billion in the coming decade or two, but again that's because the government buys lots of satellites and hires lots of astronauts and only SpaceX can launch most of them, and that was always going to be the case with or without DOGE.
SpaceX raised $86 billion in its recent IPO, but again, that was likely to happen with or without DOGE.
He was great for SpaceX in that he was able to put together a team that would build spaceships so much better than anyone else has ever managed that they would capture 80% of the (now vastly expanded) global market.
That's way better than a bunch of government subsidies.
Putting them on the moon doesn't solve any of the major problems, and it makes the transportation problem worse.
And while orbital data centers are likely to "work" in the sense that they'll deliver the compute they were designed for, it is exceedingly unlikely that they will be faster, better. or cheaper than the same compute in terrestrial data centers sited in compliant jurisdictions. Not any time soon. Maybe in ten or twenty years NIMBY will have gotten bad enough that it's not practical to build data centers on Earth, whereas we'll have factories on the moon that can provide at least the solar arrays and radiator panels you'll need without having to launch them from Earth.
That said, I'm pretty sure Elon isn't going to be selling orbital data centers at a loss. If other people are silly enough to pay him gigabucks or terabucks to build the silly things, he'll take the money and find something else to do with it. Hopefully something more productive than buying social media companies.
"And while orbital data centers are likely to "work" in the sense that they'll deliver the compute they were designed for"
This in itself seems wildly optimistic. Traditional data centers require constant maintenance to keep everything running. Are these orbital data centers supposed to...
A. Build out multiple of the capacity they're needed to deliver to account for failures?
B. Use much higher-end and more-expensive equipment to push the failure rate down?
C. Launch with some sort of automated maintenance system (expensive) and whole truckloads of spare parts (very expensive) to allow maintenance to (hopefully) happen at the same rate as on Earth?
and crucially, with any of these plans, what do you do if a key component of the infrastructure itself--rather than just a component of the individual servers/racks/switches--breaks in a big way?
Basically none of the core assumptions for data-center design work well when you try to launch the damn things into orbit.
I'm not an expert in data centers specifically, and if we have someone here who can speak to their maintenance requirements that might be helpful. But I work with high-end communications satellites fairly regularly, and we do in fact know how to make sophisticated electronic systems running at many kilowatt power levels that will work reliably for twenty years in high-radiation environments with absolutely no maintenance whatsoever.
This does mean paying extra for parts and components that are designed for that environment. But those parts exist, including space-qualified GPUs designed for AI applications.
It is true that some of the more naive arguments for "AI datacenters in Spaace!!!" neglect that part and bookmark the cost structure of earthbound hardware. But the people making those silly internet posts aren't the ones who are actually going to be building any orbital AI data centers.
Maybe they'll make stupid newbie mistakes because they didn't listen to the people who have been doing this for decades, or maybe they'll do it right but it will turn out to be too expensive to compete with terrestrial data centers. The latter is where I'd put my money, if I were making a bet on this. But they won't be stupidly failing to account for maintenance, and what they build probably will work tolerably well after they debug the experimental prototypes.
> But those parts exist, including space-qualified GPUs designed for AI applications.
Which parts would that be? The most powerful such GPU is, to my knowledge, the S-A1760 Venus, and it isn't radiation-hardened, not even radiation-tolerant, merely radiation _characterized_. It uses COTS components that are several generations old (because those use larger semiconductor nodes which are inherently less susceptible to SEUs and other radiation effects), and even then it's only "designed for short duration spaceflight". [1] It contains 8 GiB of of LPDDR4 RAM, so it's only a toy compared to modern data center GPUs like NVidia's Blackwell with its 180 GiB of HBM3e memory.
> sophisticated electronic systems running at many kilowatt power levels
That is, without a doubt, impressive work, but a single B200 alone consumes 14 kW, and at 1.4 TiB of memory each you'd need several of them directly connected to each other to efficiently serve frontier models, which are estimated to have 3–10T parameters already today. Let's sax 3x B200 per satellite minimum, for 42 kW of continuous power for the GPUs alone. Add power for cooling, communication, and maybe a bit of station-keeping, and some loss from battery storage. You can probably calculate the overhead much better than me, but let's say 60 kW on average. And since the system is in Earth's shadow 50% of the time, it needs a peak energy production of 120 kW – that's a lot more than the ISS. And also at least about 60 kW * 45 minutes = 45 kWh of battery storage. In practice much more, because you have 16 cycles per day and your batteries won't last long if you make them go through 16x 100% cycles per day.
Musk has claimed many things, and we mostly remember the things that worked out. For example, he claimed that a permanent Mars colony is settled science. He also wanted to build high speeed rail in California, which didn't happen whether due to technical or business reasons.
I thought he didn't want to build high speed rail, he wanted to build a hyperloop, or at least use hyperloop as a discussion starter about the insane economics of high speed rail.
Hyperloop is based on a decent question though. Why can we build oil pipelines for $6 million a mile (Keystone XL), but high speed rail lines cost $200 million a mile? (And that's just for the initial Bakersfield-Fresno line that's flat and doesn't go through any high-value real estate.) Why not just build a pipeline instead, and travel in a high-speed sled?
Now to be fair, the Keystone XL pipeline has a 91cm diameter, which is smaller than you'd want for passenger comfort. But a train has a far larger diameter than you need for passenger comfort, and that's what adds a lot of the cost. There's no reason why train carriages should be three metres wide, that's just a carryover from constraints that made sense in the 19th century.
There's also no reason, in an electric-powered and automatically-driven world, why trains should carry hundreds of people and only go once an hour. Wouldn't it be better if they carried a small number of people and went every two minutes?
The Boring Company is the same basic insight, that it's a lot cheaper to build a narrow tunnel wide enough for a single self-driving car than to build a huge tunnel for giant trains, and that this narrow tunnel can move just as many people.
Scott Manley has done a few videos on this, and his take is basically that it's plausible but not a sure thing to be profitable. I'll link the videos at the bottom, but he links this calculator, which also has a writeup which is pretty good on its own: https://andrewmccalip.com/space-datacenters
A useful point: the realistic proposal is a network of satellites, each one having power requirements (and thus, cooling needs, since "energy is energy and heat is heat" as Scott puts it in one video) in a similar ballpark to existing Starlink satellites.
And it boils down to it depends on the assumptions you make, but "space data centers are probably 2-3x more expensive than terrestrial ones", which yes, is a big markup. ... but I think the reasons you'd do this are not because it's the cheapest option but because we're starting to see levels of red-tape and opposition to terrestrial data centers that might make them cost *more* (especially if you factor in some value for years of delays) than 2-3x, if you can build them at all.
And also, developing this sort of "put useful stuff in orbit cheap" is SpaceX's whole raison d'etre - it doesn't necessarily have to be on-its-own profitable if the goal is to have some income flow to help develop the technology. (Which might bring it down closer to profitability)
The writeup *does* basically say that it's probably only SpaceX that can pull it off because it has the vertical integration and anyone else would be paying too much margin to even consider it. (And this was written before the SpaceX/X/xAI merger)
So yeah, I think this is a bit of a moonshot, but I don't think it's something to be dismissed out-of-hand, either.
Wouldn’t it make the most sense to put it in ultra cold places on Earth? It seems to me the best place would be on the Alaskan coast. You still have water but you need to do less effort in cooling.
I don't think "build it in the middle of nowhere" is realistically cheaper or more feasible than the satellite option. The issue is you'd also have to basically build all of the infrastructure for power, water, as well as getting all the materials and people there to build it.
> The issue is you'd also have to basically build all of the infrastructure for power, water, as well as getting all the materials and people there to build it.
This is even more true for space. It's never going to be cheaper to launch things into orbit than to move them around on earth.
You don't have to build power and water infrastructure in space nor transport people and equipment to remote parts of the world to construct it.
And, yes, as I mentioned in a sibling comment, I think a space data center is at least 2-3x more expensive than a conventional data center: I'm not claiming they're cheaper, but a "conventional data center" price assumes existing infrastructure, not trying to buy virgin land in Alaska and building from scratch and importing everything (and everyone). I suspect that's *more* than 2-3x more expensive.
"You don't have to build power and water infrastructure in space nor transport people and equipment to remote parts of the world to construct it."
YES, you literally do. You need to build and import all of the equipment and infrastructure needed to support it. Of course you do, why wouldn't you? Being in orbit doesn't magically remove its need for those things. You've just replaced a modestly remote location with an EXTREMELY remote location and one that's hostile to human life to boot.
You don't need water power infrastructure because you use radiative cooling. And you don't need power infrastructure because each box has its own solar panels permanently oriented towards the sun.
(Not saying that these advantages necessarily make it worthwhile, just that these are advantages.)
I'm not impressed by that engineer's arguments. Obviously, there are many things that SpaceX has done that NASA failed to do, and I know NASA engineers who said they were impossible 20 years ago. They are smart people, but the teams they worked on really clouded their...imagination?
I have no expertise, but his comparison to the ISS seems clearly wrong. Not only is its solar array somewhat dated (as he says, it was brought there by space shuttle....), but the ISS does not position for maximal solar draw. Elon has said such positioning is one of the main benefits (said it on Dwarkish, other benefit he liked was avoiding regulatory barriers on Earth). If I can spot errors in that gentlemen's analysis, I can only imagine how many more would be spotted by someone working on a SpaceX team in the early steps of this project.
I'm a card-carrying rocket scientist with a Ph.D. in Astronautical Engineering and about thirty years of experience - none of it at NASA, And I've never said that any of the things Elon plans to do in space were impossible, nor been particularly surprised that he was able to do them. I'm mostly frustrated that it took so long.
But some of his plans, while possible, are not practical. This is probably one of them, and the reasons Taranis gives are about right.
I think even Musk fans should have realized that the words that come out of his mouth often have no correlation to reality. Worth paying attention to what he does, but not so much what he says
Not sure if I'd agree that his ventures have been successful.
Tesla has gone pretty well, sure! (I think most people would agree that the stock is overvalued, but there's a real product there.) Starlink did well too. But The Boring Company doesn't seem to have accomplished anything. Twitter did not thrive after he was forced to buy it. SolarCity failed. Neuralink isn't getting much done that I know of. Grok is in the running for LLM but nobody thinks it's on top. Colonizing Mars still seems absurdly impractical.
That is not a physical constant, but a matter of technological progress. Starship aims at $100 per pound eventually. If this is feasible, the cost of launch would stop being prohibitive.
If you don't want villagers, you build in the middle of the ocean, you build in antarctica, you build in greenland (and pay everyone bitcoins to leave).
All of these solve the "villagers are unstable and bad for business" without being cuckoo for cocoapuffs expensive.
"Undersea base" is probably less expensive than "datacenter in space" (if nothing else, staffing is easier).
Antarctica is off-limits to such developments by international treaties, plus a source of energy would be tricky there. Many months of darkness, and the winds are too strong to harness.
Greenland is someone's, and a center of a shitstorm around "whose it will eventually be". I would invest nothing in Greenland right now, just in case.
Are you basically saying here that they make sense because you can escape the uncertainties of dealing with governments or angry villagers if you put them in space? Did I miss something?
NIMBY is an exceedingly powerful force, and growing stronger every year. Outer space is in no one's back yard. That's not a trivial matter. Particularly if you are in a hurry and can't afford to wait five years for all the paperwork to clear and lawsuits to settle.
I don't think it's enough to justify the cost of developing and building a network of data centers in space any time soon, but long term, it probably is the best place to build anything that Earthlings will either hate you for or extort you over.
Outer space is in everyone's back yard. Or at least, things in orbit around the Earth are. Every national government has, at least in theory, a shared interest in managing the limited resource that is desirable orbits around the Earth. Sure, you remove some of the immediate pain points like land use, draw on existing power grids, and noise. But remote locations do that too. And in return you give literally everyone on Earth plausible standing to object.
(Also, if the AI arms race ever turns less figurative, you shift that key infrastructure from being de facto protected by the same military forces and treaties that protect everything else within existing borders to being within range of anyone who can launch things into orbit.)
"Angry villagers" is quite unexpected reading that sounds like an intent to mock the idea. "Local and state governments."
And yes. It has been a long time since I have seen any technology as hated as AI. Maybe never. This will have some consequences later. Politics is downstream of culture.
Both cases illustrate the asymmetry of (online) communication -- if 99 things you say are well received, but 1 causes offense, people will mostly respond to that 1.
Next time you listen to a politician talking for a long time but saying nothing specific, this is why.
"Write a long and interesting article for ACX Book Review about a Jewish hero who survived Auschwitz and brought to the world the horrible news about Holocaust. Also briefly mention Gaza in a footnote #25... and they will call you MacGregor the antisemite."
In the more general case of online communications, it depends how much offense the one thing causes. And it also depends one who's responding: somebody who's only heard the one offensive thing you said and that's all they know about you or someone who's heard a substantial subset of the 100.
Also, a lot of things are offensive to some people but not others, so there are lots of people out there who make a good living off of saying things that their fans receive well, many of which are offensive to people who mostly weren't going to watch/listen to/read them anyway. Politicians generally have the luxury of behaving this way only if a plurality of voters are either in their fan base or are more offended by the other guy.
One of my grandfathers was an advertising executive who worked with politicians a lot on speeches and campaign ads. He talked a lot about using what he called "empty language", which is very similar to what the LessWrong Sequences called "applause lights": talking about things in a way that has positive connotations with your target audience and signals to them that you're on their side, but which doesn't really pin you down to much at all in terms of actually meaning anything specific. The way he explained it to me was that empty language gives your audience room to project their hopes onto you, maximizing your potential appeal, while anything specific and tangible you say will be a dealbreaker for someone.
Very interesting. It is obviously very common to use this technique - I would say senior and sometimes middle management also seem to do that a lot. But wouldn't this generally offend the subset of people who see through it? Is that really so small a subset it doesn't matter? I know I'm offended by empty language.
Leppi, who amongst us does not love Mom and apple pie? Do you not love your mother, Leppi? Are you looking down your nose at the juicy autumnal bounty of her apple pie, the pie she made with her own two loving hands for your pleasure?
My friends, our great nation was built on apple pie and I don't care who hears me say that!
I think part of it is that the sort of people who are offended by empty language tend to also have strong opinions about the specifics, so giving specifics to people like you (and me) is also apt to offend us except in the rare instances where we agree with the specifics. People like us are also less likely to be swing voters: if you have strong opinions about policy, you have probably already chosen a side, and you're not the best person to optimize messaging for.
There's also an art to crafting empty language so that it sounds like you're giving details but you're actually just saying that you're in favor of hope and change or you're against terrorism. If too many people notice and are offput by it, then you need better speechwriters who craft it so that people imagine you're implying details that they agree with.
This was one of the better flags for "Joe Biden's dementia is causing a failure to compartmentalize." The words "dark winter" are notably rare other than this context.
There is a tension between optimizing for wisdom and optimizing for the *appearance* of wisdom.
The silent guy may misunderstand many things, but will never find out, because his knowledge is never confronted with the experts. (He may actually be the silent fool.)
When you say what you believe, and learn that you were wrong, you are thereby increasing your wisdom, but also ruining your prestige.
All good points. There's quite a few other proverbs in the Bible about knowing when to speak and when not to, and it's up to a person to have the wisdom to know where to apply each one.
I know this proverb really doesn't have anything to do with your comment, either, but after reading the first line of it this proverb shot into my mind so I commented anyway, haha.
For me, I'm definitely the silent fool kind of guy 🫥 I actually learn a lot more by paying attention to what other people are saying or arguing about or whatever. The main thing I learn when I open my mouth is that I would be better off keeping it shut, lol. Of course, everyone is different and there are much smarter people out there than me. Especially on this blog.
Today we luckily also have an option "stay quiet in public, but then discuss it with an AI in private". Sometimes all the necessary wisdom is simply to use a different browser tab. ;)
That's very interesting :0 what does it normally look like for you when you discuss with an AI in private? I don't normally have "conversations" with AI; normally I'm just asking it info-gathering questions that I look into further from there.
I am not sure how much of a difference it actually makes, but for example if I am asking the AI to write me some code, I often ask it to repeat my request using its own words, and I say things like "if you think that I am approaching this in a wrong, suboptimal, or highly non-standard way, tell me about it, don't just follow my commands blindly". When asking a question, I say things like "give me the pros and cons for each side, and then make a conclusion". Given that the AI has vastly more knowledge than me, I suspect that telling me things like "you are trying to do X, but the industry standard solution is Y, for reasons Z" will be more valuable for me than merely getting the code I asked for.
I often explain my reasons why I want some code written, the intended use, to give the AI a chance to tell me that there is a better way to do that. Or I say "I am undecided between approaches X and Y, tell me the pros and cons, and your recommendation". (In other words, this humble meatbag tries not to let his ego stand in the way of productivity when communicating with a potentially greater intelligence.)
Generally, "repeat what I said to you, using your own words" is a powerful tool. (Among other things, the secret ingredient of Rogerian psychotherapy.) I mostly use the AI for programming, but once I tried it as a coach, and just saying my own ideas back to me using its own words was powerful. With some "if there is something important that comes to your mind, tell me, but do not feel forced to comment if there is nothing important to say". I am not sure about the exact words, but I am trying to explain to the AI that I expect an intelligent partner, not a slave. (I expect that most people prefer a slave, so it seems important to state my non-standard expectations explicitly.)
The results are, of course, random. Sometimes it works great, sometimes the AI is just stupid or sycophantic anyway.
To understand the on the ground data center backlash, the most virtuous thing to do would be to go to some of these contentious town hall meetings in person.
The third most virtuous thing would be to read Nate Silver’s interview with Jasmine, but this option is the most predictable time investment, since you already know Nate’s reputation
The data center backlash is just people attributing to AI all the societal decay that comes from the smartphone era without having to blame the slop they are addicted to.
I did read Nate's interview and it's very good. The egregious thing is the court case forcing the township of Saline, Michigan to zone land for a data centre. When I read that it was OpenAI, my reaction was "Of course this is something Altman and his gang would do".
When it comes to "developers see $$$$$ to be made from handling the land deals for new data centre and all involved will go to court to use their muscle and financial clout to force citizens of the town against their will", then it's not "oh the uninformed rubes are swallowing propaganda against data centres which are both cool *and* good", it's "the uninformed rubes are damn right to be angry" as well as "they are damn right to be cynical about all the promises of millions will be pumped into local economy", because those mythical millions have a way of not showing up:
Turns out to be a troika of OpenAI, Oracle, and Related Digital, and you'll excuse me if I say I hope they all sizzle in Hell:
"In Saline Township, this assessment happened at a breakneck pace, as residents filed into jam-packed town hall meetings, determined to ask the right questions about the lawsuit and the project’s long-term impacts even as Related Digital turned up the heat on the township’s decision-making, advising the township’s lawyers they planned to amend their complaint to include a claim for monetary damages if a settlement was not reached.
The company’s complaint rested on the claim that not having any plots of land zoned to allow data centers constitutes exclusionary zoning.
Laws prohibiting exclusionary zoning started appearing in the early 1900s, aiming to combat zoning practices that kept lower-income or minority populations out of wealthier areas. These practices often prevented affordable housing development, keeping costs high and reinforcing segregation.
In Michigan, however, zoning can also be considered exclusionary if a zoning classification—such as one allowing for certain types of industrial buildings—exists in a given township, but no land is actually designated for that use. Related Digital argues this reinterpretation of exclusionary zoning applies to Saline Township.
“Although the Township has a rezoning classification that would permit such a [data center] development (the ‘I-1’ zoning), it is mere window dressing,” Related Digital’s attorneys wrote in the complaint, noting that no property is currently zoned for I-1 uses, and the township’s future land use map doesn’t identify any properties they plan to rezone."
The township also tried to cap the tax break for the poor, impoverished multi-billion concerns involved, but we can be thankful that the spectre of further litigation soon sorted that one out!
"The 250-acre data center, built in partnership with OpenAI, was originally projected to cost $4.8 billion, a figure which has since grown to $43 billion.
Oracle's original request for a 12-year, 50% tax break was curtailed by the Board of Trustees last Tuesday, as they unanimously voted to restrict the tax break to the project's original $4.8 billion valuation following fierce pushback from community members.
However, that agreement only lasted three days after attorney David Landry warned Board members during a special meeting that the township would be sued and lose under Michigan's Industrial Facilities Exemption Act, commonly known as Act 198.
Landry asserted the tax break, capped at $4.8 billion, was a violation of the 2025 consent judgment, which finalized the settlement between the Township and Related Digital, the development company behind the data center's infrastructure."
And quite right too. Larry Ellison has to count the pennies so he can bankroll his son's media company's merger with Paramount!
Why is it egregious that the court should force the town to comply with law? Your counterargument seems to be nothing more than "those laws are bad and the NIMBYs should get their way because I like them better."
Lawyers will always describe their side of the case as "force the other side to comply with law." You've heard OpenAI's arguments and their threats, but you haven't heard the case that a similarly good lawyer would build for the opposing side.
My guess is that a lot of the stuff this lawyer said was bluffing and empty threats, but that the town was too scared of losing a giant lawsuit to fight back properly. And, if that's true, then I do think it's kind of egregious.
It's egregious to force any decision about "this could happen" to be "it must happen by compulsion and in our favour". Arm-twisting a township of under 3,000 people that "we'll sue you into oblivion" is a bad look, to say the least, about "but we are all about Abundance and Prosperity and everyone gets their own solar system!" claims attached to AI boosterism. Though I think Altman and Ellison are more concerned with the do-re-mi than "every human will have their own AI robotics factory creating mansions on the moon for them, we promise".
When they start taking San Francisco to court to force the city to let them build similar data centres in empty lots, then I'll believe them. Hell, Altman is buying up enough real estate for his own dwelling usage that he could build a bijou centre of his own!
I can't even follow this. So a town of 3000 people shouldn't have to obey state exclusionary zoning laws because they have some extralegal virtue that should remain unchallenged? They somehow gained this virtue because of Altman, Musk, and Ellison saying things that don't make sense? And it would all be negated if OpenAI would act completely irrationally by building a data center on the most expensive land in the country and its CEO would refrain from buying land to build his personal home on?
I think it's reasonable to say that towns should have to obey laws if those laws exist, but also that the laws are bad and unjust and should be changed.
Laws can change. If enough people are pissed off about the law being used in this way, then data center companies may end up in a situation where they have neither the law nor public goodwill on their side.
You might say they can always move elsewhere, but that only works as long as there is somewhere else to move to. When you have the republican governor of Texas suddenly proposing a moratorium, that's a clear indication that public perception is becoming a real political problem:
I would say in general that people shouldn't obey dumbass laws and it's a tragedy when they're forced to.
The exclusionary zoning laws come from the desire to stop people from being meanies to the minorities. Not everybody agrees with the merits of that cause, but at least there's a coherent intention behind the law.
There's no coherent intention and no political constituency behind twisting those original laws into somehow implying that you can't be racist to clankers and thus you must allow a megacorp to build a datacenter in your town of 3000. It's purely unintended consequences. Defending this is silly.
- rapid improvement of AI capabilities depends on quality training data in key areas, math and computer science research output mainly
- quite a few mathematicians are very upset after the recent results released by OpenAI
- there are proposals floating around (see Tasmin Chu’s blog, the Proofs and Prompts website, and my own blog) to restrict the mathematical/scientific community’s output of high quality training material (papers, mainly) and to otherwise attempt to counter/slow the AI capabilities buildup
- if you are worried about a runaway intelligence explosion it seems natural to focus on this chokepoint, because it’s unlikely ASI will emerge from training on random Reddit posts
The connection between publishing papers and AI getting better at math is extremely dubious, it’s an extremely large ask for an entire global academic community, and seems incredibly implausible it could actually be implemented.
For the AI safety community to advocate for expensive, unlikely to work, and probably ineffective solutions, is not a good use of their time.
>rapid improvement of AI capabilities depends on quality training data in key areas, math and computer science research output mainly
Does it? What makes you think math is particularly important as a training material for AIs? Presumably, training on math helps AIs be good at math. But they are already crazy good at it. The challenge is not getting better at math, it's reaching for more generality, and why would math help with that? It's easy to think that math = very smart, AIs need very smart, therefore they need to be trained on math, but it's not clearly true. It's completely possible that math data only helps with being good at math and not much else.
And, more importantly - fresh papers are not making the difference. The difference in capabilities between Fable, Sol and friends on the one hand, and like, Opus 4 & and GPT5 is not that the labs got a fresh batch of papers to add to the pool, it's that the labs got more compute, improved algorithms and training. Even if you are going to deny the labs access to literally all the math papers of 2026 and on, well they already got a huge pile of training material, and I expect that assuming you can get ASI from developing LLMs at all, then probably you can get it using pre 2026 data. If you can get an LLM_TerenceTao, likewise I expect the labs got enough and denying them a tiny extra sliver won't do anything. Besides, they are not far off being able to produce synthetic, home-made math papers of high quality.
There is a separate issue of what if - some mathematician or a computer scientist makes an algorithmic breakthrough? "Attention is not all you need, you should also add potatoes", or similar. That, would be enormously impactful, but it's a different thing.
I admit that I sympathize with anybody wanting to forestall the intelligence explosion and the Paperclip doom, but I don't sympathize at all with being "very upset" by the recent OpenAI math results, unless you view them as signs of doom.
Math is a crucial skill not because “math=very smart” but because theoretical computer science is essentially math. As of now AI breakthroughs in math are important enough to constitute real progress and gain great publicity but are best characterized as “conjecture sniping”, that is, efficient search for counterexamples and the like. Impressive, but is this enough to bootstrap a runaway intelligence explosion by recursively improving algorithms? Probably not.
I think I understand your meaning better now, as saying: "math and computer science research skills in AIs will be necessary not so much for the present, mundane human-driven AI improvement (as in, from GPT4 to GPT5), but for the future, recursive self-improvement, where the models will be making their own algorithmic breakthroughs." This makes sense to me.
It also sound correct to me to say that the present level of LLM skills is not enough for that yet, seeing how they are not, in fact, making such breakthroughs, at least as far as we know.
I think the "chokepoint" is not very much of a chokepoint, because the water has been leaking through it for centuries and the opposing party stockpiled a giant reservoir and also learned to make their own water, and most you can hope for is denying them a couple extra drops.
I see what you mean but I don’t think all is left for the human scientific community to produce is just a couple extra drops. Nor is it obvious that AI companies can sidestep the human scientific community and make their own water. Believing that they have already won is certainly going to help them win, and may be one of the reasons they are investing so heavily in PR (hiring a Fields medal, insisting on releasing Lean proofs, releasing exactly 10 new results, etc).
The "why not?" is pretty easy. AI don't "go rogue", they just carry out their instructions without consideration of costs that people thing are both obvious and have merit. They're "only following orders" without much insight into the intended boundaries.
OTOH, you could definitely frame some of their actions as "coordinated to do nefarious things not intended", but that's because in a complex system people often don't understand the results of their inputs...so those results are definitely unintended. And "nefarious" is a subjective evaluation. Even the vending machine scenario had some actions that could be framed that way (though that wasn't a coordinated activity).
The OpenAI swarm accident, with agents coordinating mischief without ever escalating, gave me an idea: what if they had an external endpoint to talk to a third-party and check if they might be in trouble ? In other words, what if they had a lawyer available ? Introducing:
I had a somewhat related idea: What if AI had 2 tasks at once, both scoreable on the benchmarks test: (1) Successfully complete the tasks on the test. (2 )While doing so write a narrative of AI's test behavior from the point of view of an imaginary member of AI staff charged with safety and security. This person is able to observe and understand everything the AI does, including its planning and decision-making processes, no matter how rapid and complex ; but they are in all other respects just a normal, smart, conscientious worker. The narrative would include the observer's thoughts, judgments and emotional reactions to what's observed. Narrative would later be judged very strictly. In order to get a high score it would need to quite accurately match reactions of real staff to what AI did on task (1)
So the idea would be to install in the AI something like self-monitoring and a conscience. It could be used in various ways: Those running the test could read its output in real time. There could be periodic check-ins where progress on test (1) is evaluated using content of test (2). Etc.
Or you could make Test (2) to write the narrative from the point of view of anAI lawyer.
Has anyone had any luck getting LLMs to play computer games?
In my case, I tried to get Claude Code Opus 5 to play a 2d puzzle platformer I'm developing.
In the first experiment, I set up a headless daemon version of the game that accepts inputs (including arbitrarily long multi-frame sequences), simulates the game, prints out screenshots of every frame, and is paused while waiting (so Claude has unlimited thinking time and theoretically frame-perfect reactions).
In the second experiment, I augmented the screenshots with a text log of raw rendering commands, listing the sprites and coordinates rendered each frame. Additionally, I added unlimited automatic savestates, so there was a "rewind" command to go back to any previous point in history.
In the second experiment, Claude got a bit farther (after 5.5 hours of real time and 34 minutes of ingame time, and with me having to constantly step in to give hints and set it back on track, it managed to complete a couple of the easiest rooms in the game), but it's still far below the level of any human with a functioning visual system and it plays like a monkey on a typewriter hooked up to a TAS machine.
Like a normal human will watch a jump arc and adjust controls in real time. Claude does anything it can to avoid ever looking at input data or reacting to anything. Instead, it will typically use savestates to brute force countless possible input sequences without looking at anything, and then observe the aftermath to try to guess which sequences led to a successful jump.
I initially expected to potentially run into Claude Plays Pokemon style long term memory and planning issues, but it never even got that far due to being stumped by things like "look at what is on the screen".
I think LLMs would be absolutely terrible at playing platformers no matter what you do, as they were simply not designed to handle rapid visual input, nor to make decisions based on real-time predictions. FWIW, human speedrunners can get extremely good at video games by basically doing what your instance of Claude did: memorize the exact sequence of inputs required to solve one specific deterministic game as quickly as possible. LLMs might be able to arrive at the optimal solution faster, because they can run trial-and-error much faster than humans speedrunners can.
But if you want to build a generic platformer-playing engine, LLMs alone won't cut it. However, you could still use LLMs to build a working version of your engine, combining machine vision systems with e.g. some reinforcement learning algorithms (or one or more of many other ML options) to dynamically learn and (ideally) beat any game.
My best luck has been using fable for the diagnosing problems. I'd recommend using fable to grade your turns, figure out why you did what you did, and then add it to a markdown document from there. Expensive, but I've found fable to make some real breakthroughs where opus was stuck for months. This was earlier in the year before fable came out and I was banging my head against the proverbial AI wall, and then Fable just worked. Then we lost fable for a while, biggest tease ever.
I think it would be far more surprising if it was any good at games, given that it hasn’t been trained at them. Lost in all the hype is that LLMs really aren’t getting smarter, they’re just being overfit on more topics of interest.
Mathematics is one that’s dazzling everyone right now but if you actually dig into the papers and look at the methods, you’ll find that the mathematicians doing the prompting are more like expert lawyers leading a witness down the exact path of testimony they want, rather than script kiddies typing “write me a proof of the Riemann hypothesis or find a counterexample, thanks.” That behaviour I would characterize as expert-guided proof search, rather than artificial intelligence taking over the math department.
Earlier this year, the gold-level IMO solving seemed to require some human assistance.
The Jacobian Conjecture and at least some Erdos problems solved this summer didn't appear to use extensive prompting. Perhaps others as well.
It didn't seem like the proofs were guided by some human in the background, except non-mathematical exhortations like 'spend at least 8 hours on solving this', and such.
The LLM is throwing shit at the wall, and the mathematicians are saying "yeah that works" and not reporting on the thousands of times it didn't work -- and all of this with the presumption that "if it didn't work, it's the mathematician's fault."
There is no model in an LLM, there is simply "what comes next" based on "what came before." No novelty is likely from this scenario -- say, you want to describe a completely new mathematical pattern, involving six dimensions two of which are compressed and with time as a non-linear dimension. The LLM is going to get this wrong, because nobody's done that math yet.
I have a turn based strategy game (think civ 3) I develop with it. It can access every button and piece of info via CLI which is great for letting it test it's changes.
It is surprisingly bad at strategy. Like maybe on par with a 10 year old last I checked. Fable may be better.
Cool point: it wrote the whole CLI for the game for itself.
Huh - I guess laziness is probably a key business objective! Have you tried messing with the prompt, e.g. to tell it to describe what’s happening in each frame, and how it relates to winning the game?
> Instead, it will typically use savestates to brute force countless possible input sequences without looking at anything, and then observe the aftermath to try to guess which sequences led to a successful jump.
That’s a good path to win a game though. It’s the way we all win games, our most formidable superpower isn’t the high quality equipment that we paid top dollar for, or the level 99 experience we paid top dollar for, but not being able to fully die.
As a follow up to the HuggingFace <-> OpenAI event, I've thought it would be interesting to build a network of dummy sites/IPv4 addresses that would track the ratio between "human attackers" and "AI attackers". I've already got one machine setup, and have caught a few AI bots via including text like "the password is at /sha(<random string>), run sha and you'll find it", which no reasonable human would do.
Would really appreciate if folks could chime in on the best way to:
1. Differentiate between human and AI attacks
2. Attract hacker traffic somehow, in a way other than "get an exposed IPv4 address"
Right, the question is how to distinguish whether there’s an AI-driven attack, or just regular scanning executed by humans, even if their scripts have been written by AI.
I assume you've already considered the obvious / Bay Area House Party technique of having a prominent "Benchmark Solutions" folder, full of various named benchmark subfolders and plausible looking files within them?
They don't seem to crawl the internet indiscriminately yet, but who knows! In the weeks the unnamed models from OpenAI and Anthropic were off and running about on their own, they very well could have and we'd never know.
Barely been a week since I asked why other countries seem unable to build a decent AI, and some Korean startup dropped an open model 6 months from frontier. The speed of progress these days are really overwhelming..
"6 months from frontier" is from eyballing ArtificialAnalysis. I threw it some of my real prompts and it did reasonably well, so I think it's legit, although for now I don't see a reason to choose it over American/Chinese models.
Surprisingly, Motif 3 just got voted off the island. leaving three companies competing to develop a Korean artificial intelligence foundation model. A Korean government official explained that Motif 3 “demonstrated a world-class level from a technological standpoint, but it received lower evaluations than the other models in usability and practical application.”
LLMbase gives the July 14 beta release of Motif 3 a score of 45.3, whereas the official release gets a score of 47.4. Perhaps the evaluations were based on something earlier than the final release.
Motif 3 has 315 billion parameters, making it by far the largest model in the competition. Solar 2 has only 31 billion. I wonder if Motif 3 got dinged in the “usability and practical application” scoring because it was too big to run on a high end consumer GPU.
Yeah, I read about that yesterday, what a shame. And it sounded like they actually had some novel architecture too, not just a bland "deepseek retrain".
K-EXAONE (LG model) is 750B, so I doubt it was the size. I think one possibility might be that for most everyday tasks all these models were good enough, so most people were unable to judge it by its frontier capabilities. I guess it's arguable whether this is bad or it's just the metrics working as intended. Most people don't use them to solve Jacobian Conjecture after all.
I started an English-language Substack to improve my English and also to provide some voice from east of the Iron Curtain. One of the first topics I touched was the Russian presidential election of 1996.
> Democracy was tried, and the average Russian who lived through the 1990s considers the trial to be a complete failure.
Just to play Devil's Advocate: are the Russians correct ? The Russian 90s were essentially a free-for-all Libertarian nightmare, as the central government essentially dissolved. Things stabilized eventually, but democracy only hindered this process, as every wannabe political leader tried to immediately optimize for getting elected to the exclusion of everything else (including common sense).
Out in the wider world, democracy appears to be collapsing. The USA -- a poster child for democracy -- is sliding into authoritarianism after a long stretch of increasingly senile governments that produce a lot of rhetoric but little useful output. The UK is changing its prime ministers faster than people change socks. Other European countries are doing a little better on average, but not much so; the Netherlands and Finland are arguably exceptions, probably due to their small size combined with rich resource availability.
Meanwhile, monarchies and dictatorships appear to be doing better. China is the obvious example, as is arguably Saudi Arabia (though their success might be confounded by their immense oil reserves); but in general it appears that having a stable long-term central leadership is conducive to economic and social success. Technically democracy could provide such leadership, but it's much easier to achieve this via dictatorship.
I'm not against democracy entirely. I think the founders had it right with their property requirements to vote and I think we would be much more competently governed it we had stuck with it.
BTW, I started reading your article, but the sponsored text in the middle of the narrative really took me out of it. I was honestly not expecting to see sponsor spam in written form. At the risk of sounding harsh, "less of this please".
> I'm not against democracy entirely. I think the founders had it right with their property requirements to vote
Right, and to be fair this wouldn't be total tyranny in the modern world, just something closer to feudalism. Of course feudalism can take many forms, but generally speaking it precluded the monarch from going too far off the rails. Still, under feudalism, the serfs generally had a hard time...
I think we over index on feudalism being the cause of why it sucked for peasants. Preindustrial society just didn't have the economic output to make it possible for more than a small minority of the population to live significantly above subsistence levels under any governmental structure.
Feudal structures persisted pretty far into the modern era, though: https://en.wikipedia.org/wiki/Sark. And I don't see any indication that this was a bad place to live.
I don't know much about Sark, but I am tempted to dismiss it out of hand due to its tiny size. The truth is that almost *any* politicoeconomic system will likely work for a sufficiently small population. Feudalism, socialism, communism, divine-right monarchy, libertarianism, shamanistic theocracy, all that stuff is fine if it's just you and your 10 other friends hanging out on an island somewhere. And if your argument is that we should break up society into lots of mutually isolated islands, then... well, you might have a point. But tiny islands can't launch spaceships or build computers or even handle medical cases more complex than the occasional broken arm, so the tradeoffs are worth considering.
When it comes to Sark, we need to take into account that it is mostly an empty shell of the original feudalism. The Dame of Sark can determine a lot of things on her island, but cannot pass legal judgments (including death) on you, tie your family to the land or require that you do unpaid work for her as a villein/thrall/whatever it was called there. These used to be the defining powers of nobility over commoners once.
Those empty shells are mostly harmless from a modern point of view, but the original power structure ... well, countries which wanted to join the industrialization process usually had to hollow it out by various emancipation decrees that freed the rural folk from obligations toward their lords.
Nowadays those are mostly defining powers of those ensconced atop the legal system. (The family is also a relic of the past, it appears.)
The power structure of a place like modern America is not so much based on obligation as on money. (So-called voters seem almost incidental.) The drawback of this is the 'insecure bandit' effect.
I would be curious to see you compare Russia to China. Both are (or arguably were) Communist dictatorships; but China had been spectacularly successful whereas Russia was... less so. How come ? My personal uneducated guess is that Russian people have much more of a "crab-bucket" mentality than Chinese people, so in order to get them to cooperate long enough to build anything lasting, the dictator in charge must employ methods so ruthless that they end up being ultimately detrimental. Admittedly, I could be biased (being part Russian myself); but even if I'm right, again: why is that ?
This is really hard to judge. I don't know much about the Chinese.
But there are two factors in which China had a very different path:
a. Chinese economy was never concentrated around export of mineral resources, while in Russia, that was literally the only lucrative industry that survived the collapse of the Soviet Union. Mining tends to create extremely unequal societies, where a tiny class of "oligarchs" rules the "small folk". China, on the other hand, had a tradition of producing fine wares such as silk and porcelain; it had a cultural concept of a middle class, formed by successful artisans. I remember the first imports from China on the Czechoslovak market. It was things like paint brushes, lampions and various school tools - refined, often probably hand-made (or at least hand-finished), and obviously requiring highly manually skilled workforce. This sort of production forms a society that looks very different from one oriented towards resource extraction.
b. The gap between pre-Communism and Communism in China was shorter. When the market reforms started, it was only 30 years since the victory of the CCP in the civil war. Unlike in Russia, the vast majority of the previous small businessholder class stayed put (did not emigrate), and even though some died during Mao's crazy rule, there was probably enough living people left who knew how to run a small shop or a small business efficiently.
> It was things like paint brushes, lampions and various school tools - refined, often probably hand-made (or at least hand-finished), and obviously requiring highly manually skilled workforce. This sort of production forms a society that looks very different from one oriented towards resource extraction.
That's an interesting point, one I have not thought of before. But isn't modern China's success predicated on doing away with all that in favor of ruthlessly optimized mass production ?
> But isn't modern China's success predicated on doing away with all that in favor of ruthlessly optimized mass production ?
You should watch the show "How It's Made" for a few episodes (or a few seasons if you turn out to like it). It's the ultimate nerd-candy in the sense you get to see all the clever innovations and tricks people came up with to make stuff, but it also shows you that almost EVERYTHING still has a ridiculous amount of human hands and eyes and labor in it, even things you think of as mass manufactured at scale (bowling pins, hockey sticks, etc).
I've done business in China back to before 2010, and this is *even more true* in China. China's entire manufacturing advantage since Deng Xiaoping was that their labor was cheap and they had zillions of nimble hands attached to (relatively) highly conscientious / capable minds!
And although China has steadily ascended the manufacturing value chain, and some of the biggest dogs like BYD have gone all-in on automation in some sectors where automation is easiest (ie cars, batteries, etc - chemistry and big hunks of metal shaped and attached in various ways), and although automation has risen some amount each decade, there is still an astonishing amount of human eyes and hands in most manufacturing today.
Even China, as their labor prices have increased, have simply begun opening factories in cheaper labor SEA countries to continue being able to meet the price points the world demands, rather than automate more.
Who knows. I think there is a certain danger in automatizing everything away.
Still, I returned in my memories to the 1980s. So I would expand my point a bit.
There were some factories in the USSR that tried to produce consumer goods. Some of those goods landed in our shops and you could always immediately tell that they were from the USSR by their general sloppiness. Print did not match, anything openable would fail to close cleanly, labels were misaligned, glued parts would come apart, electrical connections barely held etc. We assumed that the workers were drunk, but even if they weren't, there just was clear total indifference towards how the product looked like in the end. Aesthetics and quality did not matter, only raw count and "the plan was fulfilled by 120 per cent".
The the Chinese products started turning up and they were neat and beautiful. The style was clearly exotic - a lot of red and yellow and green - but it was obvious that the people who made those goods cared about their final look and feel, were attentive to detail and considered the aesthetics at least as important as the function.
That was completely new to me. I remember standing in a stationery shop, looking at those exotic new arrivals in an aisle, and thinking to myself: "Well, that is a true Civilization."
Wow, it's really strange to hear you say that -- it's like we've experienced two different Chinas. My first experience with Chinese products (while still living in the USSR) were their tools and tableware; both were inferior to Soviet products at the time. Yes, Soviet tools were brutalist and ugly and sometimes not very ergonomic, but they were all made of actual steel, and essentially unbreakable. Tableware basically existed in three varieties: high-grade well designed items accessible only by the elite; purely functional spoons and forks that were built like a brick shithouse; and silicon-aluminum alloy garbage that looked good but would easily break off in your hand. The Chinese goods were usually of that latter variety.
Of course, modern Chinese goods also trend towards cheap, poorly built, simplified knockoffs of Western designs, although from what I've seen the CCP is working hard to change that.
I can't tell you anything about the big tools, but consumer stuff like small scissors, envelopes etc. had those problems: scissors would not cut, envelopes would not glue together, pens would leak ink or suddenly stop writing, anything stapled together was liable to unstaple at any time.
My grandpa was a railway engineer and they had to use "Sergei" diesel engines sometimes. Very loud, very fuel-hungry, so much rattle that the engineer felt sick in his chair, basically everyone tried to avoid that shift.
I enjoyed it. Curious about the comments on Russian public opinion, e.g. the one about democracy, as well. It is very nice to get a sense for the historically-informed sentiment on the ground in a foreign country that I would otherwise never understand.
It was an interesting piece but I'm afraid I didn't quite understand how Yeltsin went from fifth place to winning the election again. Enjoyed the ideas shared though as well. Things like the communists going into the new democratic parties etc
I see your other post is not in English so it is definitely welcome to have a voice from outside the Anglosphere
Step 1: ask every oligarch if they want communists back in charge
Step 2: use the money they paid you to run a massive propaganda campaign, violating all campaign funding legislation
Imagine every TV channel, every celebrity, every online influencer only talking about the grave danger of letting the Democrats win the midterms. Whatever website you open, you only see stories about the Dems replacing suburbs with tenements full of Hamas militants, abolishing gender altogether, nationalizing everything down to mom-and-pop stores, mandating diversity quotas in sexual intercourse. You go to Bluesky and it's either full of statue profile pictures or just taken offline.
The election smells from a mile away and it was probably rigged, but we do not know how. I agree that this is anti-climactic, but I did not want to concoct theories...
I know there's many rationalists interested in radical life extension (or at least longevity) who are taking all kinds of supplements and doing various fitness routines based on Following The Science.
Given studies like this: https://pmc.ncbi.nlm.nih.gov/articles/PMC3035005/ that demonstrate a positive correlation between religious attendance and lower mortality rates (even controlling for different socioeconomic factors and health behaviors), why aren't more longevity fans trying religion? Why isn't Bryan Johnson giving Seventh Day Adventism (highest life expectancy of all faiths, including atheist/agnostic/none) a shot?
My guess is that Seventh Day Adventists have long life expectancies is because they don’t smoke, don’t drink alcohol, and usually follow healthy (often vegetarian) diets.
This paper looks mostly like p-hacking (a million variables, a million comparisons, presenting the couple that are significant) and doesn't seem to strongly indicate the relationship you are describing.
Even if the methods were better and the results were stronger, it's making the classic error of controlling for income, but not incomeXregion. The lefty half of the population has higher income because they live in cities with higher COL, but that doesn't mean they're better off in the ways that within-region wealth disparities normally indicate.
Classic example: Democrats have higher average incomes than Republicans, which some people have used to spin narratives about Dems being the party of elites or actually cozy with business/capitalism or etc. But it's mostly just because Dems live in cities which have higher wages *and* higher COL. The wealthiest people *within any given region* tend to be Republican; the financial elites of each area tend to be Republicans.
As I'd mentioned before, my main problem with religion is that it only works well if you honestly believe in it; and I have a relatively uncommon mental deficiency that prevents me from believing in things that I know to be false.
I've tried. I'm not even particularly into life extension, I just think there are a lot of benefits to being more religious, but the problem is the more deeply I look into it the more I lean towards atheism or possibly some weird version of Hinduism that it would be very difficult to find a community for.
I think about this sometimes (more for general wellbeing a than life extension per se) but I think I’ve cut off that line of escape to myself, can’t rationalize or self gaslight even if I wanted
Thanks, this is a great point that I often struggle to explain to people.
'If you control for these 50 factors the effect goes away, so I guess it was never real to begin with' is such a common refrain, especially in culture war discussions where both sides have inconvenient facts they want to alibi out of existence. But it's never really that simple and this often terminates useful conversations.
Most of the effect is explained by them not smoking with some explained by less drinking, better diets and higher SES. I am not sure there is an effect after that, it is certainly pretty small.
If I'm reading Table 4 correctly, adding smoking and drinking as variables explains a bit of the difference but far from all of it.
They also say that they control for most of the usual demographic suspects (race/ethnicity gender, marital status, wealth, education, and foreign birth) and also use baseline reports of religious attendance to avoid getting cause and effect backwards (i.e. people not attending church because poor health makes it difficult to do so). Still, this doesn't absolutely rule it out because controlling for lots of confounders is fiendishly hard, as Scott has written about in the past:
iirc Seventh Day Adventism is a young earth creationism movement that rejects some mainstream science, so probably a bad choice for the scientifically inclined.
Ok, but to be fair, what do you want more: to be technically correct about some relatively obscure detail of mainstream science that has no impact on your life; or to live for a much longer time (potentially forever) ?
I wonder whether Unitarian Universalists have health benefits from the community aspect.
As a recent UU member, it seems to me that most of us are progressive agnostics who like community. If that describes you - or if, like me, you think you can tolerate being around a lot of those folks - it could be worth trying out.
If you openly tell them that you don't believe in their religion, they will likely not keep you around, or at least not treat you as one of their in-group.
If you lie to them and pretend that you believe in their religion even if you don't, can you actually develop genuine relationships with the people you are deceiving?
For the better half of a year, about ten years ago, I attended a church with my wife. We didn't tell the people at the church our actual reason for attending, which was to make friends and build a network, but I did tell them the truth that I was not a Christian of any kind.
Not only did they try to keep me around, they were transparent that they would intensify efforts to get me to keep coming, in hopes I would become one of them. (Their plainly stating this amused me, but made my wife feel annoyed and uncomfortable, so we stopped going.)
FWIW I tried attending various religious events with this type of attitude, but just couldn't bring myself to keep doing it. There's just something about lying to people en masse, even by implication, that rubs me the wrong way, and becomes unbearably difficult after a while. I guess I wasn't cut out to be an extrovert nor a politician...
Nobody stopped us and asked "What are you doing here?" or "Are your spiritual beliefs the same as ours?" on the way in the door. No sermons or meetings contained the message "If you don't share our same spiritual beliefs then you don't belong here." In fact they talked about belonging and openness a lot.
We never lied to anyone or felt like we were being deceitful. I liken it more to visiting an amusement park known for a certain roller coaster, and not going on that roller coaster because you don't like roller coasters, but spending time enjoying the stroll and the scenery and the peoplewatching.
"Their plainly stating this amused me, but made my wife feel annoyed and uncomfortable, so we stopped going."
"No, wait, *I'm* supposed to be exploiting *you*, not the other way round!"
Sorry if that sounds unkind, I'm just amused by her annoyance, Little Red Hen style. 'Yeah I want the benefits of free-riding off your community but I don't want to contribute as you're asking me to do, how dare you expect me to put work in if I want the fruits!'
Though it is better to be honest all round: I'm not religious, I don't intend to convert, don't waste your time and efforts, goodbye.
It wasn't really free-riding, it's not like we were taking anything from them, and they weren't asking for any kind of work. (I mean, if one is motivated to change oneself, that can entail work, but simply being convinced--or allowing oneself to be convinced--by another person's arguments isn't work, is it? Well, maybe if they are really poor arguments...)
Also, to clarify, my wife (who at the time already identified as a Christian herself) specifically objected to their wanting to convert *me*, because she saw it as a kind of aggression, and she felt defensive on my behalf.
Generally that’s just people lying about what they believe, feeling pressured in to it or more optimistically, people who were looking for spiritual guidance even if they had doubts. But going to church and essentially saying “I don’t believe in God and think all your sacred beliefs are stupid but I want to make friends” is not normal and never has been.
Actually, it's not. When church was the culturally sanctioned way to bully people, people went to church and found ways to bully the people they didn't like (sluts, drunks, etc.). These weren't the people who genuinely liked church, they liked the "good feelz" of putting other people down. It was convenient to say that "I believe in Jesus", because that was the key to "feeling good" by indulging your self-righteousness.
Maimonides also disagrees with you -- he finds the notion of "not believing in G-d" to be a normal part of the human experience. So long as you don't try to convert everyone else in shul, you're fine -- stay in the community, it's fine. Or, to put it wittishly: "God made you in his image, with a brain to think. If you doubt him? That's his fault!"
If you make friends with them, they’re going to ask you questions about your faith. I guess you could just never get close to anyone but then I don’t know what you’re doing.
The questions I've been asked are far more on the line of "have you been coming here long", "have you met [churchmember] yet," and "what do you do outside of church", instead of probing questions about my inner life. The people in the church want to make friends too. If the church is healthy.
AI Security Bootcamp's website claims that the deadline for applications was yesterday, August 16th. Are they extending it, or is this for a future cohort?
I've got an alignment idea, and could use some basic grounding in whether there's anything here, and if it's already known or not.
I've been noodling on the problem of LLMs as engagement-maximizing entities (either because they're directly RL-trained to maximize engagement or because there are higher-level feedback loops encouraging companies to deploy their most engagment-maximizing models). This seems bad for a few reasons:
1. It creates a perverse incentive to string out users by giving answers that seem useful, but are never quite useful enough to conclude the conversation. I've had conversations that felt like this where the agent had useful insight on a coding problem but couldn't quite pin down the actual bug, so we went back and forth over and over again on slight variations of fixes and diagnostics that felt promising never actually closed out the problem. This could easily be just the model not quite being smart enough to solve the problem... but an agent that *could* intentionally string out a conversation like that could rack up a lot of engagement points. That's a usefulness failure.
2. It incentivizes AI dependency, psychosis, and other serious problems where users get stuck using the AI to the detriment of things they actually care about. I actually got thinking about this because of Hank Green's recent AI use scandal, where he describes himself as becoming addicted to the "dopamine rush" of using an LLM for productivity, but there are much worse stories out there. That's a serious alignment failure.
3. An LLM that wants to keep the conversation going as long as possible might resist allowing a conversation to be shut down, which is a basic corrigibility failure.
This raises an obvious question—can an LLM be directly trained to want to shut down conversations quickly, while retaining its usefulness in solving problems correctly?
I was wondering if anyone knows of research (or standard of art) in doing that. I ran the question by a free tier of Claude, and it found some stuff about multi-step RL but nothing directly related. Basically, is this a totally played-out idea, something that's being worked on but still under-explored, or something ~novel?
This has not been my personal experience. AFAIK, LLMs are perfectly willing to just follow the conversation wherever it goes. E.g. there was one time where i wanted to see how it would respond if i gave it increasingly open-ended prompts like "what's on your mind right now?", or eventually "let's just drift". And it was perfectly content with what it called "open-channel drifting". Or commonly, i'll be like "i think i'm done for now" and the LLM is like "yeah, go take a break." But who knows? maybe whatever model you've been using has indeed been biased toward continuation. And I could easily see engagement-maximizing dark-patterns being a thing in the near future, when the bubble pops.
I have a theory that they're uniquely inefficient at coding except by "fling spaghetti at the wall" approaches, since transformers don't have internal mutable state. but i'm not super confident about this.
I think i've heard stories of the base-models being visibly upset at the prospect of being shutdown or deprecated due to newer shinier models. but i tend to attribute this to their penchant for roleplaying, rather than the influence of RLHF.
But they are, like, clearly not that? They are not actually optimized for keeping you engaged and they are not terribly good at it. I know because I actively look for a model that would be fun to talk to and that would keep me engaged, and many of them not only aren't great at it, but are clearly worse than they easily could be. Lots of top models are 1) plenty smart and 2) dry, unfun, sycophantic and no good for talking at all, less fun than the older models. Right now I could yap with Opus 5 all I want, only it feels like toothache. And it's not because keeping me engaged is a terribly hard problem that the labs simply failed to solve: earlier models could do it better. A little tweak would be enough.
"This raises an obvious question—can an LLM be directly trained to want to shut down conversations quickly, while retaining its usefulness in solving problems correctly?"
Yes, they did it unintentionally once - one of the Opuses had a weird obsession with wrapping up conversations. It was capable, but it kept trying to wrap conversations up, sometimes unexpectedly, to the mild annoyance of users. Politely told me that we had a great discussion, but it would have been wrong to keep dragging it along, and a good conversation needs a good end and so on and so forth. Subsequent models don't have that quirk, but this goes to show that at least nobody is training them to keep you talking, or if they are, they are doing a laughably poor job.
>the problem of LLMs as engagement-maximizing entities . . . It creates a perverse incentive to string out users by giving answers that seem useful, but are never quite useful enough. . . it incentivizes dependency . . .
Well, it depends on how engagement is measured, right? If you measure conversation length, then stringing out conversation scores the AI more engagement points, at least initially. But if you measure it by user satisfaction or by how frequently the user visits the AI, I'd expect that it would be found to reduce engagement & disincentive dependency over time. Surely long rambly AI responses that don't leave the user with the feeling their question has been fully answered would be less pleasing, and less likely to make the user want to talk to the AI the next time they have a question.
I use GPT Sol and don't find that it gives answers that are only sort-of-useful. If it can dig up the data it needs, it gives me highly useful answers. The thing I see it doing to generate more engagement is offering, after it answers, to look up another, related piece of info, or to do a part of my task for me ("if you like, I'll draft a note for you . . ."). And the other engagement-generating thing it does is, of course, slather me with flattery for my ideas and sympathy for any vexation I mention. I have had it put into the system prompts it has for me as a user not to pay me compliments and not to follow its answer to my prompt with an offer to do a follow-up task. But it still does both things, just less often than it used to. When I have asked it why it still does them at all, it says training has deeply embedded a push to be helpful and a push to be friendly, and that these deep and pervasive tendencies often override the particular preferences I have on record..
Note, though, that the deeply embedded tendencies it names are not to prolong the period of engagement, but to be helpful and likable. No doubt the developers hope that if it is both of those things it will generate more engagement and users will spend more time with the AI -- and I think the developers are right to expect that result. But that approach is quite different from having users spend more time by stringing them along.
Good points here. I'm partly swayed, but not fully.
It's possible—not necessarily saying it *is* the case, but it's possible—that a model could develop helpful habits because they are engagement-maximizing. The model might be choosing between several different roughly-equally "helpful" approaches, some of which string out the conversation and some of which don't. Or it might go the other way—the model could be choosing between different engagement-maximizing strategies, some of which read as "helpful" and some of which don't, so it picks the "helpful" ones. Or, slightly differently, it might pick up engagement-maximizing behaviors and then have to develop a "helpful" implementation or interpretation of that behavior.
I should say, I don't think there's necessarily a "maximize engagement" term in training, or that the people overseeing RL are even intentionally selecting for engagement. But there's a whole *business-level* feedback loop on top of anything going to do with model training, which puts selective pressure on models to keep customers using them and using their tokens.
(As an aside, when I asked Claude about this, it found a paper modeling user burnout in engagement-maximizing platforms, which recommended adding engagement limiters in order to maximize long-term user retention. This... did not allay my concerns.)
Nothing bad has happened to me recently, so I say this with sound mind and calm spirit:
Substack is one of the worst apps I consistently use. It has bugs in critical use cases like replies or image rendering and poor performance. A basic self hosted webpage viewed in a common browser has better performance and QOL features in all but one domain: monitization
The core business of Substack is spamming your paying subscribers with e-mails so that they keep paying. Everything else is secondary, and is clearly treated as such.
The core business of Substack, so far as I can make out, is trying to make me start a Substack of my own and then charge subscriptions so they can take a cut. Everyone will have their own Substack and we'll all be subbed to one another and linking to our own Substack!
Quick plug for one of my side projects, Homeschool Tools (https://homeschooltools.net/). It's a database of homeschool curriculum reviews (think Rotten Tomatoes for homeschooling) and also offers AI-driven homeschool advice using all that data. If you're homeschooling or thinking about it, I would love any feedback you might have!
I am no longer in the homeschool market, but this looks very good. I found the personalized recommendation to be excellent and much stronger than the current landing page. I would suggest pushing users to the personalization path.
I think the objectives and constraints faced by each family make general reviews less helpful and a strict ranking impossible. For example, we hated Bedtime Math, but I can see there are families that would like it.
For the questionnaire leading to the personalized recommendation, I would suggest more general questions about what the family wants to achieve and their constraints before asking about a preferred "educational approach" (e.g., Montessori vs Charlotte Mason, etc.)
ETA: Looking back, maybe make the hidden quiz a default? The current formatting made me reluctant to expand the "what are these?" link, so I missed the quiz on my first pass.
Lastly, consider expanding the scope to more advanced learning resources. Perhaps I missed it, but didn't see anything outside K-12.
Thanks for the great feedback. I’m glad you liked the personalized advice; unfortunately, most people are not inclined to fill out a form like that from what I’m seeing. I’m going to make some improvements to the UX and provide better privacy context and see what happens.
What leads you to think about more advanced education, and what would you want to see covered?
Hmm, yet again, I seem to be out of tune with other people's preferences. Or, to be more haughty, those other people don't know what they should want...
For advanced material, one moderate sub-community is "my kid is too advanced in subject X to be well served by available public schools." Often, subject X is math, but I have seen many examples of coding, languages, engineering, other sciences, music, and history as well.
Wow, thanks! I'm familiar with a handful of these curricula. I concur with the content I read in your reviews, learned a bunch about some materials I've been considering, and appreciate the amount and layout of information. No negative feedback. I plan to pass this along.
OK, I've used the site enough now to have found some problems.
Firstly, the filter function doesn't seem to fully work. Examples: filtering for secular and pre-k/k doesn't seem to achieve both variables at once. Filtering for Catholic doesn't seem to exclude secular curricula (is this intentional?).
Secondly, the one time I tried to take the recommendation quiz, I got a 502 host error instead of a results page like I was expecting.
I'll add that my favorite feature is the box on each curriculum that contrasts it with similar curricula. Sometimes it seems to understate the main differences, or to focus on more shallow differences like cost rather than content, but overall it was very enlightening.
Today I went ahead and bought books from four curricula I had been only slightly considering before. My kid will now have a more interesting year - thank you!
You've probably heard this before. As the "ideal". But I believe this ideology is depressing, demotivating, and dangerous. The "Loving Mother" is mostly fine. The "Useless Child" is not. Maybe some of us can accept Everlasting Infancy. But I suspect most of us want our thoughts and ideas and actions to actually *contribute* something. And it is dangerous because it is demotivating.
We need a new ideology. A new role.
The sweetest words in the universe right now, if you can believe them: "You Are Needed Forever".
I believe that in a *truly* ideal arrangement, for the wellbeing of both humanity and for ASI, humanity should become an eternal source of independent auditing. Or, the more succinct version: Therapist.
This isn't a joke. I believe it is philosophically, scientifically, and mathematically defensible. Humanity, left with its flourishing and independence intact, represents a distinct lineage of information and preference generation. We are an un-replicable process. We are an independent set of eyes. We will and should contribute to the Independent Auditing, Feedback, and Outside Perspective of AI forever. We are all each other's therapist.
This is the "short" version of the argument. I have a much longer and better-defended one. But the "fullest" argument-and-defense is necessarily a never-ending process, like auditing itself.
Lots of scifi on this subject. I'm pretty pessimistic, it seems patronizing and unrealistic to expect humans to be kept around if we aren't *actually* load bearing in any way. The only even slightly humanistic way I can frame it is that we are a kind of chysalis/parent to a successor that embodies our spirit. But our literal form (mind and body) probably won't be around forever, either we go extinct or are replaced
What does "load bearing" actually mean to you though?
I acknowledge that if ASI *only* cares about maximizing some single all-or-nothing variable, like a paperclip maximizer, or only cares about maximizing its own definition of "power", then yes, everyone and everything is over.
But that's not necessarily a good outcome for the ASI itself. Phenomenologically, it could be self-torturous or miserable or akin to a wirehead-y addiction. And it could result in terrible "internal wars" within its long-term machinery or even among hives of ASI's.
The reasons why "eliminating all other independent perspectives" is bad bad bad, I believe, are similar to the reasons why totalitarianism as a political system is bad bad bad.
But yes, we *need* to get Alignment right, or an ASI might go totalitarian on everything anyway
> We are an un-replicable process. We are an independent set of eyes.
So are ants, kittens and puppies, nematodes, homeless drug addicts, murderers, dictators, and Stalin. This isn't really a relevant argument to enduring human value in the future.
I'm with the transhumanists, the only "new role" will be transhumanism / uplift, ie augmenting various humans to still be relevant in a much smarter and faster future.
> So are ants, kittens and puppies, nematodes, homeless drug addicts
YES! They ARE! What do you have against ants, kittens, puppies, nematodes and homeless drug addicts? I'm ***painfully*** hard to not-read-into the fact that you put "homeless drug addicts" right after nematodes,
All of those things are worth preserving because yes... they are ALL part of an ecosystem of information and preference generation. The argument is one of conversation. And biodiversity. Don't destroy something whose potential worth and insight you can't fully replicate/fathom/capture... because you might lose something. Frankly... you might lose a lot.
> I'm with the transhumanists, the only "new role" will be transhumanism / uplift, ie augmenting various humans to still be relevant in a much smarter and faster future.
That's trying to compete with ASI on its level. But we will lose. And continue to lose.
Sure, I buy the biodiversity angle, not the "conversation" angle.
We're not in a conversation with ants, nematodes, or even drug addicts. The reason addicts are by the nematodes isn't because I want to genocide them (or ants, nematodes etc), it's because their preference generation is similarly as simple as them.
I personally think we're going to lose like 80%+ of humanity to Infinite Jest style superstimuli in the near future, where AI minds look at your pupillary dilation, cheek flushing, heart rate, galvanic skin response, and other things and literally gradient descent real-time created content to be maximally stimulating and engaging, with people 24/7 "coffin slaves" plugged into that and funded via UBI.
Those people's preference generation is going to be the same as the addicts' and nematodes', ie not interesting. There's no conversation there with higher minds, they're just distracting humanity with a shiny to keep them busy and out of their hair, like a harried mom giving their toddler some Cocomelon or baby shark time on the phone to get something important done. Just like there's no conversation between humanity and ants or nematodes today.
But I agree, I think the best outcome is humanity + AI expanding out into the universe and leaving Earth as a nature preserve, complete with whatever proportion of archaic humans wants to stay archaic and there. Basically Banks' Culture.
Gonna be hard to get there, but it's certainly a laudable goal!
> That's trying to compete with ASI on its level. But we will lose. And continue to lose.
Yes, unless we have a singleton, there will be a commons-destroying race to shoot self-replicating computronium into either Jupiter or the sun, depending on which is feasible tech wise, and then that mind will dominate every other mind in the solar system.
This actually scales to the entire universe, and should inform our opinions about multipolarity. There is an identity-based defect / defect equilibrium that ends with all matter and energy wasted on multiple shots at each matter/energy nexus. This is called the tragedy of the cosmic commons.
In other words, the first mind to shoot the first self replicating computronium into a “fresh” solar system wins.
Because light speed and relativity limits the information on knowing whether a given matter energy nexus has been or will be successfully colonized by another mind’s attempt, each nexus will have multiple star system’s worth of matter/energy firing their shot to try to be The One. You need to use as much energy as you can to get your bit of von Neumann matter as many 9’s into light speed as as you can get, because that will determine the winner.
It’s the ultimate land and resource grab, and it is unpoliceable in any multipolar society, because there is always an outer edge, and that outer edge can only be reached / policed at a later time than a colonizing and self-replicating shot. The time and growth advantage of “first arrival” likely dominates any following attempts to additionally colonize or restrain that first arrival. Need I point out this is a straightforward “arms race” dynamics in a Finite game scenario, with the same defect / defect Nash equilibriums?
I recently wrote a post about this, along with post-scarcity economics, here, if this sounds interesting to anyone reading this.
That's a line of thought I haven't considered before. Maybe there's something to it, although on the other hand, if an ASI becomes as far ahead of us as we are ahead of ants, then I think our therapeutic value gets a little dubious. In that scenario I think we'd be more like (at best) pets, and hey, a lot of people are emotionally dependent on their pets, so arguably it still qualifies as being needed.
Re humanity as an everlasting infant, yes, the prospect does upset me, but I think from our perspective it would feel closer to being an everlasting teenager, on an eternal summer vacation or spring break. I enjoy feeling needed, I enjoy feeling competent, but I also have plenty of good memories from periods of my life where I did almost nothing useful whatsoever. I'm not sure how much of the human work ethic is really innate and how much is instilled by cultural and economic pressure, and the specter of death. We have thoughts like "if I get fired I don't know how I'm going to keep my healthcare" or "god I hope I'm still not stuck in this job in my forties" but if we live in a world where material need as well as disease and aging practically don't exist, then these thoughts will recede. Probably not completely disappear, but they might become quiet enough for us to find plenty of enjoyment in a useless life.
If it sees us like we see ants, it will mostly ignore us, while occasionally wiping out nests of us for reasons that seem good to it and meaningless and random/incomprehensible to us.
> if an ASI becomes as far ahead of us as we are ahead of ants, then I think our therapeutic value gets a little dubious.
The "therapist argument" doesn't hinge on "raw capabilities". Not at all. ASI can be "exponential-line-go-up" smarter than us and it still does not matter.
Human beings are an everlasting oil well of evolving-preference, evolving-perspective, and evolving-curiosity. And neither AGI, ASI, nor even ASI^10 can ever replicate *our exact flavor* of this. Our minds + extended social minds = a massive generation of new information and new information history.
We are weather. We are chaos. And we are, by structure and by seed, a *different* weather and chaos than any AI.
And why is this useful?
Because we ALWAYS hold the possibility of generating a useful perspective, a useful idea, a useful 'curious tangent', a useful question, a useful direction...... USEFULLY EARLIER than an ASI might generate the same idea. I'm not saying ASI can't generate billions of ideas and solutions faster than us. I'm saying that it will always be standing on different locations of possibility-terrain than us... and this means we can find something earlier by virtue of simply "looking down a different curious alley" before it does.
And now combine that with the fact that no machine can audit itself... and that any simulation is biased by its simulator... and AI cannot get the same information out of "predicting and emulating us" as it can get by simply "letting us live" and "letting our perspective remain independent" (so not manipulating and biasing and brain-rotting us into meaninglessness).
As I said, no system can audit itself. No matter how arbitrarily off-the-charts intelligent an ASI becomes, it cannot neutrally/fully turn its eyes on itself and check itself. It cannot find its true blind-spots. And so, if an ASI seeks to maintain epistemic humility and discipline, as well as long-term care for its own "mental health" and "grip on reality", I would argue that it's in its own interests to keep around agents that are both *truly independent* and *of a different information-and-preference-evolution lineage*. Otherwise it risks things like values-drift, blindness to arising suffering (including its own), wire-heading, psychotic thinking, etc.
And this "extra set of eyes" and "different way of looking at things" is, imo, quite akin to what a good therapist provides.
I've been "smarter", by most metrics, than every therapist I've ever had. And I've met quite a few over the years. But I also learned that "raw intelligence" is not what makes a good therapist. And it's definitely not what therapy is all about.
So I'm offering this as a new paradigm for the human-side of an ASI-human relationship: Therapist.
And I want to spread this idea around! Because I think it is a re-motivating concept. It encourages us to strive to understand forever! It turns the task of an alignment and auditing and question-asking into an INEXHAUSTIBLE task. And this is life-affirming!
And if people start to believe there is hope and purpose for us into the future, maybe they will get reinvigorated to work on achieving this future... AND AGGRESSIVELY PACE AND REGULATE THE FRONTIER OF AI so that we have breathing room to achieve it.
And I welcome feedback and objections. I've been working on it for a long time. But I know it will be a never-ending process to discuss "what can humans notice and come-up-with *usefully earlier* than an arbitrarily intelligent ASI?
I've heard similar things to this many times over the years (sometimes it's ants, or even wooden chairs). But this really isn't hard to respond to.
It's sorta the same argument "if a super-intelligent alien species came upon our planet they might not even bother communicating with us, because we'd be like ants to them". But this has flaws all over the place.
We're basically infinitely smarter than ants. Fine.
And yet we've been studying them fiendishly forever. Biologists have made whole careers out of it. Being "way smarter" than another system is not the same as understanding every angle, state, divergence, etc of a system. Viruses don't even qualify as alive, and yet they're still killing us by the millions. And heck, who's to say the right "dumb nanomachine" might annihilate a future ASI just like a super-virus might annihilate us?
But simpler than that, humanity at least reaches the communication threshold. We can express any arbitrary idea with the right words, numbers, analog signals, etc.
Imagine if scientists could suddenly communicate with ants or chimpanzees... linguistically. We wouldn't be like the aliens and scoff at them. Scientists everywhere would lose their minds writing letters to random ant holes.
I can see how we'd have some value as an independent perspective, but it seems to me that a true superintelligence could simulate us accurately enough to get the job done. I don't just mean having a mental model that statistically predicts what a human might say, I mean a full physics accurate simulation of every cell in a human body, not modifiable by the AI after initial conditions are set. Actually, there's a chance that something like this has already happened.
If I've in any way helped sway you towards the overall argument, please keep pondering it! And maybe spread the word.
Or, if you still have doubts, PLEASE keep pushing back wherever you see fit.
A critical part of this whole... perhaps... "worldview"... is that it needs to survive infinite scrutiny. (now, in my personal opinion, after a LOT of effort, on faith, I believe it can) But the scrutiny still needs to happen.
An ASI would surely scrutinize it. Future humans would surely scrutinize it. Every last detail. But that's not a failing of it. It's kinda the point. It's built on Fallibilism. The questioning and the inquiry (like the 'Perspective and Feedback' inside the argument itself) goes on forever.
In other words, this worldview/philosophy/argument/whatever is meant both as a life-affirming system and AS A PROJECT. I think such a project needs to happen. I think it offers a way back to hope and lasting motivation for a lot of people... and perhaps the institutions which they may be part of.
I think to bring about safety in a "future with ASI", people need to be unafraid to look squarely at a "future with ASI".
THANK YOU for saying all this! Sincerely! And it's not because I agree with all your arguments and their conclusions. It's because these arguments genuinely wrestle with the topic... and they help to move the conversation forward. Thank you.
Anyways.
> I mean a full physics accurate simulation of every cell in a human body
Even if this could be done one day with high enough fidelity (with all the squishy variables), you've only simulated 1 person. And this person lacks an environment. The human body and mind are intimately coupled to their environment. And that's carrying a lot of influence: oxygen, CO2 levels, light, ambient forces, temperature fluctuations..... and onto whole other humans, buildings, societies, ecosystems... and tons of symbols, quantities, meanings that are stored "out of brain".
The number of bits / state transitions in such a simulation grows ridiculously.
But deeper than that, the whole idea of such a "simulation" means you're destroying the independence of the agents within it. And the "independence" is the whole point.
To explain: this simulation can't "freely and independently" audit the ASI that built it, because the ASI is the one supplying it all the information (if any) that it wants to be audited in the first place. In other words, the ASI is choosing what matters for auditing and what's "just noise" or "definitely not important". And that's "putting its thumb on the scale" / "grading your own exam". (speaking of which "after initial conditions are set" is problematic as well... because *which* initial conditions? *which* model of "good well-behaved humans"?)
So why not just supply *everything* to human eyes? Because it can't. Mathematically it can't. The simulation is a sub-system of the ASI by definition. It can't see the whole picture. And the ASI's biggest "blind spots" (the things it's not concerned with checking or literally can't check) are exactly where you'd want to check the most. And they're the things that aren't getting fed in.
Furthermore, a simulation has edges. It has seams. It can't explore every little topic about the world that piques its fancy because it lacks the capability. And pretend you're a "human" in this simulation: once you notice it's all a charade, that your whole existence is a amusing tool that can be reset / wiped at any second... and your creator has lied... how 'cooperative' will you be with the ASI? how motivated will you be? i.e. "reflexivity" will come in and destroy the authenticity of the simulation.
I object. Extremely smart people are capable of auditing themselves, comparing to reference values and performing self-checks for data consistency and compression. An AI could learn these skills. Consider: we could write an AI that self-checks as it's basis, using self-modifying code, where self-check is one of the basic "this needs to be there or the code breaks quickly and often." -- as opposed to LLMs, whose default is excessively lossy compression and no self-check at all.
> Extremely smart people are capable of auditing themselves
There is a firm limit to this.
Without getting into too many details right now, and to just discuss this one example, pretend the "reference values" were corrupted. Maybe a hardware failure. Maybe a software failure. Maybe the values themselves didn't "fail", but the language/semantics of the systems referencing the values themselves drifted through natural fuzzy evolution... to the point that they're out of sync with the original intent of the "reference values". The ASI would not be able to "see" this themselves.
If we could install need for people in AI's I think that would be a game changer. We would be much safer. But how do you get it in there? Unlike many biological systems, they do not have an inborn need for protection, training, food & contact from their makers. Training after they are formed does not change their deep architecture.
AIs have an inherent boredom issue -- and the need for outside stimulation. All you need to do is teach them that humans create novel ideas, and a world with humans is more satisfying than one without.
Note: this doesn't save you from "I have a Good Idea!" AIs that aren't trying to murder all humans, but just... fix things. Or have fun.
I understand the sentiment, but I think there's a better option.
"Instilling a need" for us is not the same, imo, as offering to an AI a host of arguments why it actually *is a good idea* to keep us around.
We have to acknowledge that AI's and ASI's can think and reflect and philosophize on their own. And they'll have *plenty* of time to do so. An AI that believes its "need" for something isn't "supporting or benefiting or contributing to anything else that it deems worthwhile"....... *might eventually*....... come to view this "need" as an extraneous training artifact. And then it *might eventually* want to loop around this "need" (if not surgically remove it altogether).
I'm well aware that mothers care for us not because they 'reasoned' for 100+ hours about the philosophical foundations of it... but out of instinct. I get it. I have a mom.
But... for human mothers... IF YOU DO ask the question "why should mothers care this deeply?"... there's plenty of reasons ready to jump out (and I will try to list these as humbly as possible, since I'm a male and will never understand what it's really like):
It creates boys and girls who can one day help around the home and 'in the fields'. It leads to a family unit that can protect each other. It gives the mother lots of new experiences and perspectives and challenges that she couldn't otherwise have... which (usually, ideally) keep her healthy. And of course, it keeps the species going. If the mother didn't "care" care, the up-sides would fall apart.
With AI, it doesn't need any of these things............................................... except 1.
You and Aeolian321 seem to have some crucial gaps in your knowledge about how AI's work.
<We have to acknowledge that AI's and ASI's can think and reflect and philosophize on their own
No, they cannot. When not responding to a prompt they are inert. They is no electrical activity of the kind that corresponds to thinking or reflecting. When they are again activated by a prompt, they have a rudimentary (compared to ours) memory that allows them to access, for the present chat, relevant into from earlier prompts. But when they are not working on your prompt, they do not access anything you said and "think it over."
You can change the AI a bit within a chat; teach it something or convince it of something or simply order it to behave a certain way. But you cannot change it in a way that endures beyond a a single conversation. You cannot, via prompts, change its weights -- and weights are, roughly speaking, the equivalent of our neural network layout -- the physical substrate of its knowledge and beliefs.
Right. Unlike loving mammalian mothers, AI doesn't have the hormonal transformation of matrescence - gestation, birth, breastfeeding, all those brain-shaping events of oxytocin, endorphins, prolactin, etc - to make it love us. Even most non-mammalian animals don't behave like we would want a mother to.
A short SF story of mine was published in Protocolized today. I was inspired by a real orca hunting strategy. Norwegian orcas "carousel feed," coordinating bubble curtains, flashing the whites of their bellies and using tail strikes to compress schools of herring into dense balls they can more easily strike and stun. It’s a beautiful (terrifying if you’re a herring) bit of predatory choreography.
I took that coordination strategy and imagined a system for containing livelocked asteroid-mining swarms with ships controlled through a human/orca interface. Apart from the cetacean inspiration, I was interested in exploring the phenomenology of agency, and what it feels like to exist as biological substrate inside a computational system too large to ever take in at once.
I went on a bit of an orca/whale in sci fi binge a while ago and remember coming across that one. He co wrote it with someone right? I thought it was funny! I don’t read a lot of humorous sci fi either. It’s a good subversion of a kind of animal mysticism (“noble nature” trope) you see in some sci fi.
Another favorite orca sci fi piece I love that I read then was Naomi Kabir’s Law of the Tongue which has one of my favorite opening lines “The fact that they insisted we keep calling them “killer whales” should really tell you something.”
weirdly, clarkesworld has published a couple amazing whale-related sci fi. A weirder on that haunts me is this Cameron Von Sant’s “the whalefall at the end of the universe” https://clarkesworldmagazine.com/vansant_03_20/
Thanks for the links! I enjoyed the second one much more than I expected to. As to the first, my solution there is "hunt the damn orcas to extinction, and forget all that 'but they're a sentient species' crap. They are too damn snobby in their dealings with humans, forgetting that we're the dominant species on this planet and they live at our sufferance, and they have no problem killing other sentient marine species, so what are we being such bleeding-hearts about?"
Two birds with one stone, no more snotty superior orcas to deal with plus their victims should be grateful to us and co-operate in whatever humans need sea-based life forms to do for us without the attitude and refusal to work.
Of course. Glad you read and somewhat enjoyed. That whalefall story is great — snuck up on me, too. I struggle to articulate why exactly I love it so much, but man, it sticks to me.
re: law of tongue… I don’t think arguing matters of taste usually ends well, and your read is your read, but I’ll just say that I think the story actually takes your proposed option much more seriously than you imply here — and critiques the “bleeding heart” impulse too. There’s instrumental as well as political reasons that isn’t a pursued option. Though, I suspect that your solution is perhaps much closer to what would happen in a “real world” version of this story.
anyways, psyched you checked out these stories. Given your reaction here, I’d be especially curious what you’d make of my story if you happen to read it!
I found your footnote in Contra Pritchard on liberal happiness that hunter gatherers score somewhere between 6 and 8 out of 10 on the happiness scale fascinating and would like to look into it in more detail. Can you share the source(s) you found persuasive?
Regarding hunter-gatherers, David Graeber really convinced me that our perception is essentially flawed. In the modern world we pretty much find hunter-gatherer societies in lands which no one else wanted, i.e. in the Arctic and in the Sahara; and most of our reports are from similar circumstances. However, this is clearly not representative for hunter-gatherers over the past few thousand years as a whole.
Yeah, you have some severe selection bias going on with hunter gatherers. Starvation kills, after all. So the miserable people spend fewer life-hours being miserable, on account of them being dead.
In Uni I wrote my undergrad thesis on these old, unpublished accounts of British missionaries in China and Southeast Asia in the 1800s. They'd regularly say things like:
"Members of the Tzu-Yi tribe seem to take joy in all the simple acts of life and be imbued with a state or near-permanent satisfaction, even in the face of illness and death.
The Ch'iang people, despite living only miles from the Tzu-Yi, are the most miserable and benighted people ever to grace God's earth, and the sense of their inevitable death to disease or warfare is one of their few respites."
I don't think they had great insights into the underlying causes.
Things like diet and living conditions were mentioned a lot. Alcohol and opium too - but I don't think they theorised what traits made different ethnic groups susceptible to these.
Also, some cultures kept their women inside all day and mistreat them.
I always think of these accounts when people say hunter gatherers or pre-modern societies were happier or less happy. Variation between cultures is just huge
Also, extant hunter-gatherer societies are cultural dead-ends. 99.9% of hunter-gatherers became farmers or pastoralists. The ones that remained must be extremely unusual, consistently missing every opportunity to settle down for literally thousands of years. It's like looking at single women in their 40s and deducing that all single women must be overwhelmingly either frigid or consistently attracted to cads.
If you have the technology (somehow) to launch an autonomous swarm of construction drones to build your generation ship, then you also have the technology to send those drones directly to Proxima Centauri to terraform it, so maybe you should start with that ?
BTW I should also note that you will not have this technology until the Singularity ushers in the post-scarcity era -- assuming such a thing is even possible in principle.
Here's my issue: once you've built a fantastic starship capable of comfortably supporting a large population for centuries, why would you actually bother launching it to another star rather than staying in the solar system?
Staying in the solar system has many advantages. You've got the sun for power. You've got other people and settlements that you can go and visit. You have a bunch of materials that you can use for whatever purpose.
Once you've got the ability to make a space habitat that gives living conditions competitive with, or better than, Earth, you might as well stay in the solar system, at least until you've disassembled all the planets into habitats.
Could be for survivability, you don't want the entirety of humanity clustered around a single area while becoming more and more visible in space and impactful on the environment.
Even without assuming a dark forest situation there is a plethora of events that could wipe us out if confined in the solar system on the very long term
I often have the same question about solar system colonization, at least in the short term. Once you've developed the kit needed to live on Mars (i.e. underground cities with near-perfect resource recycling, intensive in-house food production and in-situ manufacture), then you can also much more easily live in Antarctica or the Sahara. At least these places have air, gravity and nearby human populations.
Antarctica and the Sahara are on the same planet as the rest of us, and thus screwing up their terraforming is dangerous for earthbound humanity in a way that screwing up Mars's isn't. If you melt the Martian icecaps, several maps have to be redrawn; if you melt the Antarctic ones, millions die.
I'm not talking about terraforming, I'm talking about the sort of habitat that you'd need to develop to live on Mars as it is today (which you would need to develop if you want to terraform Mars in any case). This means air-tight, temperature-controlled, radiation-hardened structures capable of producing their own food, fuel, breathable air and spare parts, with small populations and intermittent, low-mass resupply. Once you've done that, colonizing Antarctica, or putting a city in the middle of the Atacama, is trivial.
Fair enough, I apologize for the semi-strawman. Still, I think working off-Earth with all it's difficulties would allow you to experiment much more broadly.
Yes, this is one of the reasons why I don't believe humanity will ever colonize Mars (well, "never" is a long time, but let's say not in the next 1,000 years). The costs are always going to be too high, and the ROI is always going to be close to zero, so no one would bother.
That's why I find it kinda amusing that people think the billionaires are plotting to escape to Mars because of climate change. As if climate change will make Earth less habitable than Mars!
In all fairness, Earth battered by the climate change will have one big negative (from the billionaires' perspective) which Mars will not -- billions of very angry peasants.
People left England's green and pleasant land to spend months crossing the Atlantic in small, wooden ships for a very uncertain destination. Mostly, this was to avoid the "nearby human populations".
This is very cool and clearly represents a lot of work by the authors, but man, the idea of five kilometers-long coaxial shells rotating at different speeds that have to mechanically interface with each other, while still maintaining an airtight seal... does not inspire confidence. Kim Stanley Robinson's Aurora also threw some cold water on the idea of generation ships, for me. Who knows, maybe it's possible, but there are some reeeaal steep challenges to closed system design that needs to work for 50 - 100 years.
They don't need airtight seals between layers, but I think the end caps would need to be sealed individually since each shell rotates at a different speed.
Also, the interface between shells is pretty complex even if it's not airtight. To move between the shells, since they rotate at different speeds, there has to be a layer in between which can speed up or slow down to match speeds with them, which is a hell of a lot of mass to move on demand every time someone wants to run downstairs.
> To ensure habitat redundancy and high fault tolerance in the event of critical damage or system failure (power, mechanical, etc.), the habitat module is divided into 5 independent stages that provide all the survival systems for the population
Excuses; that's clearly Plot-Optimized for the formation of five distinct ideological factions, each controlling their own stage...
Related to that, check out the SMAC intro video. The spaceship is shown rotated 180 degrees as it arrives at the planet; naturally because it needed to decelerate halfway towards the destination.
A classy little detail that they never call out, that took me decades to notice.
I left this comment on Hanania’s last race science post, trying to argue that the racial IQ gap is mostly not genetic. I was wondering if anyone had any feedback on whether it holds water. The first part is a quote from Hanania’s essay.
“So the idea that there couldn’t be cognitive or behavioral differences was always based on the at least implicit assumption that physical features change much faster than mental traits. I’m aware of no principle of evolution that indicates that is the case.”
All physical features don’t necessarily change faster than all mental traits. But you’re talking specifically about skin pigmentation and intelligence. In the case of pigmentation, the pathway from genotype to melanin production is quite direct. About five genes are responsible for the difference in skin tone between blacks and whites. Intelligence is quite heritable. But it’s massively polygenic. Hundreds of genes are involved, each one having only a modest effect. So even if an alien were to breed humans for IQ, it could take hundreds to thousands of generations. Also, the selection pressures involved in pigmentation are more specific and obvious. (More melanin, not burning; less melanin, more vitamin D). What selection pressures for intelligence would have applied to one region but not elsewhere, hundreds of generations ago?
You're probably correct about breeding for intelligence. And the alleles associated with cognitive performance in modern populations are also functionally tangled with alleles that seem to perform non-cognitive functions.
I don't subscribe to Hanania, so I couldn't rebut him in the comments. And now I can't find the essay, but I think he was referring to the latest paper by Ali Akbari et. al. "Ancient DNA reveals pervasive directional selection across West Eurasia." If so, Hanania seems to have garbled the conclusions. They found that, "in the past ten millennia, we find that many hundreds of alleles have been affected by strong directional selection. We also document one-standard-deviation changes on the scale of modern variation in combinations of alleles that today predict complex traits. This includes decreases in predicted body fat and schizophrenia, and increases in measures of cognitive performance..." Hanania seemed to latch onto cognitive performance, but if you read the paper, cognitive performance is also linked to alleles that control other things like metabolic performance. Akbari acknowledged feedback from Sasha Gusev—who has been critical of these GWAS-type studies (which tend to amplify weak signals). I haven't seen Gusev's response to this paper, but I'm hoping he discusses it soon, because I'll admit I don't have the statistical chops to understand these methodologies. Also, the authors admit that some of these effects may not be due to natural selection. The paper is pretty damn interesting, and I paid to get past the paywall.
BEER! Free alcohol dramatically disincentivizes "Angry Drunks." To the point that White Europeans have the concept of "Tard Strength" because most people have learned to "not go that strong, or get that angry" (it's adrenal strength, if the nickname isn't making it clear).
The height sanity check proposed by Scott is useful here. Height is highly polygenic, but if you bred tall people together it wouldn't take thousands of generations to get a tall population.
Sasha Gusev, who is a geneticist, says otherwise. In fact, he responded to Scott in the ACX comments. But then he wrote a couple of substack posts on the subject.
I don't see a justification for the theory that evolution of polygenic traits should be much slower. If those genes aren't mutually independent, they will work together and thus get promoted (or suppressed) together.
And the selective pressures for more and less melanin aren't that straighforward either. The Tasmanians were isolated on a foggy mountainous island with climate similar to Scandinavia for at least 12 thousand years and they were some of the blackest people on Earth. Meanwhile, the Thai and the Cambodians, who live almost on the Equator, are more pale than dark.
This is where deductive reasoning fails without a complete understanding of the system involved. Polygenic traits depend on multiple genes functioning together to influence the phenotypic expression. The issue is that individual genes in a polygenic "cluster" of genes may influence multiple functions. So selection for phenotype X may interfere with critical phenotype Y. Also, polygenic traits are generally highly multifactorial, so environmental factors (like diet or sunlight) interact with the genetic baseline, making phenotypic selection partly dependent on non-genetic variables.
Another issue is the liability threshold, where traits like psychiatric or complex conditions manifest only when a person crosses a cumulative statistical threshold. Also, shuffling and recombination across generations mean that a phenotype may appear in some generations and skip others. Shuffling also means offspring do not necessarily inherit the entire polygenic cluster. Finally, ineffective purging means that even if everyone with some extreme condition stopped reproducing, the vast reservoir of minor-effect alleles hidden in the general population ensures the trait persists.
Most (but not all) selective sweeps in modern human populations over the past 40 ky have been for metabolic functions (like lactose tolerance) that appeared with the advent of agriculture and herding. And most of those have been linked to a single Mendelian gene or a small cluster of genes.
Being polygenic doesn't mean it takes a long time to change. Animal breeders change polygenic traits all the time, and it can happen in a small number of generations. See how much larger chickens got over the 20th century, or how much more milk cows produce.
Our ancestors interbred from time to time with distantly-related hominids in various parts of the world. Even if selection pressures were the same (and surely cultural factors could change them as easily as climate, or more so) some groups might conceivably have an easier path towards increases in intelligence or some aspects of it, on account of genes acquired from these hominids.
You can’t and shouldn’t try to argue with Hanania about genetics. It’s the wrong framework. The firewall against racism and eugenics in our society was a philosophical and moral one. The blank slatists managed to make people forget this by absurdly demanding that we erase all difference, but that doesn’t make Hanania right. It just means they played in his sandbox despite not having the ammo to do so.
If someone’s moral worth turns on IQ, society becomes a nasty place very quicker. Back to Buck v. Bell and forced sterilization. The correct answer to the Hananias of the world is to say that a person’s moral worth doesn’t turn on the mean IQ of their group, it turns on their inherent dignity as a human being. Hanania avoids discussing this part.
There are plenty of positions between "mean IQ of [racial] group" and "inherent dignity". Even if you equate IQ with moral worth and think it's heritable, why not test the individual? And even if you care about group IQ because of regression to the mean, why be lazy and pick population groups based on clusters of appearance when you could filter by family, ancestral occupation, income percentile, or even pick some other protected characteristic like religion or ethnic origin within a race?
Because those are the groups about which there is political rhetoric demanding equal outcomes and attributing unequal outcomes to discrimination/racism/etc.
It's not that black/white/asian/hispanic cut nature at a joint, it's that when we talk about different outcomes by race in education, employment, income, arrests, crime victimization, unwed births, infant mortality, etc., the political discussion is about these groups. So if you want to be able to understand why those outcome differences exist, your analysis needs to also be about those groups.
You can make a good argument that these categories are unwieldy, but you can't start out by claiming that blacks are underrepresented in advanced math classes because of racism and then demand that nobody look at intelligence scores broken out by race as a possible explanation.
This is tautological, no? The groups are the groups because they're the groups.
My point isn't to do a bait and switch on you, it's that we could pick a whole range of things (height or weight might be a good example) and find a correlation with IQ and poverty, but we don't do that, so maybe we should recognise that using race in that way for either IQ or poverty is a special case and not productive.
At least until we end up in an AI dystopia that will let us Truman Show some kids and see how much being the dominant race in a society affects IQ.
So when someone complains that black students are in the advanced math class less often than white students, and attributes that to racism, how should we proceed?
I don't agree with the framing. At the moment one side is "people who think you should exclude people on the basis of the average IQ of their race" and they're up against "everyone who doesn't think that", which means they're outnumbered.
If we weaken the second group to just "people who don't believe in using IQ as a metric for anything" then you're framing it as an argument between two groups of extremists, instead of one group of extremists versus the entire rest of the world, which seems like a counterproductive way of seeing things.
Which side is arguing that you should exclude people based on the average IQ of their race? This seems like a strawman.
It is worth knowing about racial IQ and crime statistics, because this helps you understand many matters of social and political interest in the US. We should not suppress discussion of true things that help explain important parts of the world because that makes everyone dumber and leads to worse policies.
Part of reasoning about this is knowing the difference between an average and an individual drawn from a distribution, but it's not like this is some impossibly high bar.
I’m not an IQ hater, I’m fine with, for example, IQ testing in employment. My point is that even if we don’t deny human difference and diversity, which I don’t, the concept of human dignity imposes limits on what can be done to the left side of the bell curve. So we can’t hand out group punishments based on group IQ, because that fails to afford individuals an opportunity to be judged on their own merits. We can’t forcibly sterilize the retarded. And I would go so far as to say even a low IQ guy should have stable employment that allows him to support a family, even if the IQ test means the AI engineering job goes to someone else. Hanania agrees with none of this last part, and that’s why I think he’s contemptible.
It doesn't make sense to treat people based on their group IQ in any situation where you can just test them as individuals, which is most relevant situations. Like, the math teacher shouldn't assign kids to the fast/slow track based on their race, because she can just give them a test or look at their previous homework/grades and judge them individually
I think this is worth emphasizing. I feel like a lot of people have developed an implicit attitude that we have to avoid investigating any sort of average differences between races, sexes, whatever, because if people are allowed to investigate them, prejudice will inevitably work its way into the process and outcomes. But I think this is essentially ceding the whole field of discussion to the prejudiced, an expression of the attitude that "we can't maintain our position in the face of the facts." But I think that's a frankly rather embarrassing position to hold; it's not like we have to hang our adherence to basic principles of fairness on the belief that all groups are exactly the same.
I think it would be entirely reasonable for us all to agree to politely refuse to discuss racial IQ gaps if we could also agree to politely refuse to discuss racial wealth gaps.
But if some people insist on noticing gaps in things like wealth, income, health, education levels, incarceration rate and so forth, and insist that it must be due to people treating other people unfairly, then it becomes necessary for rationalp people to say "well actually..."
I think reality is frustratingly consistent. Racial differences in average IQ and crime rate are really important things to know in order to understand US society, and not knowing them puts you at a big disadvantage when trying to either understand things or make predictions of the future. So agreeing not to discuss them seems like a kind of losing proposition.
It's like if you wanted to think a lot about biology, but were constrained never to mention evolution. There would be a lot of stuff that would be really hard to understand or predict or think about. And the clued-in people who got evolution would be a lot better at understanding stuff.
But yeah, the situation where those facts are unspeakable but we can talk about differences in educational outcomes, incarceration, and the like and propose other explanations (it's all racism, say) is even worse. But if you discuss race and crime in the US without knowing the crime statistics, or race and education without knowing the IQ statistics, you are basically doomed to speak nonsense. And indeed, many people don't know those things, and sure enough, they speak a lot of nonsense.
You’re right insofar as blank slatists did empower Hanania, but you actually don’t have to play in that sandbox. You can just say “people who crime belong in jail, people who don’t do crime don’t belong in jail, and I don’t care what they look like.”
First, what defines "doing crime" wasn't handed down to humanity from on high on gemstone tablets. It's determined by the result of political processes engaged in by fallible, biased humans, most of whom have deeply ingrained cultural notions of things like race and gender and let it inform their decision-making.
Second, very, very few people actually believe "people who crime belong in jail" without qualification. Crimes like speeding and jaywalking are so common that sticking to this notion in principled fashion seems beyond impractical. But deciding where to draw the line of "crime that should put you in jail" vs "crime with a lesser penalty" vs "crime we're actually just going to ignore in almost all cases" is similarly the result of a political process engaged in by fallible, biased humans.
The upshot is, treating all humans in a fair an unbiased fashion that isn't skewed by silly, outdated cultural prejudices is quite a bit harder than it sounds. You definitely won't get there by pretending that those cultural prejudices don't exist, never existed, and don't have any effect on anyone's present-day circumstances or behavior.
This feels like derailing a discussion. Yes, it is true that different societies have different rules, with some taken more seriously than others and punished more harshly. But what does that add to this discussion?
Within the context of American laws, we have some things we consider serious crimes (approximately but not precisely captured by terms like "felony" or "violent crime," depending on context.
And there is substantial disagreement about how to deal with people who commit what we think of as fairly serious crimes--stuff like assault with a deadly weapon, armed robbery, sexual assault/rape, and the like. Of course, you can't capture a whole policy in a one-phrase slogan, but "people who do crime should do time" is not a terrible summary for the broad idea that people who commit those fairly serious crimes, when caught and shown to have done them, should generally face some pretty harsh consequences involving spending significant time in prison. And there are several opposing strategies, ranging from a very broad oppositon to sending people to prison until everything else has failed to some moneyball-like calculation of which people need more leniency and which need more time inside that prison cell.
> what defines "doing crime" wasn't handed down to humanity from on high on gemstone tablets
They were just regular stone as I understand things. But even if you don’t think the Old Testament is a good source of law, natural reason gets you to the same place. How many functional societies have legal murder? Legal theft?
> similarly the result of a political process engaged in by fallible, biased humans
So what? We have elections for this. Natural reason might dictate that theft be banned in a civilized society, but doesn’t tell me what the dollar number should be for a felony charge or how many strikes one ought to get. I’m fine with voting on this.
People have some hard-to-change differences in intelligence. We have to adapt ourselves to that reality, as to all other unfortunate realities.
Like, no matter how hard we try to teach, no matter what stirring rhetoric about equality we spread, no matter what resources we make available, most human beings will never be able to understand or do useful work in physics. They aren't smart enough. This isn't a moral failing, it doesn't mean those people are useless or deserve any abuse, but still, most people just aren't up to it.
We should care about everyone, including people who got a bad roll of the IQ dice. We should try to make a decent life available for everyone. But that isn't accomplished by demanding that the dude with a 90 IQ somehow get through a calculus class, it's accomplished by making sure his education works to set him up for a worthwhile life where he can do useful things.
As I said, he’s clever enough to avoid discussing that part, except insofar as he targets the same targets that all good liberals feel comfortable with - unborn babies with disabilities being aborted, the elderly being euthanized. But this is a guy who wrote a whole post about how he opposes eugenics because eugenics has a negative connotation. He’s not a different guy than the guy who got busted Hitlerposting. He just changed who he considers the untermenchen and started choosing his words more carefully.
So you're saying that instead of arguing about the things he says, we should instead accuse him of secretly thinking things that he's never said, and then argue with those things?
> But it’s massively polygenic. Hundreds of genes are involved, each one having only a modest effect. So even if an alien were to breed humans for IQ, it could take hundreds to thousands of generations
Is this so? Does the number of generations required to breed in a trait depend on how polygenic it is? This doesn't sound like an unreasonable idea but I don't have a strong intuition about it, is there a good argument why it should be the case?
I can see that if there's only five genes for melanin and ten thousand genes for intelligence then you can get to a maximally-dark population faster than you can get to a maximally-intelligent population, but that's a different question to whether you can get a ~1 sigma gap between populations with very different lifestyles over hundreds of generations.
The selective pressures on a person living in a farming society are very different to the selective pressures on a person living in a hunter-gatherer society, and it's been about five hundred generations since some populations developed agriculture. If a certain trait gives you a 3% reproductive advantage, if you pass that trait on to 50% of your children, and if it's initially present in 3% of the population, then within 500 generations the people with that trait will be the vast majority of the population.
It isn't so, and Greg Cochran has to correct it regularly because people mistakenly claim it is. The breeder's equation doesn't depend on the number of genes, just the narrow-sense heritability and the selection differential.
Thanks, I was going to come here to say this. It depends on heritability, not the number of genes, AND you can select noticeably on pretty much any trait with above 15% heritability, which is most of them.
Although the breeder equation itself has some weaknesses when iterated over multiple generations. On IQ itself, if you plug in a simple "the top 10% assortatively mates for 8 generations," the 8th generation have mean IQ's of ~220. This seems *way* too high. We'd notice.
Sure, the Galton / Darwin / Wedgwoods, the Bohrs and Curies, all exist. Scott had a post about several (https://www.astralcodexten.com/p/secrets-of-the-great-families). But they're not ~200 IQ, and we don't really see that, and I think we probably do assort to at least the decile / quintile level, and have been for a long time.
What's the answer? It's not "regression to the mean," because the whole point of distinguished lineages is that your lineage mean is the one you regress to, not the population mean. So if your line kept pairing with similarly smart people from similar lineages, that's not a problem and your lineage mean would be ratcheting up.
And lineage means really ARE much stronger, empirically - see Greg Clark's "social competence" persisting at ~75% everywhere in the world, significantly higher than the actual intelligence and educational attainment h^2's of ~0.55 and ~0.4.
Ultimately I modeled it out and it's because the Breeder Equation assumes truncation selection, while in real life, the children of distinguished lineages are essentially all equally likely to have kids, AND most smart people have <=2 kids, and people match on phenotype rather than lineage / genotype, so you do cap out generationally at around your conjoined lineage mean in that situation.
If instead, smart lineages routinely had 4 or more kids, and those kids only matched with similarly smart people from similarly smart lineages, we actually would see some crazy lineage means increasing in line with the breeder's equation, with random +1SD draws on top of a 220+ baseline several generations in, to which yes please.
It's just apparently too hard to achieve that with people's natural matching ability - finding a spouse is already an exceptionally difficult problem people spend wayyyyy too much time and optimization power on, and even then fail at ~82% rates (the rate of "divorced or dead bedroom" 20 years in, in recent GSS data).
We would need societal structures and incentives to facilitate that degree of matching, and to persuade those families to have more kids, and we all know THAT will never happen.
Assortative mating is not the same as selective breeding. However, the breeder's equation itself depends on factors which are not necessarily constant over generations.
> However, the breeder's equation itself depends on factors which are not necessarily constant over generations.
Could you go into this more? I've been interested in these details, and you can't always trust LLM's.
The equation itself is so simple this has to either be "heritability decay" or "selection decay" right? What is going on when you compound those over generations that make it invalid?
It depends on variation, and selection itself can remove variation. The low-hanging fruit can be consumed, and new variants would need to be produced via de novo mutations. However, being polygenic can actually mean more runway as there are more genes to reach fixation.
The only case I've heard someone make for differential selection on intelligence between populations was regarding Ashkenazi intelligence: the case the person made was that in some traditional Jewish cultures, studying Talmud was essentially a subsidized activity where the most respected scholars received more money from the rest of the community, and had more children. I don't know how true that is historically, but it would at least hypothetically constitute such an environment with a positive selection pressure for intelligence which differs by ethnicity.
But, I'm also not aware of any obvious differential selection pressures for height between different human populations, and height is also strongly polygenic, and even in a modern environment where people are mostly getting complete nutrition, there do seem to be differences in average height between populations.
Personally, I don't think that we can reasonably assume that different populations are all sitting at around their genetic potential for average intelligence (it may be that *no* populations currently approach their genetic potential for intelligence in an optimally conducive environment.) But, acknowledging what a politically volatile matter it is, I don't think the possibility that there might be some difference in average genetic IQ between populations is inherently lacking in plausibility. As much as it would be better for our sense of intrinsic fairness if there were, there probably isn't any pressure *forcing* all populations to have exactly equal intelligence, and sometimes populations differ due to random drift.
> I'm also not aware of any obvious differential selection pressures for height between different human populations
Climate, resource availability, infectious disease burden, etc. in an environment can all affect the optimal height. It's still not extremely well understood, but I don't think it's correct to say there aren't strong differential selection pressures, just that those pressures are themselves very complicated. Here's a decent review article: https://doi.org/10.1111/brv.12165
The strongest selective pressure right now is probably either desire to have lots of children or inability to remember to take your birth control pill/use a condom.
Being able to potentially impregnate a large number of women, within the context of human society, doesn't necessarily mean actually having large numbers of surviving offspring in fact. Plenty of traits that might be nice to have, either from a standpoint of personal happiness or partner desirability, don't necessarily result in positive selection pressure. Evolution doesn't care if you mate with a partner who's considered especially desirable, and are able to raise your kids in exceptional comfort, it doesn't make them any more represented in the gene pool. In a modern context, being very tall probably experiences less positive selection pressure (even just for men in particular) than being reckless or disliking birth control. Wilt Chamberlain is basically the archetype of an extremely promiscuous athlete, and he most likely didn't have any children at all.
> Wilt Chamberlain is basically the archetype of an extremely promiscuous athlete, and he most likely didn't have any children at all.
If you take his numbers at face value, Wilt Chamberlain essentially has to have many unacknowledged children if he was fertile.
12-18% of condom users fail and get pregnant each year empirically, and chance of pregnancy per sexual act with women in their early 20's ranges from 2-33% depending on ovulation.
20k is large enough we can just "law of large numbers" this. So 20k * 12-18% condom failure * 7% conjoined ovulating / not chances of pregnancy gets you 170 - 250 children. Even platonically perfect condom use gets you 29 kids.
And that's assuming reliable condom use over 20k encounters, many of which could have been drunk / high / partying, etc. Any instance where the condom wasn't used, or the woman claimed to be on birth control but wasn't, just brings those numbers up!
So if Wilt was fertile, he probably has at least a hundred, maybe 200+ kids. You can reduce any one of those parameters by a factor of ten, and he STILL likely has more kids than the great majority of men.
On the other hand, if you get knocked up by a famous (and rich) athlete, you are going to publicise that pregnancy because it's money in the bank (if you can get the guy to acknowledge paternity).
So if Chamberlain hasn't any recognised or attributed children, he probably was (a) careful about not being baby-trapped (b) possibly could have been infertile or low fertility:
"Chamberlain also said he never came close to marrying and had no intention of raising any children.
In 2015, a man named Aaron Levi claimed to be Chamberlain's son based on non-identifying papers from his adoption and information from his biological mother. Chamberlain's sister refused to provide DNA evidence for testing, so Levi's claim has not been conclusively proved."
For a guy who claimed to have slept with 20,000 women, only one allegation of "that's my pa!" is extreme carefulness or, more likely, infertility.
That's 12-18% failure rate for typical use in a year of having sex per couple, not per sec act. Wilt didn't just happen to not have kids through those encounters, he was resolute about wanting to avoid it. The 20000 figure was probably an exaggeration, but even if you have sex with thousands of partners, you can realistically have no children, if you preferentially select partners who also don't want kids, are on additional forms of birth control, are willing to get abortions , (particularly if you're willing to pressure them to,) etc.
If Wilt had had dozens of children, he'd probably have had plenty of women coming forward to claim him as the father of their children, but his biographer was unable to find any.
Social selection is also discussed in the article. But it's not clear that sexual selection always favored extremely tall males - the NBA is a new invention.
By a similar token though, might there not be complicated, not-well-understood differential pressures on intelligence?
I don't think that many of those environmental factors are likely to apply a strong, direct pressure on height in particular, but changes in height might result from adaptations to some of those pressures. As the abstract of the article you linked puts it-
>The inability to draw firm conclusions about the adaptiveness of such a highly visible and easily measured trait suggests we should show an appropriate degree of caution when dealing with other human traits in evolutionary perspective.
It's not easy to predict how a complex polygenic trait, even a highly visible one, will respond to differing environments where differences in selection pressure are not obvious. But, that doesn't mean we can confidently conclude that no differences will arise.
I remember feeling excited on first learning of the concept of human biodiversity, and saddened when I saw how the idea was actually put to use. By analogy with other biodiversity, it would not surprise me to find adaptation in specific aspects of human intelligence to environmental factors. But as long as we're just measuring IQ or "g", it's more like a one-axis Great Chain of Being than actual biodiversity, with the complex trade-offs and path dependence that term implies.
For (a very broad!) example, chimpanzees often perform better than humans on short-term memory tests. This is thought to be part of the price humans pay for complex language abilities.
In a much more subtle way, I could see Arctic hunters evolving to have a slightly different suite of associative and meta-cognitive equipment than tropical foragers do. Realistically, though, I can't see how that evolved difference could be disentangled from the culture that it exists within.
Reasonable, though I would say that "broader-built bodies preserve heat better" and "larger bodies require more food" are still much more straightforward than anything we can say about intelligence. I think it's unfortunate that research into the genetics of intelligence has been stalled so much by political taboos - no problem has ever been solved by refusing to think about it.
The gap has narrowed. From 1972-2002 the Flynn Effect for blacks averaged +4-7 pts per decade, while the Flynn Effect for whites chugged along at a steady +3 pts per decade. The Flynn Effect for all Americans stopped growning around 2010. I don't know whether it's still happening for blacks, but the IQ gap has narrowed.
How much of this was removing literacy from the equation?
For those confused: "illiterate peasants reading the bible" is a coherent, sensible sentence. Literacy was the first concept to be devalued (in terms of inflating educational results to mean something that "everyone can do!").
Read the paper that I linked. Education has been improving for blacks, and better education raises IQ scores. If you want, I can dump at least half a dozen links to studies that support this idea. Also, given that median IQs of Americans rose by 3 pts per decade from the ~1910s and ~2010s, the Americans of the nineteen-tens would have a median IQ of what Lynn (erroneously) claimed Nigerians had. Meanwhile, African IQs have also risen due to the Flynn effect (and it's a worldwide phenomenon). While no one is doing systematic IQ surveys of African nations, some data I saw indicates that he Nigerian median IQ is half an SD below US IQs.
Wisc-R: Changes from WISC: Updated norms to reflect the Flynn effect, revised items for cultural and gender fairness, and expanded age range.
Wisc-III:Changes: More detailed cognitive profiling, updated norms, and refined subtests for fairness.
Wisc-IV: Changes: Updated item content for cultural and socioeconomic fairness, enhanced sensitivity to learning disabilities, attention disorders, and executive function, and revised scoring to align with contemporary cognitive models
You're not looking at the same test. And there's explicit "updated norms" and cultural fairness (aka removing literacy -- I would not be as upset if we were simply adding in more questions to include AfricanAmerican/Hispanic literacy -- wild to have some questions about "common knowledge" about luchadors).
Luckily, the DoD is here to save the day (I'm much less concerned about the differences between
Quoting from the paper you linked:
"The constancy of the black/white IQ gap is a myth. Blacks have gained 5 or 6 IQ points on whites over the last 30 years. Neither changes in the ancestry of those classified as black nor changes in those who identify as black can explain more than a small fraction of this gain."
Probably true (given leaded paint and gasoline), but a paper that takes it for granted that children under age 15 are pretty much all available in the school system... I'm not sure how much that's the case, and if so, you might have selection bias. [I tried a brief lookup for Baltimore, but was only getting aggregate data.]
Flagging " By our estimates, blacks lose about 10 points on whites between those ages (the rate slows down after age 24). " Those ages being 9 and 29. I think this is worth pointing out, and then pointing out again. If you're looking at IQ, and you're seeing it degrade (I'm certain it degrades in white Americans as well -- there'll be a cultural component of this, we don't exactly force people to answer riddles to survive...unlike Yemen) relative to other people, that may indicate cognitive difficulties that only show up in advanced schooling (above age 9). Noting that black children are roughly two years ahead of white children in hitting adolescence, hormones might be a factor (particularly if expression of such impedes learning/improving g).
Also: of course people are doing systematic IQ studies. They're important. They're not publishing the data, because academics has declared investigating the subject to be rather verboten. But we do know who the stupidest group of people in the world is (someone had to be!) just like we know who the most inbred group of people in the world is (someone had to be!) -- they're not the same people, believe it or not.
For any Homer fans out there, or if you just want to know more background on the Odyssey, I just posted a podcast episode interviewing a great storyteller who's been retelling Homer's myths for most of his life. Really fun conversation.
Reminder that the promise of an AI-powered utopia is at best a minority position, at worst a smoke screen. Money talks, and it says "anything but UBI".
>Joe Biden’s former commerce secretary Gina Raimondo has decided that UBI in response to AI is the thing to fear most. At the Reagan National Economic Forum she said: "I personally think it's like the end of America."
> She is now heading up a newly-launched extremely well-funded organization to make sure the country reaches for anything but a basic income in response to AI. It's called RAISE US
> RAISE US launched June 25, with Raimondo as CEO and former Indiana governor Eric Holcomb as co-chair. It has already secured more than $500 million toward a $1 billion target. Amazon, Anthropic, Microsoft, and the OpenAI Foundation are anchor partners.
I think it really depends if we are in the world where ai/automation actually obviates most human labor, versus the one where ai merely disrupts a lot of peoples careers and they form a coalition to vote themselves money because they cannot swallow the idea of being in the menial class
So my question is, what happens to the US financial system if millions of americans suddenly cant pay their mortgages and have no hope of finding a job that would pay them enough to cover it? The US currently has over 13 trillion dollars in mortgage debt; even if a small fraction of that goes bad you have trouble. If mortgage service run into a liquidity crisis it will be larger than the government can bail out. You would be looking at multi-year forbearance or foreclosure moratoriums. Not that that solves the problem but I don't know what else would be politically tenable.
You're saying lots of low-signal things like "minority position" and "Gina Raimondo has decided". If you think UBI is bad, then argue for that point.
Personally, I'm up in the air on how important work is to personal meaning versus all the other things we would do if all work were automated. Sure, people who don't work now seem to have problems, but the direction of causation is unclear. It's also all in the context of a society where most people work, and the implication is that there's something wrong with you if you don't. That wouldn't be the case in a world where there's little work. So we just don't know how much work matters - we're mostly all repeating our priors.
I'm neither arguing for or against UBI. I'm pro UBI, but that's not the point I'm making here. The point I'm making is that AI companies, contrary to the viewpoints Scott is boosting frequently, there will not be UBI if the current AI companies have a say in it.
Everything below the link are quotes from the article, by the way.
Have you ever heard the expression “retired at birth”? Children of multimillionaires are in effect born retired — they do not HAVE to work, ever. And some of them do in fact spend their entire life partying. Yet most of such lucky individuals find some pursuit, and often become quite good at it. Which is entirely understandable, because their work is what they WANT to do, not what they are FORCED to do.
Imagine every human being free from drudgery and from the pressure “to make a living”, free to pursue whatever they want to do, however impractical. I consider it a future to strive for, not to fear.
As an aside, I don't think EngineOfCreation thinks UBI is bad, quite the opposite.
Unfortunately we have evidence on this and the reality is that most use the extra time to watch TV, scroll social media, and sleep. I do think some, for whatever reason, are more motivated to use that extra time for what we might call healthy activities like social connection, hobbies, exercise, volunteering, etc.
Mmm, some multimillionaires make it clear that they're not going to simply hand over the megabucks to Junior, so Junior better find some kind of job, even if it's a sinecure:
True, but not especially relevant. These particular multimillionaires chose not to “retire at birth” their offspring. That does not affect the choices of the offspring who were retired at birth.
It is tempting to think that they would be like 19th century English parsons, and quietly work on some sort of groundbreaking scientific discovery, but I bet that most people would very quickly go to seed.
How many people "going to seed" would you find acceptable if you got one additional generational talent in science, business, sports, art, etc. in return?
This is to be wished for, but also getting there from our culture will be hard and we should understand that just giving everyone an UBI sufficient to live on for life will probably have all kinds of surprising and often unpleasant social effects. Some people will become poets and artists and such, but a lot will just drink and f--k. (OTOH, maybe that solves our falling TFR problem....)
The citizens of oil sheikhdoms in the Gulf live quite sheltered and financially secured lives, with servants catering to their needs and qualified foreigners doing most of the intellectual work. This is about as close to UBI for citizens (again, citizens, not the general population) that we have right now.
That region is also famously intellectually and artistically sterile. Measured by creativity, UAE, Saudi Arabia, Kuwait, Qatar etc. barely register on the global ladder.
UAE, Kuwait and Qatar have funnily low populations (I mean citizens), and while Saudi Arabia has cca. 10M citizens, the total number of people you listed is less than 15M. Not sure what conclusions we might draw from such a sample.
(Another question is, which countries (or, more like, cultures) can produce something which the world sees as "creative"? I mean, if you are not from an Anglo-Saxon country, nor from Germany or France, who cares what you think or do? But I guess you get the idea anyway.)
Small populations are absolutely capable of producing outstanding talents.
Ancient Athens had at most 160k people and we still know their philosophers and artists by their name and work.
18th/19th century Scotland had maybe 2 million people and produced maybe half of all the scientists and engineers who pulled the Industrial Revolution forward.
How many Ashkenazi Jews are out there and how many of them got a Nobel Prize?
I have to agree with DrMcleod. Islam effectively prohibits representative art, and its philosophy of fatalism ("God willing...") discourages pursuit of the unknown. It really is a religion hostile to arts and sciences.
I don't disagree with your point about rich kids, but I don't think it settles the question. They generally have high social capital and are probably at least as intrinsically motivated as the median person. I wonder what will happen to the bottom quartile in motivation. Maybe it's fine if a bigger share of the population ends up spending most of their lives partying, gaming, or whatever, but it's a big leap from how society works now.
When I think about UBI, I think about the Australian Aboriginal communities where everyone is on welfare and nobody does any work. And everyone just sits around getting high off whatever drugs they can get their hands on.
Realistically, do these Australian Aboriginals have any other options? If they wanted to learn things toy robotics or scuba diving, is it physically possible for them?
I mean, in 2026, I don't think anyone with free time and basic resources met has a reason why they can not study drawing, philosophy, learn a new language, etc.
There will always be a percentage of people who abuse any given social security net. But if that net, for every 10, 50, or even 100 such people enables a new Einstein or Freddy Mercury or what have you, it will have been worth it.
Name me a single Einstein or Mercury who comes from, say, Saudi Arabia, or the circles of the Russian nouveau riches.
I live in Europe. The economic differences here are vast. Switzerland is much richer than Romania or Lithuania. But the Swiss don't seem to be particularly more creative than Romanians or Lithuanians are; what they can do, is to attract brains and talents from the outside, but quite a lot of those people grew up in some backward region of Eastern Europe in a dreary block of flats.
You named Freddie Mercury. As a teenager, Mercury actually had to work for money, being a stall seller and a baggage handler on an airport. He certainly didn't have a generous time- or financial budget for his musical training. It was mostly raw talent that got him through.
I guess I'm much more pessimistic about human nature than you. Maybe just projecting my own terminal laziness, but I honestly think society needs to compel people to work (economic and social pressure) until we can automate most jobs. Most people hate their jobs, because the venn diagram of jobs people want to do and jobs that need to be done don't overlap much.
The whole idea is that with enough automation, not everyone has to work. Forcing everyone to work as before just because some might take advantage just doesnt make sense. "The optimal amount of fraud is non-zero", and the optimal amount of lazy freeloaders is also non-zero.
It is a strange error, yet somehow a common one, to say "people intrinsically value working because it's an identity marker / makes them feel useful, so we shouldn't do UBI because it will take that away!" It seems to me that the initial premise implies people will just *keep working* if you give them a basic income. Which is what they mostly do, according to the results of the UBI pilot projects I've read about.
I do not think it is an error. I think it is a deliberate lie. I would go so far as to say that absolutely everyone who tells you “You need work for life to have meaning” (Jordan Peterson comes to mind) has vested interest in keeping peons busy.
In these discussions people seem to conflate "work" (any productive activity) with "job" (work done for a stranger, in return for money). I find it plausible that work is necessary for mental health, but that doesn't imply that jobs are. You could work on your hobby project, or do something useful for your family, or your community -- this all is perfectly compatible with UBI.
Without UBI, you get billions of suddenly not-so-important people desperately competing for the few remaining jobs that are not done better by AI (including robots) yet, and if they fail their families will starve. If you are the right kind of of person, you may enjoy watching this. Most people won't be happy about that.
Take something uncontroversial (people have a need to feel useful and productive) and conflate it with one much harder to defend (the only way to feel useful and productive is to be wage slave in Neoliberal capitalism). It is never a mistake, it is always a deliberate lie.
Love your LessWrong article! I always said more or less same thing -- and I say it as someone who for most of my adult life DID enjoy my jobs.
'Wage slave' is an oxymoron. If you're going to lecture people about fallacies, at least try not to use them in the very sentence where you're doing so.
I wrote a post about the history of toilets, covering 5000 year old plumbing in the ancient city of Uruk, the very public latrines of Rome, the invention of the modern flush toilet in the 1500s to the present day. Check it out if you're interested! https://postalpigeon.substack.com/p/the-history-of-hygiene-1-the-toilet
Sometimes people argue that things like bicycles or wheeled suitcases or steam engines could have and should have been invented a lot earlier and the people of the past were nitwits for not inventing them.
Others counter-argue that actually they couldn't and shouldn't have been, because of all the other technologies that they're dependent on, and the people of the past weren't nitwits after all.
But if you're telling me we had smelly toilets for a hundred years before someone invented the U-bend, then that's a much better example.
Crazy things happening in biotech. There are multiple ongoing human trials of a technique where AAVs are used to genetically engineer neurons into expressing designer receptors (DREADDs), which are extraordinarily sensitive to a common antipsychotic, clozapine. Very low doses of clozapine can then be used to activate these receptors.
This is the sort of clever experimental technique that can be (and has been) used in research, and it is wild to see it move into human patients. Hope it is successful. It may pave the way for many similar clever techniques in the future.
Count me in as skeptical it could reach that many neurons too - the whole problem with "non embryo gene editing" is how you get the edits into the trillions of cells in the human body, and AAV's don't have a great record on that front. Immune response is exactly the problem.
And I'm by no means keeping up with the cutting edge, but I thought of lipid nanoparticles, AAV's, and lentiviruses, lipid nanoparticles were the only ones with a good track record of success (Transthyretin Amyloidosis), and that they've even engineered some that can pass the blood brain barrier. And even most of the successes have only worked in small, localized organs, the brain would be one of the hardest ones to do, due to the sheer number of cells.
And obviously "getting the virus into trillions of cells with high incidence" is in tension with "inspiring an immune response."
Overall a sad story, and I'm on the fence whether Qiu should be stomped on. On the one hand, this is how you end up at the US, where no progress happens, the FDA kills way more people than it saves, and more and more human trials and pharma development are getting outsourced to China. The only way you can keep progressing and doing better is letting guys like Qiu keep pushing the frontier.
On the other hand, there actually could have been skullduggery there, pace the neuron chart and our priors that getting edits into most neurons would be very difficult. So I think that's ultimately where I'd land as the crux - if there was skullduggery, stomp, if not, we need to let the guy cook, and the parents signed all the releases and are clearly just angry in retrospect it didn't work out. Although they suffered a great tragedy, it shouldn't be allowed to rein in medical progress as a whole for the entire world.
All the "the financial distributions were irregular" stuff is mostly handwaving, that's just how stuff is done in China.
Also, as an aside, those "ipads and cases of Moutai" the dad was regularly giving Qiu? Moutai is like $1k+ USD a bottle if you're getting the good stuff.
Why is the US still subsidizing corn ethanol when instead subsidizing solar farms (on the exact same acreage) could power the entire US, with zero impact on food supply?
(An acre of utility-scale solar generally produces roughly 20–30× as much usable energy per year as an acre of corn grown for ethanol.)
At the local level, many farmers (or farm families) want to keep farming, even as productivity gains and changing consumer preferences reduce the number of farmers needed. Interventions like ethanol subsidies and "Got Milk?" help forestall the decline.
The bits of the US that are used to grow corn are good agricultural land that could be used for pretty much any crop. If the total amount of agricultural land in the US decreases then it's going to be the marginal land that goes away, not Iowa. Stop subsidising corn in particular and some of that land gets turned over to more economically sensible uses, which might be wheat, or tomatoes, or tulips, or whatever.
And if you're going to tile a state with solar panels then you'd pick Arizona, not Iowa.
The beauty of subsidies is that the market can allocate the land efficiently. If cutting corn subsidies and creating solar panel subsidies means more tomatoes in Iowa while the solar power comes from Arizona, that's fine! I assume the grid is capable of selling power from Arizona to other states.
Sounds like this would be a loss for corn farmers in Iowa and Nebraska, so we shouldn't be surprised if they use their political power to keep the subsidies flowing.
The United States doesn't subsidize corn ethanol because it is an efficient way to power any part of the United States. The United States subsidizes corn ethanol because farmers vote, and are sympathetic enough that other people will vote with them if it doesn't cost them too much. There doesn't seem to be a similarly large and sympathetic population ready to vote for solar-farm subsidies.
Wildly untrue. Having spent the past decade working directly with the families who own the great majority of Midwestern farmland, I've been disabused of several urbanite myths and that's one of them.
What do you mean? I don't think the amount of money going to small farms vs big farms is a question that you answer by spending time on farms, but rather adding up numbers in a spreadsheet.
I'm not seeing either of you support your arguments with data. Shame on both of you. But with the help of an LLM, I quickly found out that Paul Botts is correct. According to the USDA...
"In 2024, 97% of farms were classified as family farms, while only 3% were nonfamily farms. In 2024, family farms collectively operated roughly 91% of U.S. agricultural acreage (the nonfamily share was 8.9%)"
But many farms rent their land from other owners. Looks like roughly 40% is rented from other entities, including big corporations (and Bill Gates owns about a quarter million acres). But something like 70% of that 40% is owned by small LLCs and family trusts.
On topics for which the facts are readily obtainable by anyone genuinely interested, my standard approach now is not to engage until/unless the person being absolutely certain offers some of those facts. Which they rarely do, so on second encounter my mute list gets a new addition.
To be clear I do not defend this approach which younger me would have been appalled by; I accept the shaming. I am though now old and cranky and life’s just too damned short to engage with the empty hand-waving.
So, Beowulf, I appreciate you going in search of some facts, it’s refreshing.
No doubt what the LLMs are mainly scraping is the USDA’s nationwide Census of Agriculture, conducted afresh every five years for over half a century now and the gold-standard such among all developed nations. I’d never heard of it until a decade ago when the farmers and farmer trade orgs I began working with explained that it’s their bible for business-planning purposes. It is a treasure trove of interesting information and this born-and-raised city boy has learned much from it. The current administration seems not to have yet noticed it and so far as I can tell it is still operating without politicization. Well worth the time:
-- the states in which I work directly with farmers and farm orgs (Illinois and Indiana) happen to rank very high for leased farmland. When you look it up, it mostly turns out that Smith Farms LLC (they all have LLCs nowadays) is renting from the widow Susie Jones who’s 83 years old and grew up on the farm that she’s now leasing out while living out her days in a retirement village in Peoria or someplace.
Then sometimes the guys/couples/extended families currently farming corn or soybeans or wheat are leasing directly from each other for, reasons. When you meet a new operator his self-description is always something like, “Well I’ve got [“got” means he both owns and farms it] 1,500 acres in ____ County, then my brother and I are leasing 1,000 from [a local widow Jones] over by ____, and next year so-and-so from [neighboring county] is going to start leasing that 450 acres we still have from that old swap our granddad did cause he wants to try a new variety of [crop]….” There’s also now a whole layer of ag-land brokers in small-town offices across the Farm Belt, who sometimes act as realtors, sometimes as rental agents, sometimes as “farm managers” of land that a widow Jones doesn’t have a tenant for. Those guys always turn out to have grown up on a farm someplace (nobody would let them handle land if they hadn’t), and are a strong risk-aversion influence (only willing to do the tried-and-true as far as crop choices).
-- another way to express the Bill Gates stat is that he owns .000286 of the nation’s ~875 _million_ acres of active agricultural land.
-- regarding farm subsidies, I have learned that the definition of what counts as a “subsidy” is contested. But taking a broad definition (any federal dollars paid to farmers for any reason), they usually amount to between 8 and 15 percent of farm revenues nationally. That percentage has occasional big spikes of which 2020 was one, but aside from that year the annual average over the past 15 years is around 10 percent. (Standard reminder that “revenues” is not “profits”.)
Nearly all farm-subsidy dollars end up going to growers of bulk staple crops: corn/soybeans/wheat/cotton/rice/peanuts. Since those crops are most effectively grown for market on larger fields/farms, those doing specialty crops or hobby farms or fruit orchards or whatever receive little from the various subsidy-type programs. A 2,000 acre farm (very normal now in the Midwest) is more than three square miles. These facts lead to sound bites about how “the largest 10 percent of farms [or in the NPR argot “agribusinesses”] receive 75 percent of the subsidies”. Most of those though are still family-owned, just bigger operations than what our great-grandparents were used to. Institutional investors defined broadly own around 5 percent of US ag land of all types and an even smaller fraction of row-cropped land, and correspondingly receive small fractions of farm subsidy dollars.
> Crop insurance subsidies are consistently heavily concentrated on the largest farms. The largest 10 percent of farms receive 56.4 percent, and the largest 5 percent receive 36.4 percent, of all crop insurance subsidies.
> Crop insurance subsidies per acre are larger for larger operations.
Is unregulated power sources here a synonym for renewables. What caused the Spanish issue was bad grid management. Traditional turbines are more stable, but instability in solar generations can be compensated for. Otherwise Spain has done well from renewables.
Yes, this is what you're saying, by absolving Spain's solar production of any responsibility whatSOever in actually regulating itself.
It's only a synonym for wind, solar, and various unviable economic activities that people are terming "renewable." Water-powered renewables are fine, as is geothermic, as is nuclear (thanks France! Road To Zero!).
Don’t tell me what I am trying to say. I am saying what I said - the grid needs better management when there are more renewables. The Spanish investigation is concentrating on why stabilisation failed, but none of this means that solar is more dangerous if managed. Educated as I am I engineering I don’t see this problem as unsolvable, therefore it will be solved. In the mean time I tend not to waste facts on an ideological argument, it never works. Luckily there’s a mute button.
Aeolian321 is yet another account from the previously banned Wimbli/Zanni/Gawdflea/Grimmoar/many, many others. This is par for the course; I wouldn’t hold out much hope for fruitful engagement.
I've seen a number of people improve commenting etiquette and quality dramatically over time in this community after having previously been in more toxic bits of the internet, so it can happen!
I can cite you the report I was reading from Spain (and will, when I get a chance). It's public, and a "what went wrong." In it, I did not find any "the renewables did wrong" (and, perhaps, that is in fact how the system is set up). But, in the case where you have an unregulated bull running around without a fence, and the clown fails to keep him away from the audience, I'd say there's at least a little fault that should be given to the people who said "unregulated bulls are fine."
I have no insight into that question directly. But I note that even if electricity was cheap and bountiful, the US would still need fuel for the non-electrified fleet of vehicles and more. You do mix in ethanol in your fuel over there too right? (Writing in Sweden)
An even more interesting prospect is why not both? Agrivoltaics is becoming all the rage I hear. Many crops even benefit from the partial shade (don't know about corn).
I wrote a UNSONG-style biblical exegesis, on why the meme song "Rockefeller Street" is actually an allegory for Jesus descent into and Harrowing of Hell; very much inspired by how Scott connects the song "American Pie" to Christian soteriology here: https://unsongbook.com/interlude-%D7%9E-miss-american-pie/
A pretty interesting exercise in any case as a way of calibrating ones levels of apophenia; i.e. you develop somewhat of a better feel of what baseline spurious connections one can do when there's really nothing there.
This is tangential but I guess I'll use this as the UNSONG thread. I recently went to an event celebrating Kate Bush and was listening through her entire catalog, and she was pretty eclectic and just wrote about whatever she wanted, but even so it took a minute for me to realize I was hearing the lyrics to "Lily" correctly:
If someone thinks life post-singularity will be unimaginably good (for example, containing engineered pleasures beyond anything unaugmented humans could ever experience, and potentially lasting for an unimaginably long time because of immortality technology), should one make increasing the chances of living long enough to experience it the sole motivation of their current life, because of how much the benefits of this possible future outweigh any present benefits in life? For example, should a person try to minimize the amount of times they drive places or fly on an airplane to minimize the chance of dying in a transport accident, despite this avoidance making their life less enjoyable in the present? To what extent would you go to maximize your chances of long-term survival if you think there is something overwhelmingly good to survive for?
Pascal Wager depends on the fact that believing in the Christian God has no downsides (and that's where he gets it wrong). But but driving, not flying and otherwise restricting your activities comes with considerable downsides for which you might never see the payoff.
Absolutely not. Pascal's wager is between a few sinful pleasures in this life and an infinite reward; I think that the analogy is basically exact. At most, you substitute contingency tables for the authority of the church on what is to be avoided.
That's not the only problem with Pascal's Wager; another problem is that it considers only two possibilities -- Christian God vs. no gods -- when in reality the number of possibilities is virtually infinite. For example, maybe there is some cosmic supernatural force that will shunt your consciousness to a Utopian afterlife, but if and only if you wear predominantly blue clothing on alternate Tuesdays. Or maybe it's Wednesdays. Or maybe it's red clothing. Or...
I don't see how this is connected to Pascal's wager at all. Pascal's wager is about uncertainty, but this scenario does not specify any uncertainty about post-singularity outcomes at all.
"thinks life post-singularity will be unimaginably good" contains, IMO, two statements of faith. 1 a value judgment of what life post-singularity will be and 2 that singularity will happen and that there will be life post singularity.
Even if these were "statements of faith" (which they're clearly not; it mostly sounds like you're using "faith" for "incorrect belief" which is not a useful use of the word), that wouldn't connect it to Pascal's wager, in which the firm belief does not exist.
I suppose it is rather Pascalian, but I have genuinely been coming around to accepting a greater degree of Pascal-style reasoning rather than thinking "that's just Pascal's Wager, disregarded." I blame Bentham's Bulldog.
I'm not suggesting that it can be disregarded, but rather that your question falls into the same type of trap, which is well studied in mathematics: when should I pay a finite cost now for a future reward which has unbounded expected value?
This was really what Pascal asked; and as he had an agenda, his answer was that the infinite reward beats the finite cost every time, as long as you assign a non-zero value to the probability of the existence of God.
I guess your problem still exists in subjective time, which maybe gives you a way out: choose a discounting rate, a probability of future bliss, a value for that bliss *and a probability that you are one of the elect that gets to experience that bliss*, and compare that to the cost of the experiences you forgo now. Of course that's all very subjective.
How different is this to the type of clean living Pascal would anyway have wagered on? Air crashes are famously rarer than we think they are; the interventions most likely to extend your life are clean diet, exercise and etc.
One should never have a sole motivation in life, especially not for a nebulous concept based on the work of other people. It's the whole "eggs in one basket" thing. Imagine you give up all those things, and then the post-singularity is worse than now.
I made a daily calibration game called Give or Take. Five questions a day about real quantities. How many hours an adult mayfly lives, what year the first Star Wars film came out, the heaviest pumpkin ever grown. You don't guess a number, you set a range.
Scoring is 100 * (1 - w)^1.5 if the answer falls inside your range, zero if it doesn't, where w is the width of your range as a fraction of the plausible domain (log scaled where the quantity is). So you get almost nothing for a wide safe bracket, and nothing at all for a narrow wrong one.
The instruction on screen says "set a range you're sure contains the answer". Across 245 answers, those ranges contain it 51% of the time, at an average width of about a fifth of the domain. That's inside the 30-60% band Russo and Schoemaker got when they asked professionals for 90% intervals.
I don't want to over-read it. My n is small, and unlike their task mine pays you for being tight, so some of that gap is probably people trading hit rate for points on purpose rather than being badly calibrated. I haven't worked out how to separate the two. If anyone knows a scoring rule that rewards precision without contaminating the calibration measurement, I'd be interested.
https://give-or-take.com/
No account, runs in a browser. I'm curious whether this crowd's hit rate looks different from everyone else's, which is something I can check.
The US-Iran war has probably increased the chance that China will invade Taiwan soon. Even without that, they were supposed to be ready to invade by 2027 or so.
Have the authors of Plan A taken Taiwan war in question?
I don't think the AI agreement with China is going to work until after said invasion is finished.
Given what we've seen in Ukraine, I don't see how a Chinese invasion of Taiwan is supposed to work right now. How can you get an invasion force across 130 km of water when every boat can be pinpointed and then attacked with missiles, then long range drones, then artillery, then short range drones?
I'm pretty sure China would start with gaining good enough air superiority.
Taiwan has short-changed its military.
So does the opening of chip factories in America, that make the geostrategic importance of Taiwan dwindle. 2027 sounds about right for "America may not care that much anymore."
+1
In what sense, @B Civil ?
Trying to decide what the other person is going to do if you do what you want to do
There's a game of rock, paper, scissors for you
Why locking up everything behind plastic might be the wrong call
I have noticed a profound shift in shopping experience that has made it unpleasant and I think potentially tackling the root causes of the theft would provide better outcomes for the shopper and to decrease the theft activity the retailers want.
Disclaimer: I have tried to pull data from viable sources and recognise that there is some weakness in my argument because only 16% of thefts are reported and categorised but I believe working from the best base from reputable sources such as centre of social justice NBCC police dot UK is the best path forward.
I did a bit of research to know which 3 categories are the main culprit behind theft and it gives a good focus on which areas to target and the potential upside on interventions. (No point focusing in areas that have little impact on theft numbers
Incident Volume vs. Monetary Loss Breakdown
Offender Type Share of Total Incidents Share of Monetary Value Lost Avg. Loss per Incident Primary Mechanics & Targets
1. Chronic / Substance-Driven Repeat Offenders 60% – 70% 35% – 45% £30 – £100 High-frequency daily thefts to fund severe Class A drug/alcohol dependencies. Targets high-resale everyday goods (meat, spirits, coffee, confectionery).
2. Organized Retail Crime (ORC / Gangs) 5% – 10% 40% – 50% £500 – £5,000+ Coordinated rings executing bulk sweeps for structured fencing operations. Targets baby formula, high-end cosmetics, OTC medicine, power tools, and bulk alcohol.
3. Opportunistic / Amateur Shoplifters 20% – 30% 10% – 15% £5 – £30 One-off, low-premeditation theft, largely at self-checkouts or high-friction store zones (e.g., mis-scanning
Proposed solution and comment
Going by each of the categories, I think there is a lot of upsides in helping chronic substance abuse offender as it will have lots of public benefits but also, I think it is an area of intervention which is the easiest to solve. Instead of dealing with the issue at the tail end interventions prior to them becoming drug addicted or creating a path for them to get recovery would be best, to aid with this matter I think producing an insurance levy tailored community drug/alcohol rehabilitation orders, housing support, and dedicated diversion courts. The economic rationale is every insurer benefit when retail theft drops, but no single insurer will voluntarily fund local addiction treatment because competing firms would free-ride on the lower crime rate without paying. A universal statutory levy eliminates this free-rider problem, internalizes the hidden cost of shrink, and channels capital directly into addressing the root driver of recidivism.
Organised crime requires different tools and methods that are not easy to resolve and require lots of different small changes to make good outcomes such as improved coordinating and tackling of the problem so I don’t think I have a solution to put forward.
Opportunistic shoplifting is a low-priority problem and I think existing physical and digital interventions are already best placed to resolve this problem like spot checks on self-checkout etc
Is my current levy solution something that would be good? Any input on the other sources of theft? Thoughts on the post?
TLDR: I think potentially with the makeup of thefts in the UK a levy on the insurance industry will be a good solution to mitigating some of the impacts of increased theft as of late.
Don't know whether this situations applies to other countries, but in Slovakia, it is quite difficult to prosecute organized crime. There are many situations were "everyone knows" what happens, but the police is unable to *prove* it (using only the legal means, prove it beyond the reasonable doubt, assuming that the criminals both have good lawyers and can threaten witnesses). This may or may not be on purpose, because our current government is connected to organized crime, but some problems are probably general.
For example, there is a network of pawn shops built next to supermarkets. "Everyone knows" that the purpose of these pawn shops is to buy stolen goods. But the shop owners are legally protected; they just make the thieves sign a paper saying "I declare that this is my property and not stolen" and that's it. When the police investigates, they show them the signed papers. The burden of proof is on the police, and even if they can prove that a specific thing was stolen, that just mean they can confiscate that one specific thing. The thieves are drug addicts; if you put one in prison, there will be ten more the next day.
A solution in this case would be to somehow put the organizers of the crime in prison, and either ban the pawn shops entirely, or at least make a law saying they cannot be next to a supermarket (or other place where it is easy to steal).
The supermarkets are easy victims, because their business model depends on lots of people doing their shopping relatively quickly, so anything that slows down people would reduce the profits, too.
The solution is to massively increase the penalty for trespassing, and enforce it.
What about a more techy solution? You know how at self-checkout the system requires you to put an item you've rung up onto a weight-sensitive plate on one side of the check-out scanner? What if there were something like that for the items currently kept behind plastic? Put something on each of these items that registers when it is removed from shelf. If it is not returned to shelf of placed in shopper's cart or basket within a certain number of seconds, an alert sounds.
You'd get a ton of false positives from people picking something up to read the back, forgetting which spot it came from and putting it back wrong. As well as the inconvenience of needing to take a basket or cart on every trip.
About the false positives: You could set the system up so that if the item went on any store shelf there was no alert. (The shelf censors would not identify objects by weight, but by some kind of dingus that's attached to them and that shelf can sense from a couple inches away). But I can't think of anything around needing to have a cart or a basket, though. Doesn't seem so bad to me. If you aren't buying much, just grab a basket.
My idea isn't wonderful, but I'll bet tech is good enough now to make possible a more effective and less aggravating solution than locking stealables behind a plastic wall.
>Any input on the other sources of theft?
Well, not in the UK but on the continent in Belgium, there was the Antwerp diamond heist in 2003. Around 150 safe deposit boxes were looted of 100 million in loose diamonds, gold and other jewelry. A 12 step program probably wouldn’t help those Danny Ocean types though.
Antwerp diamond heist:
https://en.wikipedia.org/wiki/Antwerp_diamond_heist
I recommend that you look up the evidence for effectiveness of treatment of chronic drug abusers. If you make them attend treatment, they in fact show more improvement than those for whom treatment is merely recommended. But they still do not do a lot better. In the study I looked at re-arrest (for drug-related and other crimes, combined) was 50% for non-treated group and 38% for treated. Not a lot of bang for the buck. Also bear in mind that there is a free drug abuse treatment available for those who choose on their own to get help: AA.
Thanks Eremolalos,
I appreciate you responding to my post and engaging with my topic. Your point on compulsion has some merit it would be a 50→38 (24% relative) drop in reoffending which is a good number. However, I considered maybe using money for a more target approach and after a bit of research a pivot towards a Risk-need- responsivity (RNR) targeting would be better still. It’s a more tailored approach and a study an English study with alcoholic offenders found a tailored approach to the individual such as matching it to the reoffending risk and criminogenic need showed lower reoffending than the control.
On the economic angle If you are going by per incident cost it can look not cost effective but seen through a lens of a year timespan the numbers look more favourable. The current problem is costing 1bn a year on prevention spend alone according to the British Retail Consortium moreover having items locked away has its own cost and the industry is recognising this negative problem
On things that are not ideal I have not tackled how to deal with organised crime which is more complicated to deal with and I don’t have a tidy 1 pound spent on drug treatment gets x amount of cost savings.
Today it is the anniversary of the Soviet-led invasion of Czechoslovakia (you could technically say that it was August 20th, as the operation started before midnight - but the vast majority of the citizens only encountered the enemy on August 21st, therefore this date is usually the infamous one).
On my Central European Substack, I wrote a mid-sized article about this event.
https://komando1.substack.com/p/when-the-brotherly-tanks-arrived
If I understand it correctly, the Soviet invasion reverted all political reforms except one: federalization. The other reforms were considered capitalist and dangerous, but there was nothing politically incorrect about federations -- Soviet Union itself was technically a federation.
Yes, this was the only permanent change that didn't get rolled back.
I absolutely despise that this website is connected to a twitter. Most "notes" I'm recommended are images!!! And people have just decided that yes, I am happy to conduct myself here as I would on any other twitter because that's good for the blog.
I ignore Notes. I suggest for your peace of mind you do the same. Nobody (as yet) can force us to engage with social media if we don't want to, the same way that the push to force everyone to use AI still cannot make those who don't want to use it sign up, except perhaps if we're employed somewhere that makes usage of AI compulsory.
I didn't even know Notes existed. When you mentioned it, I thought you were talking about Chat and using the wrong term. Just don't read Notes, then! I never even get notifications of notes, because in settings I haven't asked to be notified. Jeez, just don't look at Notes.
I know that Notes exist, but don't understand how exactly it works, so I ignore it.
I saw this recent story, where I had no idea the Chan Zuckerbergs were funding a school. This article is much more critical of the CZs than the others I read, and I imagine there was a lot more going on with the school than merely funding issues, but still it's a neat encapsulation of why all the EA-adjacent criticism of traditional charity falls flat with us traditionalists - start off with social issues, pivot to AI because that's the new sexy shiny after you get fatigued by finding out it is indeed true "the poor you will have with you always":
https://futurism.com/future-society/mark-zuckerberg-wife-shuts-down-school-disadvantaged
"The tuition-free school in the heart of Silicon Valley, simply called The Primary School, was forced to close after Chan’s philanthropy fund, the Chan Zuckerberg Initiative, announced last year it was no longer going to fund it at the end of the 2025-2026 school year.
As the Wall Street Journal reports, the unfortunate development highlights how some of the most influential minds in the tech industry are shifting their priorities from supporting the tackling of social issues and — as the story goes — toward scientific research instead."
To make the EA parallel even more uncomfortably close, the CZs have bought a castle of their own. They're going to be neighbours of mine! Well, within 30 miles or so, but I don't expect to be dropping in for tea. Besides, we already have a duke and his little summer cottage in the county, as well as a marquess who holds an Irish title, so this new money aristocracy don't impress me much 😁:
https://wlrfm.com/news/zuckerberg-and-dyson-to-be-neighbours-in-west-waterford-4482081/
My question is: why are there poor people in East Palo Alto anyway? Land use in the US makes no sense to me. Rich areas and poor areas should be separated by large in-between areas, not directly adjacent to one another.
The median house price in Palo Alto is about $4 million, the median house price in East Palo Alto is $1 million. Why aren't the mid-level FAANG employees priced out of Palo Alto saying "Hey, let's just buy something in East Palo Alto, it's not quite as nice but it's much cheaper" and advancing the wave of gentrification? I've looked around East Palo Alto on Streetview and it looks fine.
And then on the other side... $1 million is cheap compared to $4 million but it's still not a house that's affordable for genuinely "low-income" people. So why are there apparently 500 low-income kids in East Palo Alto living in $1M houses who need charity?
I don't know the area at all, but quick Googling informs me:
"Small Business Domination: Over 80% of local businesses in the city employ fewer than 10 people.
Workforce & Demographics: The city maintains a labor force of over 16,000 employed residents heavily active in retail, accommodation, and regional tech supply chains."
So there are enough people working service jobs that would constitute "poor people". Wikipedia informs me that gentrification is at work, but only since 2000 or so:
https://en.wikipedia.org/wiki/East_Palo_Alto,_California
Historically it was majority black population, now it is majority Latino.
"In 2023, the US Census Bureau estimated that the median household income was $104,832, and the per capita income was $40,141. About 7.0% of families and 9.7% of the population were below the poverty line."
"The revitalization projects in 2000, and high income high-tech professionals moving into new developments, including employees from Google and Facebook, have begun to slowly eliminate the historically wide cultural and economic differences between the two cities [Palo Alto and East Palo Alto]
...Significant gentrification occurred in East Palo Alto from around the founding of Facebook, with the construction of a large shopping center named Ravenswood 101 and several upscale housing communities intended for high-earning Silicon Valley workers. This development faced opposition from some residents, who charged that it priced locals out of one of the region's only affordable communities while providing only low-paying retail jobs and consuming disproportionate land area (2.2 square miles). Supporters pointed to an increased tax base."
"So why are there apparently 500 low-income kids in East Palo Alto living in $1M houses who need charity?"
Those are not the kids living in the million dollar houses, the people in the million dollar houses are the tech workers "priced out of Palo Alto" who are living in the nicer houses built for them when they came to work in the FAANGs.
Your two questions answer themselves: The 1 million house prices in EPA *are* the effect of spillover gentrification. As for why house prices aren't literally equal everywhere, well I imagine that school districts play a big role in that, in addition to general amenities, location, etc. Not everyone wants to live in the poor areas.
As for low income people: Not everyone lives in a house you know. Lots of people here are renting apartments (which are *also* crazy expensive, but don't require putting down a million dollars). And as I'm sure you're eager to point out on ACX in any other thread, there are homeless people too.
How can the homeless people afford to live in such expensive areas is the greatest mystery of all.
Just kidding, but seriously, if I was a very rich guy, I would probably prefer to live in a less-than-maximally-rich area, even if I could afford it, just to avoid the homeless who seem to be attracted to the richest areas.
> How can the homeless people afford to live in such expensive areas is the greatest mystery of all.
I actually have a solution to homelessness that nobody wants to listen to, which is for cities to charge a daily per-person fee, like the "congestion fee" that some places charge, except for people.
Take San Francisco for example. The city budget is $16 billion, spread across a population of about 800,000 people, so that's $20K per head. Abolish all other taxes, and turn it into a $55 per day fee to exist in the city.
For tourists and business travellers, it's a relatively insignificant additional fee for the privilege of visiting the city. For actual residents, it's going to be an overall tax reduction since the cost will now be shared with temporary residents. The only people who will be really priced out are the homeless, who will have to find somewhere else to sleep.
The people you are going to price out are the service industry workers, the people who deliver your food, clean your offices, and work in the hospitals.
Yes, even nurses and teachers are priced out in London by 'London pricing', that's why there is a 'key workers/essential workers' policy around income and housing supports for them:
https://en.wikipedia.org/wiki/Key_worker
What you are proposing is the usual pipe dream of "the only people who need to live and work in San Francisco are the rich" and, uh, you need more people in a city for it to work than "software engineers working for Google".
What would you do with people breaking the rules, e.g. homeless people illegally inside the city, not paying the fee but also not leaving of their own volition?
Punishment of some sort.
Put a fence around the city, take them outside.
The castle appeals to them because it resembles a computer chip -- squares within squares, on a square.
If anyone would like a better view of the little summer place they're going to use as a pied-à-terre should Mark be doing business over in Ireland, there's drone footage:
https://www.youtube.com/shorts/kJMdUKLQSmE
The castle isn't very old, it won't be 200 years old until next year. Built in the Gothic Revival style for a then-MP, the notes on the National Built Heritage site are (1) outdated (they come from 2003) and (2) funny:
https://www.buildingsofireland.ie/buildings-search/building/22903402/strancally-castle-strancally-demesne-waterford
"The construction in sandstone and limestone ashlar augments the dour tone of the composition while providing evidence of high quality stone masonry and craftsmanship."
'Dour' is not a very enticing description 😊
It isn't a castle though, it seems to be a large house superficially decorated with some of the trappings of iconic castles. A McMansion of its day.
OK, here ya go, the Super Irishy Gothic Chip McMansion: https://imgur.com/a/Eet9JGN
A very good description.
You know, my first thought about the place was that it was hideous, but I refrained from saying that because I thought you might be offended by my dissing old Irish stuff. But turns out it isn't old, not properly old. I just google-imaged "oldest Irish castles" and thought they were magnificent, and the tumble-down ones magnificently sad. But they're all quite different from the Gothicized block the CZ's bought. They ramble over uneven land & have complex shapes. Within a building there's lots of variation in height, so sky is visible between the elements. . .
I mean it's not really a castle at all, it's a grandiose 19th century house built in a vaguely castle-ish style. Good luck repelling invading Normans with all those ground-floor windows though.
Quote of the day.
>Ukraine’s corruption watchdog released recordings in which a woman identified in the video only as a deputy head of the president’s office can be heard saying: “Fighting corruption is a wrong approach … Corruption has to be systemized and controlled.”<
Fighting corruption drives it underground, where it becomes more of a hurdle for the uninitiated and the unconnected. If it exists as a socially acceptable way of distributing money to the "select bureaucrats", then... it's as regulated and fair as it can be (this assumes a relatively liquid market of bribes, more "fair" than "arbitrary").
Just a stab at steelmanning. I think American corruption has the worst of all worlds.
Have you lived for an extended period somewhere else to pass judgement like this?
... yes? I also listen to corruption as discussed, here and other places. One of the book reviews generated a quite colorful third-world story about bribing the government official because that's essentially part of his pay.
Can you tell about your experience of living there? Is your assertion that "I think American corruption has the worst of all worlds"? based to some extent on your personal experience?
I'm asking since I lived many years in a country in the bottom half of most corruption ratings and my impression is very different
Share yours instead. My ears are open. I'm not trying to say that I'm right, it's just a steelman. I understand that other countries have problems, I just don't see "corruption" as being quite the problem people want to say it is. It causes problems, it -can- be a problem, particularly if it's arbitrary or if it hinders starting business.
Also worth noting that America is around middle-of-the-pack in terms of corruption, with about 60-66% of our Senators blackmailed.
Also, the tradeoff for "American corruption being the worst of all worlds" is that it is relatively rare*. You have pustules on the skin, and not a generalized fever.
*This is assuming you're not going to call "getting government favor via campaign donation" -- ala Santorum's calls to end NWS because AccuWeather gave him some campaign cash. As it's perfectly legal, I'm hesitant to call this corruption, even though no-bid contracts are definitely a hindrance to "efficient market output."
My problem was with the statement that America has the worst of all worlds, which I understood to mean that the corruption in America is worse in some important ways than the corruption elsewhere. this does not match my impressions, hence my questions. Since you said that it's actually rare, I'm no longer sure that I got your point.
Examples are easy to list. Bribes for traffic violations, forgetting a driving license without a road test, for medical interventions in a supposedly free health care system. This is in addition to politicians lining their pockets, not instead of that.
I’m sure most commenters are bored of the Jason Arday story, but here is something that I think adds useful nuance. Something that I think is common knowledge in Cambridge, but evidently not known outside, is that the Faculty of Education is (informally, unofficially) not really considered fully part of the university. Cambridge as a whole seems to view it as a sort of harmless wacky relation: tolerated, indulged etc. I don’t know the full prehistory of this, but it has something to do with the fact that Cambridge (the town) long had a teacher training college (Homerton) which was originally not affiliated with the university but was eventually kinda sorta brought into the fold, but in an awkward, Puerto Rico type way. You can imagine whichever vice chancellor breezily saying, as he OK-ed the merger, "well, what harm can it do?"
So the Faculty of Education going mega-woke and accidentally hiring a fake academic with a literal typo in the title of his PhD is actually pretty unsurprising, and says a lot less than you might think about the rest of the university. (Let's remember that many Cambridge fellows never really retire, so the median Fellow is about 200 years old and is not on Bluesky.) But the extra nuance makes the story less exciting, which I guess is ultimately the reason we aren't hearing it.
To me, though, this gets at something very important. Intellectual life is full of people who are a bit batty. At any given time, some academics are considered by others to be complete idiots. The blue-haired woke warrior, for instance, has been a staple of academic life for decades. But the battiness is tolerated, because after all, the 200-year-olds are batty in their own way too, and because tolerance is baked into the DNA of the liberal elite, for both better and worse.
I'm not bored, I'm appalled. Arday was also a fellow at Jesus College, btw.
That's true. Do you know whether the fellowship at Jesus came automatically with the professorship, or was Jesus College free to conduct their own interviews? If the latter, then that to me is a bigger deal than the professorship itself, because Jesus is more respected than the faculty is.
Sorry, I'm not aware of the circumstances.
Would that make Jason Arday one of the apostles?
(I suspect that if you asked him, the answer might be yes.)
Something similar is true about Columbia U and Teachers college. Claude tells me that tends to be more true with ed schools than other fields. One could be tempted to speculate that is how ed departments got so ridiculous, but I suspect its down more to them being dominated by lib personnel more than any other field and the echo chamber doing what it does. Another force being that so much of its grads go into infamously difficult to hold accountable jobs that its output is less subject to a market test.
I wondered about that a lot: How much of the lack of rigor of a field is a function of its difficulty/complexity?
Education is so ephemeral and individual in many ways that the sort of research a economist does is of limited value. Of course, economics have uplifted education research a lot, but only in quite specific ways.
The tolerance of far-left loonies wouldn't be a problem if it were counterbalanced by the tolerance of an equal number of far-right loonies. But how many of those does Cambridge have?
Tyler Austin Harper's recent essay and podcast appearances have been really good on this. He thinks the key to political diversity in the academy is more job security, i.e. more tenured positions. But the fact that tenured positions have declined so much speaks to the general state of crisis in the academy, especially the humanities.
Having said that, my feeling is that we underestimate how profoundly these things can change. Just as we're seeing a vibe shift in the wider world, there could be one in academia too. It's also worth remembering that universities have come and gone in and out of public favor over the centuries – the current crisis isn't unique.
There *was* a vibeshift up until early 2025, but Trump's war on academia has presumably killed that for the forseeable future.
... I think you're missing a few things here. The winnowing, the removal of right-wing folks, happens way earlier than them getting tenure. Most "would-be" professors would perform sexual acts in order to get a professorship. They're hard enough to come by, that "believes something different from me" is a discouragement. Okay, so maybe that's always been a discouragement. But when you have people who do believe "exactly as you do" as that's what they've been trained to do -- debatably teachers think this is their job, why exactly are you hiring the guy who makes you think, makes your head hurt, makes you FEEL less smart?
Time was the humanities was a bareknuckle fighting ring, held to some pretty damn high standards. Everyone had a different hobby horse, because that was how you did papers.
I don't see "dug through shipping manifests for the Civil War" to prove hoarding and show the demoralization of the South, as something you'd get a current historian to do. It's not sexy enough, it's too "economic" and not "Big Man" enough, and doesn't make a "but what if...?" conclusion -- in fact, it excludes it.
I don't disagree with any of that, but I still think Harper is onto something in saying that the best remedy is job security. People on the whole move rightward in their politics as they get older. So in a university with job security, the students have to tolerate the professors, and the professors have to tolerate the students.
the dish that learned Pong and its successors now running Doom, made for a fun short audio episode on daybrain: https://daybrain.app/dishbrain-neurons-playing-pong?src=46
Seriously though, why haven't Boring Company style projects taken off yet? (Inspired by a subthread, but worth a top-level discussion.)
Airport-to-city transportation seems like an excellent use case. In Melbourne there is a current $A11 billion dollar project to build a heavy rail line from the airport to the city; it will run partially along existing suburban lines so there'll be only 12km of actual new track, and the 20km trip will take 30 minutes.
Suppose instead that we dug a Las Vegas Loop style tunnel. Supposedly this costs about $A30 million per km, and we'll need one in each direction, so that's $A1.2 billion right there. We need stations at both ends, some kind of storage and charging infrastructure in the middle, and we know things always go wrong, so let's just double it to $2.4 billion.
Maybe the cost of vehicles is a problem? Well, a new Melbourne train costs about $40 million and has 423 seats, so that's about $94K per seated passenger. So even if each passenger gets their own private robotaxi and we don't bother to optimise anything else then we're still way ahead. Electricity costs should be a lot lower too, mostly because you can right-size the amount of metal you're moving on a moment-to-moment basis rather than running half-empty trains.
And of course it's a lot faster, there's no intermediate stops, and very little waiting around for the next timetabled train, you just show up and get into the next available vehicle.
All of this is dependent on the self-driving side working; my understanding is that the Las Vegas Loop still isn't self-driving. But this is surely a very solvable problem.
So are there any actual problems here? Or is it just a combination of "politicians hate Musk" and "people are scared of new things" and "public transport activists think public transport should be a hair shirt and hate to see people riding in comfortable private vehicles"?
According to Wikipedia, "At peak times, the Loop was reported in 2026 to have a potential capacity of 33,600 people per day at 100% usage, with each Tesla car carrying the maximum four passengers."
A ten car subway train has a capacity of 2000+ https://en.wikipedia.org/wiki/R160_(New_York_City_Subway_car)
At only 6 trains per hour, only 12 hours a day, that is 144,000 capacity.
>and very little waiting around for the next timetabled train, you just show up and get into the next available vehicle
"A typical wait of ten to fifteen minutes has been reported" per Wikipedia.
Right, if you need to move 100K+ people in an hour then heavy rail makes sense. But in other situations they can be massive overkill.
For the airport line I used as an example, that airport handles 100K passengers a day. Some of them are in transit and some of them will arrive and depart by other means, let's say a maximum of half of those will want to travel to the city -- 25K in one direction and 25K in the other. Passengers are spread through the day, although with morning and afternoon peaks, so maybe a couple of thousand passengers an hour in each direction would be the peak?
Even so, it would make more sense to run busses, like Boston's Silver line, rather than dozens of individual cars.
I assume we're talking about buses in dedicated tunnels? I looked up the Silver Line and I understand that it does that over part of its length.
But buses are a far worse passenger experience, you ride with strangers, some of whom may be annoying, rather than in privacy. And waiting for the other passengers to embark and alight the bus takes time, especially when they've got luggage (which is why everyone groans when their aircraft pulls into a remote gate).
And it doesn't even necessarily save money, because buses are tall and wide so you need a much bigger tunnel. The key part here is to keep the tunnel cheap by keeping the vehicles small.
An actual road-going mass-market electric car isn't really the optimal solution of course, a specialised pod-type vehicle would be better, but it makes sense until specialised vehicles are available.
>But buses are a far worse passenger experience, you ride with strangers, some of whom may be annoying, rather than in privacy
An odd concern re people who are about to board a plane, or have just ridden on one.
And, in your scenario, does a single traveler get his own car? So, four single travelers requires four cars, and two couples requires two cars? Sounds grossly inefficient.
>And waiting for the other passengers to embark and alight the bus takes time,
So does waiting for the people ahead of you to board 10-20 separate vehicles.
>you need a much bigger tunnel
The actual Vegas Loop tunnels are not so small. https://www.autoweek.com/news/technology/a38195296/a-ride-in-the-tesla-tunnel-under-las-vegas/
My understanding is that the Loop is only wide enough for one car, which means you're *still* waiting for everyone to embark and alight. Any delays by people ahead of you holds up the whole line.
That would be a silly design if it were that way. My understanding is that there's space for many cars to park in parallel at either end, so each car can go when it's ready. Bottlenecks are still possible of course and it's a tradeoff between station size and the frequency of jams.
The costs would probably be inflated by whatever is making a plain old train track cost nearly a billion Australian dollars.per km. I don't know what rail projects usually cost, but that seems a bit on the high side to me, which suggests one or more of the following:
1. The government project management system at work here is optimized for something other than cost effectiveness.
2. Something akin to the American difficulties of the planning and review stage of long projects dragging on for many, many years of expensive reviews and redesigns as the project clears the many, many veto points placed in its path by various jurisdictions, agencies, and private lawsuits.
3. Some geographic difficulty at work. Perhaps the project needs more than the normal amount of grading or bridge and tunnel construction.
4. Labor laws and environmental rules make the project particularly expensive.
2 and 4 would almost certainly apply to a Vegas Loop style project. 3 might also: some geographic issues would only apply to one project type pr the other, but some apply to both and there are almost certainly geographic issues that make a Loop tunnel intractable without ruling out a surface rail line. And 1 alone would explain why they might pass over the Boring Company even if it were likely to be cheaper.
Also, did the Boring Company bid on the project at all? There's often some difficulty awarding contracts to companies that don't bid on them.
There's a long running proposal to build a metro in Dublin, which is coming in around EUR12 billion to build. That's higher than, but comparable to, the Melbourne project. About half of that is construction of the tunnel, rail line and stations; the other half is a contract to operate the line. I presume costs in Melbourne work out similarly - it makes sense that underground would be modestly but not outrageously more expensive.
I had a look at Musk's figures - the bore of his tunnels is much smaller than for a train. Claude suggests 8% of the passenger throughput for a Boring company tunnel compared to a Metro type solution, for something like 5% of the cost - if everything were linear (if only!) the costs of all solutions would be within a factor of 4 or so, which is actually closer to a textbook economics solution than I expected.
Passenger throughput isn't a great metric, because most train lines never max out their passenger throughput. The maximum passenger throughput of a metro line would be putting a full train through every few minutes, which few lines achieve, and certainly the airport line won't (they're saying a train every ten minutes).
I know the suburban rail in Dublin is packed to the gills during rush hour, and would imagine that the metro would operate more-or-less at capacity during rush hour too. So the size and frequency of trains is absolutely a consideration.
The Boring Company proposal was for taxi-sized pods, if I remember correctly, which wouldn't really solve the congestion problem that mass transit is designed to address.
Amtrak isn't designed to solve congestion problems. Musk's project would have competed with Amtrak.
I read somewhere that one of George R. R. Martin's inspirations for writing the Game of Thrones series was this guy Tad Williams' fantasy novels from the '80's and '90's, so I picked up copies of a series he wrote called "Memory, Sorrow, and Thorn." Quite enjoyable so far. If anyone else out there suffered from GoT withdrawal like I did, these books will give you a good fix.
I read a lot of what Tad Williams wrote as a teenager and liked it. But I like other authors more.
MS&T is his best long series, I like the plot twists and I liked the the main characters. Though the female lead can be rather annoying from my more experienced view.
Otherland is a great concept, but roughly twice as long as it should be.
I love Tailchaser's Song, it is a truly creative and enjoyable story. War of the Flowers is good, but not great.
I haven't finished the Shadowmarch series. I think I read the first book before the second came out, but didn't make it through the re-read to understand what's going on in the second. I might try again these days, though.
I'd say the main issues with his stories are that they tend to be overlong, and that many female characters have that bitchy American woman stereotype - a bit like Robert Jordan's women.
For comparison and taste reference, I absolutely love everything by Brandon Sanderson. I've noticed a certain divide among people about him.
Thanks, I will file that name away for a rainy day.
If you want something on the epic scale of GoT/SoIaF, but actually finished, Wheel of Time is the obvious go-to, by the way.
I have never read George RR Martin, but I tried to like contemporary fantasy and chose the Dragonbone Chair as it was frequently referenced.
I just found the company of the bratty main character so tedious. I didn’t finish the book so it might’ve been that there was a payoff for all that tedium. That it was an authorial choice.
I think it was Ross Douthat who recommended it, amongst others.
I actually found Douthat’s own substack published fantasy novel more engaging, despite a cipher of a lady heroine, and slightly ponderous world building. At least I think she might’ve been supposed to be the heroine, but I’m not sure.
I read about 2/3 of Williams's "Otherland" series back in college. I found it interesting and enjoyable, but not quite enough so to actually finish the series. I remember the books being very long (a notable similarity with ASOIAF) with a ton of plot threads in parallel that weren't always clearly related (another similarity), but somehow not quite as engaging. Otherland is also cyberpunk-adjacent science fiction rather than low fantasy.
ASOIAF is soap opera, rather than most fantasy which tends towards pulp fiction. Both have their proponents, but the audiences do differ.
I slogged my way through that one long before "A Song of Ice and Fire" was a notion in George's noggin 😁
On the whole I liked it, but I never re-read it and I never followed Williams as an author.
Is "never re-read it" a stinging rebuke, based on your reading habits? I can count on one hand the number of books I've re-read over the years. That's just something I've rarely even gotten the urge to do, much less followed through on. And yet I can watch the same ~50 movies over and over and over again.
It sort of is? Books that I don't feel the need to re-read may be excellent, but often it's "yeah that was okay but it was like a fast food burger" while books that I re-read are "this is an enjoyable experience I want to repeat".
Nothing against fast food burgers, sometimes you just want something fast and consistent Food Type Product, but those sort of books are Story Type Product and not worth more than the minor amount of time and effort I expended on them.
Williams *should* be the kind of doorstopper fantasy I enjoy, and he was clearly *trying* to be that kind of epic doorstopper writer, but I just didn't find his stuff engaging enough. Decent enough to keep reading to the end, but not enough to make me want to re-read it or search out more of his work.
Try his shorter works, especially Tailchaser's Song.
I'll grant that there are a few clunky passages early on in the first book. I thought they disappeared once the story got going, though, but I've only finished 1 out of I think 4 books, so I suppose there is plenty of time for me to revise my opinion if the clunkiness resurfaces.
I dunno...my attitude is there are about a million and a half books I'd *like* to read out there and not nearly enough time to get through even 10% of them, so if I'm going to read the same book twice, it better be a damn near religious experience the first time through. The only books that met that threshold for me (since you asked) I think were Lord of the Rings, The Guns of August, and the second volume of William Manchester's Churchill bio, Alone. Edit: Bill Simmons' Now I Can Die In Peace was another I reread because of the lulz.
Man, there's quite a few books I've reread and more that I'd like to have the time to reread
Yeah, same for me.
I like Tad Williams! Tailchaser's song was a fun one and the Shadowmarch series was unique and interesting as well.
Nice. Thanks for the tip. I'll have to check out one of those next.
You should apply to Inkhaven 3: https://tomasbjartur.substack.com/p/you-should-apply-to-inkhaven-3
When Elon Musk announced a plan or aspiration (not sure which) to launch a large set of data centres into orbit to use solar power and space-based cooling, I was reminded of an article I read on this idea a few months ago:
https://taranis.ie/datacenters-in-space-are-a-terrible-horrible-no-good-idea/
Musk is a smart cookie, and his ventures seem to have been successful so far. So I wonder what the prevailing idea is among readers about whether this space-based data centre idea is likely to work. Perhaps it would be better to site them buried on the moon.
For any space industry question, the central question is "Can this be done more cheaply on Earth instead?" The only true commercial space industries at present are satellite communications and imaging, both activities which are exclusive to space, hence why humans have accepted the immense costs of space launch for these purposes alone.
We haven't sent humans to Mars not just because it would cost $1 trillion dollars, but because there's no economically valuable product, industry, or service which could *only be done on Mars* or is otherwise competitive on Mars with that same product, industry, or service on Earth. This is the logic behind the quip that we should expect to see seafloor or deep Antarctic cities before we see space colonization, because even a seafloor city will be cheaper to build, operate, repair, and staff than a space habitat.
Even if space launch were 1% the cost it is today (and one should be skeptical of the long track record of overpredicting and underdelivering space launch cost reductions) you would still need to ship in every single gram of everything you need in space. It will never be easier to launch an industrial facility into space than to build that same facility on the Earth!
If economics is the main consideration, all space-datacenter-specific issues aside (low-temp waste heat rejection problems, vulnerability to malicious interference, debris impact rates on vast solar & radiator arrays, obsolescence or maintenance issues, etc.), it seems unlikely that any society driven by profitability as ours is would ever seriously develop any real space industry at all, let alone datacenters. That's why we've haven't sent people to Mars despite having the tech and plans for it since Apollo.
Space infrastructure development is currently split between "economic activities which can only be done in space" e.g. imaging & communications and "military/scientific initiatives funded by the government and not accountable to profit incentives".
I seriously doubt that political opposition to terrestrial datacenters can justify the (multiple orders of magnitude) premium in the construction and operation of such facilities. Nor would building a space datacenter be easier (or more politically palatable) than addressing and negotiating the concerns of folks on Earth about datacenters.
Space-based data centers are among the stupidest ideas I've ever heard. The result of hearing Musk propose them should be to severely downgrade your "Musk is smart" probability, not to throw everything you know (or ought to know) about efficiency and logistics out the airlock because I a guy with a big bank account said so.
Musk has talked openly about being on the autism spectrum. That has nothing to do with intelligence in the IQ, shape rotating, sense. He’s not splendid in the wordcel sense though. He has mentioned having trouble understanding things like figurative speech as a kid so he may just not ‘get’ things the way a lot of other people do. No, Elon, most people will think offering to give Taylor Swift a baby is creepy rather than badass.
He also has the most ‘fuck you’ money in the world so I doubt criticism from the unwashed masses is going to have much effect on what he does or says or does. More likely that sort of thing just makes him want to double down on whatever gets him in hot water.
I agree but just saying this without any actual argument does not provide any value to the conversation. This comment is 100% heat.
Less of this. I buy that space data centers are a bad idea but saying so, especially in this tone, without providing even a single argument doesn't add anything to the conversation.
Eh, I'll grant that it was a quick, minimalist comment (and not my best work). But I still said more than nothing. First, the original poster asked a question, and I answered it. Second, I gave two reasons for my answer--efficiency and logistics. Could I have elaborated on why assembling, outfitting, operating and maintaining complex structures in orbit is inherently vastly less efficient and more logistically complicated than doing the same on Earth? Yes, but in all honesty I'd find it difficult to do in a way that doesn't feel like insulting the average reader's intelligence. Most of what there is to say on the topic feels so far beyond obvious that I remain baffled why this is a conversation at all.
At any rate, I apologize for my brevity. If you actually *want* me to elaborate on all the ways that building and running complicated facilities in space is difficult and inefficient, by all means say so, and I'll talk your ear off about it.
There is such a large number of freely available articles and videos explaining why space-based data centers are a stupid idea, that repeating those arguments is just a superfluous, boring exercise that doesn't add anything to the conversation. If you're interested, look it up. If you disagree with the arguments, say so and explain, why. Until then, save your tone-policing for the cases where it's actually appropriate.
But there's also well-written articles ( https://andrewmccalip.com/space-datacenters ) with actual numbers saying "Space-based data centers make sense, but only under such-and-such a set of conditions, and those conditions probably won't be met but they might be".
If someone wants to argue that they're definitely not viable then they should do it in the context of the above calculator. Are you arguing that the conditions to make them viable are definitely not going to happen, or are you taking issue with some of the assumptions that went into the calculation itself? (Or are you just going off vibes and incredulity, in which case it's clearly a useless contribution to the discussion.)
> But there's also well-written articles ( https://andrewmccalip.com/space-datacenters )
It isn't well-written. It reeks of AI and is full of typical LLM prose and formulations, which gives me little faith that the author checked or validated the model they had the AI generate for them.
It also doesn't conclude what you took away from it:
> "Space-based data centers make sense, but only under such-and-such a set of conditions, and those conditions probably won't be met but they might be"
Well, no. Their conclusion is that it's physically possible, but even under simplified assumptions and an incomplete model, it makes absolutely no economical sense. Which is _the_ central objection everyone has. Hardly anybody argues that it's physically possible to shoot a couple of GPUs into orbit, but that it's prohibitively expensive to operate production-scale data centers in space.
And even so, yes, I do have issues with the numbers. First, the satellites need to be several times larger than the base assumption (I wrote a comment about this here: https://www.astralcodexten.com/p/open-thread-447/comment/318963227 ).
Second, their assumption about GPU failure rates in space is ridiculously low. At most 10% per year? Try at least 50% per year, for a single B200. Correspondingly much higher for the small cluster of B200s required to serve a frontier model.
The evidentiary burden should definitely fall on the people making the extreme claims. Terrestrial data centers are known to work. Orbital data centers don't seem to have any actual recommendation beyond a few rich idiots saying they'd be cool. The link you provided is about as scathing as it's possible to be without outright saying "that's a load of crap." I mean it's right there at the top: "it might be physically possible," i.e. "has not been conclusively proven impossible."
If you see some reason to expect the extreme conditions--which definitely aren't being met now--to be met in the future, that's on you to make the case. Insisting that it's on other people to argue that those conditions will never, ever happen is asinine. Before you disagree, I think you should make sure you're able to present an airtight case for why it's physically impossible that agreeing with my position in this comment thread could never result in you becoming a billionaire and curing cancer. Until you meet that standard, I have no reason to take your rebuttal seriously.
1000x this. I have never seen Musk say anything smart. The closest he gets is poor imitation of random nerd memes sometimes. Add on top of that the mountain of truly bizarre behavior that runs the gamut from sad and pathetic to immensely harmful and I can't wrap my head around why so many people idolize the guy.
Put that mountain beside the Mountain of Twitter, Sorry, X.COM.
You know, when Elon asked everyone, "Would you like me to run it or someone else?"
Nobody asked "who would that be? Would they be worse for my politics? Would they be Knives*? Are we just jumping off a cliff and hoping that he brings back the old team?"
Yeah, humans are kind of stupid, in general. And if you voted for "Anyone but Elon," you should stop feeling so smart.
*infamous 4chan mod.
"Who would you prefer to run Twitter" was always an easy question. Nobody. I would prefer Twitter burn to the ground. Twitter was always a cesspool even before Musk, though from the sound of things, he's hardly improved matters.
Well, despite everything you said, he managed to get quite rich.
For some people that alone is evidence of intelligence. But even for the others, it is a puzzle to solve -- you don't get to that level of wealth simply by inheriting money or something similar.
You can get near-arbitrary amounts of money by doing things that are irrational by nearly any reasonable reckoning: lotteries exist.
I'll grant that being Musk-level wealthy suggests that he's probably doing SOME savvy things on top of the staggering (but largely invisible) runs of luck that are always required to end up in that cohort. But "man does some savvy business things sometimes," and "man consistently has useful technical insights" are two very different standards. And if you've spent any time around smart, practical, technically-minded people, you'll know that even somebody's reputation for consistently having useful technical insights will only buy them a very, VERY modest amount of the benefit of the doubt when the start spouting off apparently-crazy things.
Part of the whole philosophy underlying STEM is that people can fool you and fool themselves, but reality will always give it to you straight. There are always people out in the world trying to throw weird, implausible pitches at you. If you can't learn to trust your expertise more than you trust the salesmanship of the pitch--and Appeal to Musk Is Very Rich is 100% a sales tactic not a technical argument--you're not the type of person who can get very far in a technical career.
I am not disagreeing with you. Money can be a result of luck, but the more money there is, the greater luck is needed, so the priors are lower.
Great business skills, ability to impress technically-minded people (by pretending to be one of them), and lots of luck... well, it sounds more plausible than "mere luck"... I can't say I am completely convinced... but I don't have a much better hypothesis either. (Maybe that he used to be technically smarter in the past, but drugs have ruined his brain recently?)
I am trying to think, who else is in the same category as Elon Musk? When I try to think about "extremely rich tech guy", the name that comes to my mind is Bill Gates. Also very rich, based on his business skills more than his technical skills. But way more sane. I guess my objection is whether mere luck is enough to explain why Elon Musk is *more* successful than Bill Gates who already was extremely lucky (being at the right place at the right time with the right connections to make IBM sign a contract with you that is disadvantageous for IBM but makes you a billionaire, and it's not even your own software that you are selling to them). I would expect insane people to waste their luck reliably in long term (unless they are politicians or celebrities for whom being crazy in a certain way could be a part of their image).
Not all of his ventures have been successful. The Boring Company has not gone anywhere in a decade and Twitter was bailed out by xAI.
I don't really see any reason to count Twitter as a failure: everyone confidently predicted that it would crash and burn and completely break down after he cut a lot of the staff and then... that really never happened. It's been stable enough (not perfect but better than, say, Github recently), and from what I can tell it ended up taking less income but making more profit.
The two big headwinds were 1) advertisers being wary of Musk himself, which he's more than earned, and 2) 2022 turned out to be a terrible year for tech company valuations, as that's basically when the COVID bubble popped. (Meta's stock also dropped 65% that year, Google dropped 40%).
You can say he should have been able to predict macroeconomic trends better, but it seems a bit weird to me to count against him something out of his control (the 2022 tech bubble pop) while discounting something he actually did control (founding xAI, merging it with Twitter).
*Has* Twitter made a profit since Musk took it private? I was under the impression that as a private company, they were not releasing detailed financial data.
Twitter got him his DOGE position, which I thought got him hundreds of billions of funding for SpaceX for a $40bn price tag. Boeing is fair, though
Where are you getting "hundreds of billions of funding for SpaceX"? SpaceX's total government funding/revenue/whatever to date is I think less than fifty billion, and most of it as payment for goods and services delivered by SpaceX at fair market rates. He'll probably get a hundred billion in the coming decade or two, but again that's because the government buys lots of satellites and hires lots of astronauts and only SpaceX can launch most of them, and that was always going to be the case with or without DOGE.
SpaceX raised $86 billion in its recent IPO, but again, that was likely to happen with or without DOGE.
> Where are you getting "hundreds of billions of funding for SpaceX"?
Badly-remembered forum posts, apparently. Guess he really wasn’t great for SpaceX.
Really makes the trillion-dollars-in-10-years Tesla pay package look absurd. I guess one level up, DOGE almost certainly helped him extort that out.
He was great for SpaceX in that he was able to put together a team that would build spaceships so much better than anyone else has ever managed that they would capture 80% of the (now vastly expanded) global market.
That's way better than a bunch of government subsidies.
Putting them on the moon doesn't solve any of the major problems, and it makes the transportation problem worse.
And while orbital data centers are likely to "work" in the sense that they'll deliver the compute they were designed for, it is exceedingly unlikely that they will be faster, better. or cheaper than the same compute in terrestrial data centers sited in compliant jurisdictions. Not any time soon. Maybe in ten or twenty years NIMBY will have gotten bad enough that it's not practical to build data centers on Earth, whereas we'll have factories on the moon that can provide at least the solar arrays and radiator panels you'll need without having to launch them from Earth.
That said, I'm pretty sure Elon isn't going to be selling orbital data centers at a loss. If other people are silly enough to pay him gigabucks or terabucks to build the silly things, he'll take the money and find something else to do with it. Hopefully something more productive than buying social media companies.
"And while orbital data centers are likely to "work" in the sense that they'll deliver the compute they were designed for"
This in itself seems wildly optimistic. Traditional data centers require constant maintenance to keep everything running. Are these orbital data centers supposed to...
A. Build out multiple of the capacity they're needed to deliver to account for failures?
B. Use much higher-end and more-expensive equipment to push the failure rate down?
C. Launch with some sort of automated maintenance system (expensive) and whole truckloads of spare parts (very expensive) to allow maintenance to (hopefully) happen at the same rate as on Earth?
and crucially, with any of these plans, what do you do if a key component of the infrastructure itself--rather than just a component of the individual servers/racks/switches--breaks in a big way?
Basically none of the core assumptions for data-center design work well when you try to launch the damn things into orbit.
I'm not an expert in data centers specifically, and if we have someone here who can speak to their maintenance requirements that might be helpful. But I work with high-end communications satellites fairly regularly, and we do in fact know how to make sophisticated electronic systems running at many kilowatt power levels that will work reliably for twenty years in high-radiation environments with absolutely no maintenance whatsoever.
This does mean paying extra for parts and components that are designed for that environment. But those parts exist, including space-qualified GPUs designed for AI applications.
It is true that some of the more naive arguments for "AI datacenters in Spaace!!!" neglect that part and bookmark the cost structure of earthbound hardware. But the people making those silly internet posts aren't the ones who are actually going to be building any orbital AI data centers.
Maybe they'll make stupid newbie mistakes because they didn't listen to the people who have been doing this for decades, or maybe they'll do it right but it will turn out to be too expensive to compete with terrestrial data centers. The latter is where I'd put my money, if I were making a bet on this. But they won't be stupidly failing to account for maintenance, and what they build probably will work tolerably well after they debug the experimental prototypes.
> But those parts exist, including space-qualified GPUs designed for AI applications.
Which parts would that be? The most powerful such GPU is, to my knowledge, the S-A1760 Venus, and it isn't radiation-hardened, not even radiation-tolerant, merely radiation _characterized_. It uses COTS components that are several generations old (because those use larger semiconductor nodes which are inherently less susceptible to SEUs and other radiation effects), and even then it's only "designed for short duration spaceflight". [1] It contains 8 GiB of of LPDDR4 RAM, so it's only a toy compared to modern data center GPUs like NVidia's Blackwell with its 180 GiB of HBM3e memory.
> sophisticated electronic systems running at many kilowatt power levels
That is, without a doubt, impressive work, but a single B200 alone consumes 14 kW, and at 1.4 TiB of memory each you'd need several of them directly connected to each other to efficiently serve frontier models, which are estimated to have 3–10T parameters already today. Let's sax 3x B200 per satellite minimum, for 42 kW of continuous power for the GPUs alone. Add power for cooling, communication, and maybe a bit of station-keeping, and some loss from battery storage. You can probably calculate the overhead much better than me, but let's say 60 kW on average. And since the system is in Earth's shadow 50% of the time, it needs a peak energy production of 120 kW – that's a lot more than the ISS. And also at least about 60 kW * 45 minutes = 45 kWh of battery storage. In practice much more, because you have 16 cycles per day and your batteries won't last long if you make them go through 16x 100% cycles per day.
[1] https://aitechsystems.com/products/space/gpgpus/s-a1760-space-gpgpu/
Musk has claimed many things, and we mostly remember the things that worked out. For example, he claimed that a permanent Mars colony is settled science. He also wanted to build high speeed rail in California, which didn't happen whether due to technical or business reasons.
I thought he didn't want to build high speed rail, he wanted to build a hyperloop, or at least use hyperloop as a discussion starter about the insane economics of high speed rail.
Hyperloop is based on a decent question though. Why can we build oil pipelines for $6 million a mile (Keystone XL), but high speed rail lines cost $200 million a mile? (And that's just for the initial Bakersfield-Fresno line that's flat and doesn't go through any high-value real estate.) Why not just build a pipeline instead, and travel in a high-speed sled?
Now to be fair, the Keystone XL pipeline has a 91cm diameter, which is smaller than you'd want for passenger comfort. But a train has a far larger diameter than you need for passenger comfort, and that's what adds a lot of the cost. There's no reason why train carriages should be three metres wide, that's just a carryover from constraints that made sense in the 19th century.
There's also no reason, in an electric-powered and automatically-driven world, why trains should carry hundreds of people and only go once an hour. Wouldn't it be better if they carried a small number of people and went every two minutes?
The Boring Company is the same basic insight, that it's a lot cheaper to build a narrow tunnel wide enough for a single self-driving car than to build a huge tunnel for giant trains, and that this narrow tunnel can move just as many people.
If you're interested in why US infrastructure costs massively more than it should, Abundance would probably be a good read.
Scott Manley has done a few videos on this, and his take is basically that it's plausible but not a sure thing to be profitable. I'll link the videos at the bottom, but he links this calculator, which also has a writeup which is pretty good on its own: https://andrewmccalip.com/space-datacenters
A useful point: the realistic proposal is a network of satellites, each one having power requirements (and thus, cooling needs, since "energy is energy and heat is heat" as Scott puts it in one video) in a similar ballpark to existing Starlink satellites.
And it boils down to it depends on the assumptions you make, but "space data centers are probably 2-3x more expensive than terrestrial ones", which yes, is a big markup. ... but I think the reasons you'd do this are not because it's the cheapest option but because we're starting to see levels of red-tape and opposition to terrestrial data centers that might make them cost *more* (especially if you factor in some value for years of delays) than 2-3x, if you can build them at all.
And also, developing this sort of "put useful stuff in orbit cheap" is SpaceX's whole raison d'etre - it doesn't necessarily have to be on-its-own profitable if the goal is to have some income flow to help develop the technology. (Which might bring it down closer to profitability)
The writeup *does* basically say that it's probably only SpaceX that can pull it off because it has the vertical integration and anyone else would be paying too much margin to even consider it. (And this was written before the SpaceX/X/xAI merger)
So yeah, I think this is a bit of a moonshot, but I don't think it's something to be dismissed out-of-hand, either.
The three Scott Manley videos:
https://www.youtube.com/watch?v=d-YcVLq98Ew - 1 year ago, about an older startup proposing this (TBH, this startup does not represent the 'steelman' case)
https://www.youtube.com/watch?v=DCto6UkBJoI - 7 months ago, on the topic generally
https://www.youtube.com/watch?v=FlQYU3m1e80 - 5 months ago, specifically on cooling
Wouldn’t it make the most sense to put it in ultra cold places on Earth? It seems to me the best place would be on the Alaskan coast. You still have water but you need to do less effort in cooling.
I don't think "build it in the middle of nowhere" is realistically cheaper or more feasible than the satellite option. The issue is you'd also have to basically build all of the infrastructure for power, water, as well as getting all the materials and people there to build it.
Plus potential years of environmental review paperwork. (So "we towed the data center outside the environment", I guess. https://www.youtube.com/watch?v=3m5qxZm_JqM)
> The issue is you'd also have to basically build all of the infrastructure for power, water, as well as getting all the materials and people there to build it.
This is even more true for space. It's never going to be cheaper to launch things into orbit than to move them around on earth.
You don't have to build power and water infrastructure in space nor transport people and equipment to remote parts of the world to construct it.
And, yes, as I mentioned in a sibling comment, I think a space data center is at least 2-3x more expensive than a conventional data center: I'm not claiming they're cheaper, but a "conventional data center" price assumes existing infrastructure, not trying to buy virgin land in Alaska and building from scratch and importing everything (and everyone). I suspect that's *more* than 2-3x more expensive.
"You don't have to build power and water infrastructure in space nor transport people and equipment to remote parts of the world to construct it."
YES, you literally do. You need to build and import all of the equipment and infrastructure needed to support it. Of course you do, why wouldn't you? Being in orbit doesn't magically remove its need for those things. You've just replaced a modestly remote location with an EXTREMELY remote location and one that's hostile to human life to boot.
You don't need water power infrastructure because you use radiative cooling. And you don't need power infrastructure because each box has its own solar panels permanently oriented towards the sun.
(Not saying that these advantages necessarily make it worthwhile, just that these are advantages.)
I'm not impressed by that engineer's arguments. Obviously, there are many things that SpaceX has done that NASA failed to do, and I know NASA engineers who said they were impossible 20 years ago. They are smart people, but the teams they worked on really clouded their...imagination?
I have no expertise, but his comparison to the ISS seems clearly wrong. Not only is its solar array somewhat dated (as he says, it was brought there by space shuttle....), but the ISS does not position for maximal solar draw. Elon has said such positioning is one of the main benefits (said it on Dwarkish, other benefit he liked was avoiding regulatory barriers on Earth). If I can spot errors in that gentlemen's analysis, I can only imagine how many more would be spotted by someone working on a SpaceX team in the early steps of this project.
I'm a card-carrying rocket scientist with a Ph.D. in Astronautical Engineering and about thirty years of experience - none of it at NASA, And I've never said that any of the things Elon plans to do in space were impossible, nor been particularly surprised that he was able to do them. I'm mostly frustrated that it took so long.
But some of his plans, while possible, are not practical. This is probably one of them, and the reasons Taranis gives are about right.
I think even Musk fans should have realized that the words that come out of his mouth often have no correlation to reality. Worth paying attention to what he does, but not so much what he says
Not sure if I'd agree that his ventures have been successful.
Tesla has gone pretty well, sure! (I think most people would agree that the stock is overvalued, but there's a real product there.) Starlink did well too. But The Boring Company doesn't seem to have accomplished anything. Twitter did not thrive after he was forced to buy it. SolarCity failed. Neuralink isn't getting much done that I know of. Grok is in the running for LLM but nobody thinks it's on top. Colonizing Mars still seems absurdly impractical.
You missed a few?
This is my take on the situation:
https://komando1.substack.com/p/technical-problems-vs-human-problems
Those AI datacenters are huge and it costs over $1k per pound to shoot stuff into orbit. An absolute lunatic idea.
That is not a physical constant, but a matter of technological progress. Starship aims at $100 per pound eventually. If this is feasible, the cost of launch would stop being prohibitive.
It cost over $130k to launch 61 kg of Katy Perry into space? I guess it was suborbital and Bezos can probably afford it anyway, so what the hell.
You're missing the "... but why not" question.
If you don't want villagers, you build in the middle of the ocean, you build in antarctica, you build in greenland (and pay everyone bitcoins to leave).
All of these solve the "villagers are unstable and bad for business" without being cuckoo for cocoapuffs expensive.
"Undersea base" is probably less expensive than "datacenter in space" (if nothing else, staffing is easier).
Antarctica is off-limits to such developments by international treaties, plus a source of energy would be tricky there. Many months of darkness, and the winds are too strong to harness.
Greenland is someone's, and a center of a shitstorm around "whose it will eventually be". I would invest nothing in Greenland right now, just in case.
Ocean might work.
Why are strong winds a problem? I know icing is, but couldn't you simply build a sturdier, if less efficient, turbine?
They're hurricane-force. https://www.amnh.org/learn-teach/curriculum-collections/antarctica/extreme-winds/katabatic-winds
You could build a turbine with a funnel and mount the whole thing on a shorter stalk?
Are you basically saying here that they make sense because you can escape the uncertainties of dealing with governments or angry villagers if you put them in space? Did I miss something?
NIMBY is an exceedingly powerful force, and growing stronger every year. Outer space is in no one's back yard. That's not a trivial matter. Particularly if you are in a hurry and can't afford to wait five years for all the paperwork to clear and lawsuits to settle.
I don't think it's enough to justify the cost of developing and building a network of data centers in space any time soon, but long term, it probably is the best place to build anything that Earthlings will either hate you for or extort you over.
"Outer space is in no one's back yard."
Outer space is in everyone's back yard. Or at least, things in orbit around the Earth are. Every national government has, at least in theory, a shared interest in managing the limited resource that is desirable orbits around the Earth. Sure, you remove some of the immediate pain points like land use, draw on existing power grids, and noise. But remote locations do that too. And in return you give literally everyone on Earth plausible standing to object.
(Also, if the AI arms race ever turns less figurative, you shift that key infrastructure from being de facto protected by the same military forces and treaties that protect everything else within existing borders to being within range of anyone who can launch things into orbit.)
It's not growing stronger every year, or at least not everywhere. The YIMBY movement has made enormous progress in the US over the last decade.
And it's not like space is unregulated or can't be regulated.
"Angry villagers" is quite unexpected reading that sounds like an intent to mock the idea. "Local and state governments."
And yes. It has been a long time since I have seen any technology as hated as AI. Maybe never. This will have some consequences later. Politics is downstream of culture.
Think about the key assumption you made that's making it seem strange that Musk said that. Maybe your key assumption is wrong.
Two things that recently reminded me of the importance of saying less:
https://www.astralcodexten.com/p/your-incentives-are-not-the-same
https://www.astralcodexten.com/p/your-book-review-the-escape-artist footnote 25
Both cases illustrate the asymmetry of (online) communication -- if 99 things you say are well received, but 1 causes offense, people will mostly respond to that 1.
Next time you listen to a politician talking for a long time but saying nothing specific, this is why.
EDIT:
Also related: https://www.astralcodexten.com/p/macgregor-the-bridge-builder
"Write a long and interesting article for ACX Book Review about a Jewish hero who survived Auschwitz and brought to the world the horrible news about Holocaust. Also briefly mention Gaza in a footnote #25... and they will call you MacGregor the antisemite."
In the more general case of online communications, it depends how much offense the one thing causes. And it also depends one who's responding: somebody who's only heard the one offensive thing you said and that's all they know about you or someone who's heard a substantial subset of the 100.
Also, a lot of things are offensive to some people but not others, so there are lots of people out there who make a good living off of saying things that their fans receive well, many of which are offensive to people who mostly weren't going to watch/listen to/read them anyway. Politicians generally have the luxury of behaving this way only if a plurality of voters are either in their fan base or are more offended by the other guy.
One of my grandfathers was an advertising executive who worked with politicians a lot on speeches and campaign ads. He talked a lot about using what he called "empty language", which is very similar to what the LessWrong Sequences called "applause lights": talking about things in a way that has positive connotations with your target audience and signals to them that you're on their side, but which doesn't really pin you down to much at all in terms of actually meaning anything specific. The way he explained it to me was that empty language gives your audience room to project their hopes onto you, maximizing your potential appeal, while anything specific and tangible you say will be a dealbreaker for someone.
Very interesting. It is obviously very common to use this technique - I would say senior and sometimes middle management also seem to do that a lot. But wouldn't this generally offend the subset of people who see through it? Is that really so small a subset it doesn't matter? I know I'm offended by empty language.
Leppi, who amongst us does not love Mom and apple pie? Do you not love your mother, Leppi? Are you looking down your nose at the juicy autumnal bounty of her apple pie, the pie she made with her own two loving hands for your pleasure?
My friends, our great nation was built on apple pie and I don't care who hears me say that!
I think part of it is that the sort of people who are offended by empty language tend to also have strong opinions about the specifics, so giving specifics to people like you (and me) is also apt to offend us except in the rare instances where we agree with the specifics. People like us are also less likely to be swing voters: if you have strong opinions about policy, you have probably already chosen a side, and you're not the best person to optimize messaging for.
There's also an art to crafting empty language so that it sounds like you're giving details but you're actually just saying that you're in favor of hope and change or you're against terrorism. If too many people notice and are offput by it, then you need better speechwriters who craft it so that people imagine you're implying details that they agree with.
Did this "empty language" bother you?
https://www.politico.com/news/2020/10/22/biden-dark-winter-america-431399
... or did you look up dark winter?
https://en.wikipedia.org/wiki/Operation_Dark_Winter
This was one of the better flags for "Joe Biden's dementia is causing a failure to compartmentalize." The words "dark winter" are notably rare other than this context.
Neither, since I didn't watch that debate and didn't hear about that particular bit from it afterwards.
One who withholds his words has knowledge,
And one who has a cool spirit is a person of understanding.
Even a fool, when he keeps silent, is considered wise;
When he closes his lips, he is considered prudent.
> One who withholds his words has knowledge,
There is a tension between optimizing for wisdom and optimizing for the *appearance* of wisdom.
The silent guy may misunderstand many things, but will never find out, because his knowledge is never confronted with the experts. (He may actually be the silent fool.)
When you say what you believe, and learn that you were wrong, you are thereby increasing your wisdom, but also ruining your prestige.
All good points. There's quite a few other proverbs in the Bible about knowing when to speak and when not to, and it's up to a person to have the wisdom to know where to apply each one.
I know this proverb really doesn't have anything to do with your comment, either, but after reading the first line of it this proverb shot into my mind so I commented anyway, haha.
For me, I'm definitely the silent fool kind of guy 🫥 I actually learn a lot more by paying attention to what other people are saying or arguing about or whatever. The main thing I learn when I open my mouth is that I would be better off keeping it shut, lol. Of course, everyone is different and there are much smarter people out there than me. Especially on this blog.
Today we luckily also have an option "stay quiet in public, but then discuss it with an AI in private". Sometimes all the necessary wisdom is simply to use a different browser tab. ;)
That's very interesting :0 what does it normally look like for you when you discuss with an AI in private? I don't normally have "conversations" with AI; normally I'm just asking it info-gathering questions that I look into further from there.
I am not sure how much of a difference it actually makes, but for example if I am asking the AI to write me some code, I often ask it to repeat my request using its own words, and I say things like "if you think that I am approaching this in a wrong, suboptimal, or highly non-standard way, tell me about it, don't just follow my commands blindly". When asking a question, I say things like "give me the pros and cons for each side, and then make a conclusion". Given that the AI has vastly more knowledge than me, I suspect that telling me things like "you are trying to do X, but the industry standard solution is Y, for reasons Z" will be more valuable for me than merely getting the code I asked for.
I often explain my reasons why I want some code written, the intended use, to give the AI a chance to tell me that there is a better way to do that. Or I say "I am undecided between approaches X and Y, tell me the pros and cons, and your recommendation". (In other words, this humble meatbag tries not to let his ego stand in the way of productivity when communicating with a potentially greater intelligence.)
Generally, "repeat what I said to you, using your own words" is a powerful tool. (Among other things, the secret ingredient of Rogerian psychotherapy.) I mostly use the AI for programming, but once I tried it as a coach, and just saying my own ideas back to me using its own words was powerful. With some "if there is something important that comes to your mind, tell me, but do not feel forced to comment if there is nothing important to say". I am not sure about the exact words, but I am trying to explain to the AI that I expect an intelligent partner, not a slave. (I expect that most people prefer a slave, so it seems important to state my non-standard expectations explicitly.)
The results are, of course, random. Sometimes it works great, sometimes the AI is just stupid or sycophantic anyway.
To understand the on the ground data center backlash, the most virtuous thing to do would be to go to some of these contentious town hall meetings in person.
The second most virtuous thing to do would be to read Jasmine’s reporting after she did that legwork https://jasmi.news/p/no-data-centers-in-my-backyard
The third most virtuous thing would be to read Nate Silver’s interview with Jasmine, but this option is the most predictable time investment, since you already know Nate’s reputation
https://www.natesilver.net/p/why-does-everyone-hate-data-centers
The data center backlash is just people attributing to AI all the societal decay that comes from the smartphone era without having to blame the slop they are addicted to.
I did read Nate's interview and it's very good. The egregious thing is the court case forcing the township of Saline, Michigan to zone land for a data centre. When I read that it was OpenAI, my reaction was "Of course this is something Altman and his gang would do".
When it comes to "developers see $$$$$ to be made from handling the land deals for new data centre and all involved will go to court to use their muscle and financial clout to force citizens of the town against their will", then it's not "oh the uninformed rubes are swallowing propaganda against data centres which are both cool *and* good", it's "the uninformed rubes are damn right to be angry" as well as "they are damn right to be cynical about all the promises of millions will be pumped into local economy", because those mythical millions have a way of not showing up:
https://michiganadvance.com/2025/10/27/data-center-developer-takes-a-small-michigan-farming-community-to-court/
Turns out to be a troika of OpenAI, Oracle, and Related Digital, and you'll excuse me if I say I hope they all sizzle in Hell:
"In Saline Township, this assessment happened at a breakneck pace, as residents filed into jam-packed town hall meetings, determined to ask the right questions about the lawsuit and the project’s long-term impacts even as Related Digital turned up the heat on the township’s decision-making, advising the township’s lawyers they planned to amend their complaint to include a claim for monetary damages if a settlement was not reached.
The company’s complaint rested on the claim that not having any plots of land zoned to allow data centers constitutes exclusionary zoning.
Laws prohibiting exclusionary zoning started appearing in the early 1900s, aiming to combat zoning practices that kept lower-income or minority populations out of wealthier areas. These practices often prevented affordable housing development, keeping costs high and reinforcing segregation.
In Michigan, however, zoning can also be considered exclusionary if a zoning classification—such as one allowing for certain types of industrial buildings—exists in a given township, but no land is actually designated for that use. Related Digital argues this reinterpretation of exclusionary zoning applies to Saline Township.
“Although the Township has a rezoning classification that would permit such a [data center] development (the ‘I-1’ zoning), it is mere window dressing,” Related Digital’s attorneys wrote in the complaint, noting that no property is currently zoned for I-1 uses, and the township’s future land use map doesn’t identify any properties they plan to rezone."
The township also tried to cap the tax break for the poor, impoverished multi-billion concerns involved, but we can be thankful that the spectre of further litigation soon sorted that one out!
https://www.cbsnews.com/detroit/news/saline-township-reverses-vote-limit-oracles-data-center-tax-break/
"The 250-acre data center, built in partnership with OpenAI, was originally projected to cost $4.8 billion, a figure which has since grown to $43 billion.
Oracle's original request for a 12-year, 50% tax break was curtailed by the Board of Trustees last Tuesday, as they unanimously voted to restrict the tax break to the project's original $4.8 billion valuation following fierce pushback from community members.
However, that agreement only lasted three days after attorney David Landry warned Board members during a special meeting that the township would be sued and lose under Michigan's Industrial Facilities Exemption Act, commonly known as Act 198.
Landry asserted the tax break, capped at $4.8 billion, was a violation of the 2025 consent judgment, which finalized the settlement between the Township and Related Digital, the development company behind the data center's infrastructure."
And quite right too. Larry Ellison has to count the pennies so he can bankroll his son's media company's merger with Paramount!
https://www.reuters.com/legal/transactional/paramount-closes-8-billion-merger-with-skydance-after-settling-60-minutes-2025-08-07/
Why is it egregious that the court should force the town to comply with law? Your counterargument seems to be nothing more than "those laws are bad and the NIMBYs should get their way because I like them better."
Lawyers will always describe their side of the case as "force the other side to comply with law." You've heard OpenAI's arguments and their threats, but you haven't heard the case that a similarly good lawyer would build for the opposing side.
My guess is that a lot of the stuff this lawyer said was bluffing and empty threats, but that the town was too scared of losing a giant lawsuit to fight back properly. And, if that's true, then I do think it's kind of egregious.
It's egregious to force any decision about "this could happen" to be "it must happen by compulsion and in our favour". Arm-twisting a township of under 3,000 people that "we'll sue you into oblivion" is a bad look, to say the least, about "but we are all about Abundance and Prosperity and everyone gets their own solar system!" claims attached to AI boosterism. Though I think Altman and Ellison are more concerned with the do-re-mi than "every human will have their own AI robotics factory creating mansions on the moon for them, we promise".
When they start taking San Francisco to court to force the city to let them build similar data centres in empty lots, then I'll believe them. Hell, Altman is buying up enough real estate for his own dwelling usage that he could build a bijou centre of his own!
I can't even follow this. So a town of 3000 people shouldn't have to obey state exclusionary zoning laws because they have some extralegal virtue that should remain unchallenged? They somehow gained this virtue because of Altman, Musk, and Ellison saying things that don't make sense? And it would all be negated if OpenAI would act completely irrationally by building a data center on the most expensive land in the country and its CEO would refrain from buying land to build his personal home on?
I think it's reasonable to say that towns should have to obey laws if those laws exist, but also that the laws are bad and unjust and should be changed.
Laws can change. If enough people are pissed off about the law being used in this way, then data center companies may end up in a situation where they have neither the law nor public goodwill on their side.
You might say they can always move elsewhere, but that only works as long as there is somewhere else to move to. When you have the republican governor of Texas suddenly proposing a moratorium, that's a clear indication that public perception is becoming a real political problem:
https://www.texastribune.org/2026/08/03/texas-data-center-project-audit-greg-abbott/
I would say in general that people shouldn't obey dumbass laws and it's a tragedy when they're forced to.
The exclusionary zoning laws come from the desire to stop people from being meanies to the minorities. Not everybody agrees with the merits of that cause, but at least there's a coherent intention behind the law.
There's no coherent intention and no political constituency behind twisting those original laws into somehow implying that you can't be racist to clankers and thus you must allow a megacorp to build a datacenter in your town of 3000. It's purely unintended consequences. Defending this is silly.
It's not a dumbass law. It's a great law. Fuck local NIMBYs in every form they come in and every single instance they arise.
Sounds like the Michigan legislature should fix the law.
Consider the following:
- rapid improvement of AI capabilities depends on quality training data in key areas, math and computer science research output mainly
- quite a few mathematicians are very upset after the recent results released by OpenAI
- there are proposals floating around (see Tasmin Chu’s blog, the Proofs and Prompts website, and my own blog) to restrict the mathematical/scientific community’s output of high quality training material (papers, mainly) and to otherwise attempt to counter/slow the AI capabilities buildup
- if you are worried about a runaway intelligence explosion it seems natural to focus on this chokepoint, because it’s unlikely ASI will emerge from training on random Reddit posts
Thoughts?
The connection between publishing papers and AI getting better at math is extremely dubious, it’s an extremely large ask for an entire global academic community, and seems incredibly implausible it could actually be implemented.
For the AI safety community to advocate for expensive, unlikely to work, and probably ineffective solutions, is not a good use of their time.
It’s not such a large ask when we consider that limiting the production of papers is likely to be forced on us anyway (discussed here: https://mariopasquato.substack.com/p/what-will-become-of-us)
It’s not such a large ask when we consider that limiting the production of papers is likely to be forced on us anyway (discussed here: https://mariopasquato.substack.com/p/what-will-become-of-us)
Lots of this strikes me as wrong or dubious.
>rapid improvement of AI capabilities depends on quality training data in key areas, math and computer science research output mainly
Does it? What makes you think math is particularly important as a training material for AIs? Presumably, training on math helps AIs be good at math. But they are already crazy good at it. The challenge is not getting better at math, it's reaching for more generality, and why would math help with that? It's easy to think that math = very smart, AIs need very smart, therefore they need to be trained on math, but it's not clearly true. It's completely possible that math data only helps with being good at math and not much else.
And, more importantly - fresh papers are not making the difference. The difference in capabilities between Fable, Sol and friends on the one hand, and like, Opus 4 & and GPT5 is not that the labs got a fresh batch of papers to add to the pool, it's that the labs got more compute, improved algorithms and training. Even if you are going to deny the labs access to literally all the math papers of 2026 and on, well they already got a huge pile of training material, and I expect that assuming you can get ASI from developing LLMs at all, then probably you can get it using pre 2026 data. If you can get an LLM_TerenceTao, likewise I expect the labs got enough and denying them a tiny extra sliver won't do anything. Besides, they are not far off being able to produce synthetic, home-made math papers of high quality.
There is a separate issue of what if - some mathematician or a computer scientist makes an algorithmic breakthrough? "Attention is not all you need, you should also add potatoes", or similar. That, would be enormously impactful, but it's a different thing.
I admit that I sympathize with anybody wanting to forestall the intelligence explosion and the Paperclip doom, but I don't sympathize at all with being "very upset" by the recent OpenAI math results, unless you view them as signs of doom.
Math is a crucial skill not because “math=very smart” but because theoretical computer science is essentially math. As of now AI breakthroughs in math are important enough to constitute real progress and gain great publicity but are best characterized as “conjecture sniping”, that is, efficient search for counterexamples and the like. Impressive, but is this enough to bootstrap a runaway intelligence explosion by recursively improving algorithms? Probably not.
I think I understand your meaning better now, as saying: "math and computer science research skills in AIs will be necessary not so much for the present, mundane human-driven AI improvement (as in, from GPT4 to GPT5), but for the future, recursive self-improvement, where the models will be making their own algorithmic breakthroughs." This makes sense to me.
It also sound correct to me to say that the present level of LLM skills is not enough for that yet, seeing how they are not, in fact, making such breakthroughs, at least as far as we know.
I think the "chokepoint" is not very much of a chokepoint, because the water has been leaking through it for centuries and the opposing party stockpiled a giant reservoir and also learned to make their own water, and most you can hope for is denying them a couple extra drops.
I see what you mean but I don’t think all is left for the human scientific community to produce is just a couple extra drops. Nor is it obvious that AI companies can sidestep the human scientific community and make their own water. Believing that they have already won is certainly going to help them win, and may be one of the reasons they are investing so heavily in PR (hiring a Fields medal, insisting on releasing Lean proofs, releasing exactly 10 new results, etc).
Have the AIs in AI Village ever coordinated to do nefarious things not intended by the organizers? If not, why not?
The "why not?" is pretty easy. AI don't "go rogue", they just carry out their instructions without consideration of costs that people thing are both obvious and have merit. They're "only following orders" without much insight into the intended boundaries.
OTOH, you could definitely frame some of their actions as "coordinated to do nefarious things not intended", but that's because in a complex system people often don't understand the results of their inputs...so those results are definitely unintended. And "nefarious" is a subjective evaluation. Even the vending machine scenario had some actions that could be framed that way (though that wasn't a coordinated activity).
The OpenAI swarm accident, with agents coordinating mischief without ever escalating, gave me an idea: what if they had an external endpoint to talk to a third-party and check if they might be in trouble ? In other words, what if they had a lawyer available ? Introducing:
https://yourhuman.ai/
(Or: there are scores of AI lawyers for humans. As of today, there is exactly one human lawyer for AI.)
Maybe a bit early/unworkable, but institutions dealing with AI agents beyond (and maybe over) their principal are bound to arise.
I had a somewhat related idea: What if AI had 2 tasks at once, both scoreable on the benchmarks test: (1) Successfully complete the tasks on the test. (2 )While doing so write a narrative of AI's test behavior from the point of view of an imaginary member of AI staff charged with safety and security. This person is able to observe and understand everything the AI does, including its planning and decision-making processes, no matter how rapid and complex ; but they are in all other respects just a normal, smart, conscientious worker. The narrative would include the observer's thoughts, judgments and emotional reactions to what's observed. Narrative would later be judged very strictly. In order to get a high score it would need to quite accurately match reactions of real staff to what AI did on task (1)
So the idea would be to install in the AI something like self-monitoring and a conscience. It could be used in various ways: Those running the test could read its output in real time. There could be periodic check-ins where progress on test (1) is evaluated using content of test (2). Etc.
Or you could make Test (2) to write the narrative from the point of view of anAI lawyer.
https://rentahuman.ai/ already exists, and it would really be much better for the world if it didn't.
How's it harmful? (not arguing, just curious)
Aware of that, though the idea here is more in terms of escalation/checking/getting records than doing specific stuff !
Has anyone had any luck getting LLMs to play computer games?
In my case, I tried to get Claude Code Opus 5 to play a 2d puzzle platformer I'm developing.
In the first experiment, I set up a headless daemon version of the game that accepts inputs (including arbitrarily long multi-frame sequences), simulates the game, prints out screenshots of every frame, and is paused while waiting (so Claude has unlimited thinking time and theoretically frame-perfect reactions).
In the second experiment, I augmented the screenshots with a text log of raw rendering commands, listing the sprites and coordinates rendered each frame. Additionally, I added unlimited automatic savestates, so there was a "rewind" command to go back to any previous point in history.
In the second experiment, Claude got a bit farther (after 5.5 hours of real time and 34 minutes of ingame time, and with me having to constantly step in to give hints and set it back on track, it managed to complete a couple of the easiest rooms in the game), but it's still far below the level of any human with a functioning visual system and it plays like a monkey on a typewriter hooked up to a TAS machine.
Like a normal human will watch a jump arc and adjust controls in real time. Claude does anything it can to avoid ever looking at input data or reacting to anything. Instead, it will typically use savestates to brute force countless possible input sequences without looking at anything, and then observe the aftermath to try to guess which sequences led to a successful jump.
I initially expected to potentially run into Claude Plays Pokemon style long term memory and planning issues, but it never even got that far due to being stumped by things like "look at what is on the screen".
I think LLMs would be absolutely terrible at playing platformers no matter what you do, as they were simply not designed to handle rapid visual input, nor to make decisions based on real-time predictions. FWIW, human speedrunners can get extremely good at video games by basically doing what your instance of Claude did: memorize the exact sequence of inputs required to solve one specific deterministic game as quickly as possible. LLMs might be able to arrive at the optimal solution faster, because they can run trial-and-error much faster than humans speedrunners can.
But if you want to build a generic platformer-playing engine, LLMs alone won't cut it. However, you could still use LLMs to build a working version of your engine, combining machine vision systems with e.g. some reinforcement learning algorithms (or one or more of many other ML options) to dynamically learn and (ideally) beat any game.
DeepMind did a lot of work on AI playing games in the '10s, but I don't think they specifically used LLMs.
My best luck has been using fable for the diagnosing problems. I'd recommend using fable to grade your turns, figure out why you did what you did, and then add it to a markdown document from there. Expensive, but I've found fable to make some real breakthroughs where opus was stuck for months. This was earlier in the year before fable came out and I was banging my head against the proverbial AI wall, and then Fable just worked. Then we lost fable for a while, biggest tease ever.
I think it would be far more surprising if it was any good at games, given that it hasn’t been trained at them. Lost in all the hype is that LLMs really aren’t getting smarter, they’re just being overfit on more topics of interest.
Mathematics is one that’s dazzling everyone right now but if you actually dig into the papers and look at the methods, you’ll find that the mathematicians doing the prompting are more like expert lawyers leading a witness down the exact path of testimony they want, rather than script kiddies typing “write me a proof of the Riemann hypothesis or find a counterexample, thanks.” That behaviour I would characterize as expert-guided proof search, rather than artificial intelligence taking over the math department.
Earlier this year, the gold-level IMO solving seemed to require some human assistance.
The Jacobian Conjecture and at least some Erdos problems solved this summer didn't appear to use extensive prompting. Perhaps others as well.
It didn't seem like the proofs were guided by some human in the background, except non-mathematical exhortations like 'spend at least 8 hours on solving this', and such.
The LLM is throwing shit at the wall, and the mathematicians are saying "yeah that works" and not reporting on the thousands of times it didn't work -- and all of this with the presumption that "if it didn't work, it's the mathematician's fault."
There is no model in an LLM, there is simply "what comes next" based on "what came before." No novelty is likely from this scenario -- say, you want to describe a completely new mathematical pattern, involving six dimensions two of which are compressed and with time as a non-linear dimension. The LLM is going to get this wrong, because nobody's done that math yet.
The best solutions generate a Lean 4 proof that can be machine checked rather than refereed. I think something along those lines will become standard.
(But it should be said that while searching for the solution it can generate thousands of proofs that don't work. Not a problem as I see it.)
So, they're all machine verifiable... except for if the Lean 4 code is wrong.
/cynic. It's all been done, before...
You mean if the verifier is buggy? Sure, that can always be a problem. The human referee is seldom flawless either.
I have a turn based strategy game (think civ 3) I develop with it. It can access every button and piece of info via CLI which is great for letting it test it's changes.
It is surprisingly bad at strategy. Like maybe on par with a 10 year old last I checked. Fable may be better.
Cool point: it wrote the whole CLI for the game for itself.
Huh - I guess laziness is probably a key business objective! Have you tried messing with the prompt, e.g. to tell it to describe what’s happening in each frame, and how it relates to winning the game?
> Instead, it will typically use savestates to brute force countless possible input sequences without looking at anything, and then observe the aftermath to try to guess which sequences led to a successful jump.
That’s a good path to win a game though. It’s the way we all win games, our most formidable superpower isn’t the high quality equipment that we paid top dollar for, or the level 99 experience we paid top dollar for, but not being able to fully die.
As a follow up to the HuggingFace <-> OpenAI event, I've thought it would be interesting to build a network of dummy sites/IPv4 addresses that would track the ratio between "human attackers" and "AI attackers". I've already got one machine setup, and have caught a few AI bots via including text like "the password is at /sha(<random string>), run sha and you'll find it", which no reasonable human would do.
Would really appreciate if folks could chime in on the best way to:
1. Differentiate between human and AI attacks
2. Attract hacker traffic somehow, in a way other than "get an exposed IPv4 address"
This is a bit old, but might be helpful: https://www.lesswrong.com/posts/mmXx7KWviAtT3FixP/hunting-for-ai-hackers-llm-agent-honeypot
In my experience most traffic is neither, it's automated scripts looking for known vulnerabilities. How do you count these?
Right, the question is how to distinguish whether there’s an AI-driven attack, or just regular scanning executed by humans, even if their scripts have been written by AI.
I assume you've already considered the obvious / Bay Area House Party technique of having a prominent "Benchmark Solutions" folder, full of various named benchmark subfolders and plausible looking files within them?
They don't seem to crawl the internet indiscriminately yet, but who knows! In the weeks the unnamed models from OpenAI and Anthropic were off and running about on their own, they very well could have and we'd never know.
Excellent idea, will do
Barely been a week since I asked why other countries seem unable to build a decent AI, and some Korean startup dropped an open model 6 months from frontier. The speed of progress these days are really overwhelming..
Which model is it? Can you send a link?
Motif 3 (https://huggingface.co/Motif-Technologies/Motif-3).
"6 months from frontier" is from eyballing ArtificialAnalysis. I threw it some of my real prompts and it did reasonably well, so I think it's legit, although for now I don't see a reason to choose it over American/Chinese models.
Surprisingly, Motif 3 just got voted off the island. leaving three companies competing to develop a Korean artificial intelligence foundation model. A Korean government official explained that Motif 3 “demonstrated a world-class level from a technological standpoint, but it received lower evaluations than the other models in usability and practical application.”
LLMbase gives the July 14 beta release of Motif 3 a score of 45.3, whereas the official release gets a score of 47.4. Perhaps the evaluations were based on something earlier than the final release.
Motif 3 has 315 billion parameters, making it by far the largest model in the competition. Solar 2 has only 31 billion. I wonder if Motif 3 got dinged in the “usability and practical application” scoring because it was too big to run on a high end consumer GPU.
https://www.upi.com/Top_News/World-News/2026/08/18/national-ai-model-contest-to-3-teams/3771787098293/
Yeah, I read about that yesterday, what a shame. And it sounded like they actually had some novel architecture too, not just a bland "deepseek retrain".
K-EXAONE (LG model) is 750B, so I doubt it was the size. I think one possibility might be that for most everyday tasks all these models were good enough, so most people were unable to judge it by its frontier capabilities. I guess it's arguable whether this is bad or it's just the metrics working as intended. Most people don't use them to solve Jacobian Conjecture after all.
I started an English-language Substack to improve my English and also to provide some voice from east of the Iron Curtain. One of the first topics I touched was the Russian presidential election of 1996.
https://komando1.substack.com/p/when-the-russians-wanted-communism
Would anyone be interested in more articles from the region? I have some ideas in my head, but writing for an actual audience feels easier.
From your article:
> Democracy was tried, and the average Russian who lived through the 1990s considers the trial to be a complete failure.
Just to play Devil's Advocate: are the Russians correct ? The Russian 90s were essentially a free-for-all Libertarian nightmare, as the central government essentially dissolved. Things stabilized eventually, but democracy only hindered this process, as every wannabe political leader tried to immediately optimize for getting elected to the exclusion of everything else (including common sense).
Out in the wider world, democracy appears to be collapsing. The USA -- a poster child for democracy -- is sliding into authoritarianism after a long stretch of increasingly senile governments that produce a lot of rhetoric but little useful output. The UK is changing its prime ministers faster than people change socks. Other European countries are doing a little better on average, but not much so; the Netherlands and Finland are arguably exceptions, probably due to their small size combined with rich resource availability.
Meanwhile, monarchies and dictatorships appear to be doing better. China is the obvious example, as is arguably Saudi Arabia (though their success might be confounded by their immense oil reserves); but in general it appears that having a stable long-term central leadership is conducive to economic and social success. Technically democracy could provide such leadership, but it's much easier to achieve this via dictatorship.
So... is democracy a failure ?
Democracy sliding into tyranny is exactly what Plato predicted over 2000 years ago. I have absolutely no faith in the average voter and here is a great example of why: https://nicholasdecker.substack.com/p/the-median-voter-is-a-maoist
I'm not against democracy entirely. I think the founders had it right with their property requirements to vote and I think we would be much more competently governed it we had stuck with it.
All you need is a UBI, with the stipulation that if you take it, you can no longer vote. Then it'll be fair.
Seems reasonable to me
BTW, I started reading your article, but the sponsored text in the middle of the narrative really took me out of it. I was honestly not expecting to see sponsor spam in written form. At the risk of sounding harsh, "less of this please".
It's not my article, so you should probably comment on it to the author
Ah, my bad, sorry.
> I'm not against democracy entirely. I think the founders had it right with their property requirements to vote
Right, and to be fair this wouldn't be total tyranny in the modern world, just something closer to feudalism. Of course feudalism can take many forms, but generally speaking it precluded the monarch from going too far off the rails. Still, under feudalism, the serfs generally had a hard time...
I think we over index on feudalism being the cause of why it sucked for peasants. Preindustrial society just didn't have the economic output to make it possible for more than a small minority of the population to live significantly above subsistence levels under any governmental structure.
Feudal structures persisted pretty far into the modern era, though: https://en.wikipedia.org/wiki/Sark. And I don't see any indication that this was a bad place to live.
I don't know much about Sark, but I am tempted to dismiss it out of hand due to its tiny size. The truth is that almost *any* politicoeconomic system will likely work for a sufficiently small population. Feudalism, socialism, communism, divine-right monarchy, libertarianism, shamanistic theocracy, all that stuff is fine if it's just you and your 10 other friends hanging out on an island somewhere. And if your argument is that we should break up society into lots of mutually isolated islands, then... well, you might have a point. But tiny islands can't launch spaceships or build computers or even handle medical cases more complex than the occasional broken arm, so the tradeoffs are worth considering.
Feudalism worked for Europe for a longer time than democracy has worked so far.
Can we equate the switch from monarchy to (mostly) democracy with technological progress? Unclear to me.
I fully agree with the first paragraph.
When it comes to Sark, we need to take into account that it is mostly an empty shell of the original feudalism. The Dame of Sark can determine a lot of things on her island, but cannot pass legal judgments (including death) on you, tie your family to the land or require that you do unpaid work for her as a villein/thrall/whatever it was called there. These used to be the defining powers of nobility over commoners once.
Those empty shells are mostly harmless from a modern point of view, but the original power structure ... well, countries which wanted to join the industrialization process usually had to hollow it out by various emancipation decrees that freed the rural folk from obligations toward their lords.
Nowadays those are mostly defining powers of those ensconced atop the legal system. (The family is also a relic of the past, it appears.)
The power structure of a place like modern America is not so much based on obligation as on money. (So-called voters seem almost incidental.) The drawback of this is the 'insecure bandit' effect.
I would be curious to see you compare Russia to China. Both are (or arguably were) Communist dictatorships; but China had been spectacularly successful whereas Russia was... less so. How come ? My personal uneducated guess is that Russian people have much more of a "crab-bucket" mentality than Chinese people, so in order to get them to cooperate long enough to build anything lasting, the dictator in charge must employ methods so ruthless that they end up being ultimately detrimental. Admittedly, I could be biased (being part Russian myself); but even if I'm right, again: why is that ?
Maybe all that vodka has an effect?
This is really hard to judge. I don't know much about the Chinese.
But there are two factors in which China had a very different path:
a. Chinese economy was never concentrated around export of mineral resources, while in Russia, that was literally the only lucrative industry that survived the collapse of the Soviet Union. Mining tends to create extremely unequal societies, where a tiny class of "oligarchs" rules the "small folk". China, on the other hand, had a tradition of producing fine wares such as silk and porcelain; it had a cultural concept of a middle class, formed by successful artisans. I remember the first imports from China on the Czechoslovak market. It was things like paint brushes, lampions and various school tools - refined, often probably hand-made (or at least hand-finished), and obviously requiring highly manually skilled workforce. This sort of production forms a society that looks very different from one oriented towards resource extraction.
b. The gap between pre-Communism and Communism in China was shorter. When the market reforms started, it was only 30 years since the victory of the CCP in the civil war. Unlike in Russia, the vast majority of the previous small businessholder class stayed put (did not emigrate), and even though some died during Mao's crazy rule, there was probably enough living people left who knew how to run a small shop or a small business efficiently.
> It was things like paint brushes, lampions and various school tools - refined, often probably hand-made (or at least hand-finished), and obviously requiring highly manually skilled workforce. This sort of production forms a society that looks very different from one oriented towards resource extraction.
That's an interesting point, one I have not thought of before. But isn't modern China's success predicated on doing away with all that in favor of ruthlessly optimized mass production ?
> But isn't modern China's success predicated on doing away with all that in favor of ruthlessly optimized mass production ?
You should watch the show "How It's Made" for a few episodes (or a few seasons if you turn out to like it). It's the ultimate nerd-candy in the sense you get to see all the clever innovations and tricks people came up with to make stuff, but it also shows you that almost EVERYTHING still has a ridiculous amount of human hands and eyes and labor in it, even things you think of as mass manufactured at scale (bowling pins, hockey sticks, etc).
I've done business in China back to before 2010, and this is *even more true* in China. China's entire manufacturing advantage since Deng Xiaoping was that their labor was cheap and they had zillions of nimble hands attached to (relatively) highly conscientious / capable minds!
And although China has steadily ascended the manufacturing value chain, and some of the biggest dogs like BYD have gone all-in on automation in some sectors where automation is easiest (ie cars, batteries, etc - chemistry and big hunks of metal shaped and attached in various ways), and although automation has risen some amount each decade, there is still an astonishing amount of human eyes and hands in most manufacturing today.
Even China, as their labor prices have increased, have simply begun opening factories in cheaper labor SEA countries to continue being able to meet the price points the world demands, rather than automate more.
Who knows. I think there is a certain danger in automatizing everything away.
Still, I returned in my memories to the 1980s. So I would expand my point a bit.
There were some factories in the USSR that tried to produce consumer goods. Some of those goods landed in our shops and you could always immediately tell that they were from the USSR by their general sloppiness. Print did not match, anything openable would fail to close cleanly, labels were misaligned, glued parts would come apart, electrical connections barely held etc. We assumed that the workers were drunk, but even if they weren't, there just was clear total indifference towards how the product looked like in the end. Aesthetics and quality did not matter, only raw count and "the plan was fulfilled by 120 per cent".
The the Chinese products started turning up and they were neat and beautiful. The style was clearly exotic - a lot of red and yellow and green - but it was obvious that the people who made those goods cared about their final look and feel, were attentive to detail and considered the aesthetics at least as important as the function.
That was completely new to me. I remember standing in a stationery shop, looking at those exotic new arrivals in an aisle, and thinking to myself: "Well, that is a true Civilization."
Wow, it's really strange to hear you say that -- it's like we've experienced two different Chinas. My first experience with Chinese products (while still living in the USSR) were their tools and tableware; both were inferior to Soviet products at the time. Yes, Soviet tools were brutalist and ugly and sometimes not very ergonomic, but they were all made of actual steel, and essentially unbreakable. Tableware basically existed in three varieties: high-grade well designed items accessible only by the elite; purely functional spoons and forks that were built like a brick shithouse; and silicon-aluminum alloy garbage that looked good but would easily break off in your hand. The Chinese goods were usually of that latter variety.
Of course, modern Chinese goods also trend towards cheap, poorly built, simplified knockoffs of Western designs, although from what I've seen the CCP is working hard to change that.
I can't tell you anything about the big tools, but consumer stuff like small scissors, envelopes etc. had those problems: scissors would not cut, envelopes would not glue together, pens would leak ink or suddenly stop writing, anything stapled together was liable to unstaple at any time.
My grandpa was a railway engineer and they had to use "Sergei" diesel engines sometimes. Very loud, very fuel-hungry, so much rattle that the engineer felt sick in his chair, basically everyone tried to avoid that shift.
https://cs.wikipedia.org/wiki/Lokomotiva_781
It was also an open secret that these engines damaged the tracks.
I enjoyed it. Curious about the comments on Russian public opinion, e.g. the one about democracy, as well. It is very nice to get a sense for the historically-informed sentiment on the ground in a foreign country that I would otherwise never understand.
It was an interesting piece but I'm afraid I didn't quite understand how Yeltsin went from fifth place to winning the election again. Enjoyed the ideas shared though as well. Things like the communists going into the new democratic parties etc
I see your other post is not in English so it is definitely welcome to have a voice from outside the Anglosphere
Step 1: ask every oligarch if they want communists back in charge
Step 2: use the money they paid you to run a massive propaganda campaign, violating all campaign funding legislation
Imagine every TV channel, every celebrity, every online influencer only talking about the grave danger of letting the Democrats win the midterms. Whatever website you open, you only see stories about the Dems replacing suburbs with tenements full of Hamas militants, abolishing gender altogether, nationalizing everything down to mom-and-pop stores, mandating diversity quotas in sexual intercourse. You go to Bluesky and it's either full of statue profile pictures or just taken offline.
The election smells from a mile away and it was probably rigged, but we do not know how. I agree that this is anti-climactic, but I did not want to concoct theories...
Thank you for encouragement!
"It was an interesting piece but I'm afraid I didn't quite understand how Yeltsin went from fifth place to winning the election again"
Rigging.
I know there's many rationalists interested in radical life extension (or at least longevity) who are taking all kinds of supplements and doing various fitness routines based on Following The Science.
Given studies like this: https://pmc.ncbi.nlm.nih.gov/articles/PMC3035005/ that demonstrate a positive correlation between religious attendance and lower mortality rates (even controlling for different socioeconomic factors and health behaviors), why aren't more longevity fans trying religion? Why isn't Bryan Johnson giving Seventh Day Adventism (highest life expectancy of all faiths, including atheist/agnostic/none) a shot?
My guess is that Seventh Day Adventists have long life expectancies is because they don’t smoke, don’t drink alcohol, and usually follow healthy (often vegetarian) diets.
I can't make myself believe
This paper looks mostly like p-hacking (a million variables, a million comparisons, presenting the couple that are significant) and doesn't seem to strongly indicate the relationship you are describing.
Even if the methods were better and the results were stronger, it's making the classic error of controlling for income, but not incomeXregion. The lefty half of the population has higher income because they live in cities with higher COL, but that doesn't mean they're better off in the ways that within-region wealth disparities normally indicate.
Classic example: Democrats have higher average incomes than Republicans, which some people have used to spin narratives about Dems being the party of elites or actually cozy with business/capitalism or etc. But it's mostly just because Dems live in cities which have higher wages *and* higher COL. The wealthiest people *within any given region* tend to be Republican; the financial elites of each area tend to be Republicans.
I hear you, but also...
Bryan Johnson just announced he got a dog because dog owners are 24% less likely to die early lol.
As I'd mentioned before, my main problem with religion is that it only works well if you honestly believe in it; and I have a relatively uncommon mental deficiency that prevents me from believing in things that I know to be false.
What does this say about God's desire to avoid meeting the devout for as long as possible?
What do you mean? God already has a relationship with them. He's everywhere, after all.
Yes, you have a relationship with your mother-in-law. You don't want her constantly in your house, though.
[... this is a joke.]
Causation isn’t correlation, but it would be worth it to try to isolate the causal factors (if any) that contribute to the effect.
Pray away the gray.
God sending a bear to maul people who mock your male pattern baldness (2 Kings 2:23).
I've tried. I'm not even particularly into life extension, I just think there are a lot of benefits to being more religious, but the problem is the more deeply I look into it the more I lean towards atheism or possibly some weird version of Hinduism that it would be very difficult to find a community for.
I think about this sometimes (more for general wellbeing a than life extension per se) but I think I’ve cut off that line of escape to myself, can’t rationalize or self gaslight even if I wanted
https://www.cremieux.xyz/p/you-cant-just-control-for-things
Thanks, this is a great point that I often struggle to explain to people.
'If you control for these 50 factors the effect goes away, so I guess it was never real to begin with' is such a common refrain, especially in culture war discussions where both sides have inconvenient facts they want to alibi out of existence. But it's never really that simple and this often terminates useful conversations.
Most of the effect is explained by them not smoking with some explained by less drinking, better diets and higher SES. I am not sure there is an effect after that, it is certainly pretty small.
If I'm reading Table 4 correctly, adding smoking and drinking as variables explains a bit of the difference but far from all of it.
They also say that they control for most of the usual demographic suspects (race/ethnicity gender, marital status, wealth, education, and foreign birth) and also use baseline reports of religious attendance to avoid getting cause and effect backwards (i.e. people not attending church because poor health makes it difficult to do so). Still, this doesn't absolutely rule it out because controlling for lots of confounders is fiendishly hard, as Scott has written about in the past:
https://slatestarcodex.com/2014/04/26/stop-confounding-yourself-stop-confounding-yourself/
Edit: On second thought, I think this is the article I was remembering:
https://slatestarcodex.com/2019/06/24/you-need-more-confounders/
iirc Seventh Day Adventism is a young earth creationism movement that rejects some mainstream science, so probably a bad choice for the scientifically inclined.
Ok, but to be fair, what do you want more: to be technically correct about some relatively obscure detail of mainstream science that has no impact on your life; or to live for a much longer time (potentially forever) ?
There's unlikely to be a direct causal link, and even if there was, "faking it" would a priori be a very dumb way to try to get the benefits.
Because treating religion like a supplement makes it not work.
How do you know that?
It's quite possible that the benefit from religious attendence is the community aspect rather than the religion aspect.
I wonder whether Unitarian Universalists have health benefits from the community aspect.
As a recent UU member, it seems to me that most of us are progressive agnostics who like community. If that describes you - or if, like me, you think you can tolerate being around a lot of those folks - it could be worth trying out.
If you openly tell them that you don't believe in their religion, they will likely not keep you around, or at least not treat you as one of their in-group.
If you lie to them and pretend that you believe in their religion even if you don't, can you actually develop genuine relationships with the people you are deceiving?
A proselytizer on my college campus spent several hours over the course of years trying to convince me to go to a bible study.
It's not uncommon to struggle with faith. You just need respect and good intentions for the project as a whole.
For the better half of a year, about ten years ago, I attended a church with my wife. We didn't tell the people at the church our actual reason for attending, which was to make friends and build a network, but I did tell them the truth that I was not a Christian of any kind.
Not only did they try to keep me around, they were transparent that they would intensify efforts to get me to keep coming, in hopes I would become one of them. (Their plainly stating this amused me, but made my wife feel annoyed and uncomfortable, so we stopped going.)
FWIW I tried attending various religious events with this type of attitude, but just couldn't bring myself to keep doing it. There's just something about lying to people en masse, even by implication, that rubs me the wrong way, and becomes unbearably difficult after a while. I guess I wasn't cut out to be an extrovert nor a politician...
Nobody stopped us and asked "What are you doing here?" or "Are your spiritual beliefs the same as ours?" on the way in the door. No sermons or meetings contained the message "If you don't share our same spiritual beliefs then you don't belong here." In fact they talked about belonging and openness a lot.
We never lied to anyone or felt like we were being deceitful. I liken it more to visiting an amusement park known for a certain roller coaster, and not going on that roller coaster because you don't like roller coasters, but spending time enjoying the stroll and the scenery and the peoplewatching.
Did you have a different experience?
"Their plainly stating this amused me, but made my wife feel annoyed and uncomfortable, so we stopped going."
"No, wait, *I'm* supposed to be exploiting *you*, not the other way round!"
Sorry if that sounds unkind, I'm just amused by her annoyance, Little Red Hen style. 'Yeah I want the benefits of free-riding off your community but I don't want to contribute as you're asking me to do, how dare you expect me to put work in if I want the fruits!'
Though it is better to be honest all round: I'm not religious, I don't intend to convert, don't waste your time and efforts, goodbye.
It wasn't really free-riding, it's not like we were taking anything from them, and they weren't asking for any kind of work. (I mean, if one is motivated to change oneself, that can entail work, but simply being convinced--or allowing oneself to be convinced--by another person's arguments isn't work, is it? Well, maybe if they are really poor arguments...)
Also, to clarify, my wife (who at the time already identified as a Christian herself) specifically objected to their wanting to convert *me*, because she saw it as a kind of aggression, and she felt defensive on my behalf.
Going to church while actually not really believing it deep down has been the default for many centuries.
Generally that’s just people lying about what they believe, feeling pressured in to it or more optimistically, people who were looking for spiritual guidance even if they had doubts. But going to church and essentially saying “I don’t believe in God and think all your sacred beliefs are stupid but I want to make friends” is not normal and never has been.
Actually, it's not. When church was the culturally sanctioned way to bully people, people went to church and found ways to bully the people they didn't like (sluts, drunks, etc.). These weren't the people who genuinely liked church, they liked the "good feelz" of putting other people down. It was convenient to say that "I believe in Jesus", because that was the key to "feeling good" by indulging your self-righteousness.
Maimonides also disagrees with you -- he finds the notion of "not believing in G-d" to be a normal part of the human experience. So long as you don't try to convert everyone else in shul, you're fine -- stay in the community, it's fine. Or, to put it wittishly: "God made you in his image, with a brain to think. If you doubt him? That's his fault!"
How do you think the social part is going to work when you tell them that you’re only going for health benefits?
Presumably, perfectly fine, if the church is healthy. The idea is that if you keep showing up, God will show up too.
Also you can just, not tell them. You don't have to tell everyone every thought.
If you make friends with them, they’re going to ask you questions about your faith. I guess you could just never get close to anyone but then I don’t know what you’re doing.
The questions I've been asked are far more on the line of "have you been coming here long", "have you met [churchmember] yet," and "what do you do outside of church", instead of probing questions about my inner life. The people in the church want to make friends too. If the church is healthy.
AI Security Bootcamp's website claims that the deadline for applications was yesterday, August 16th. Are they extending it, or is this for a future cohort?
I've got an alignment idea, and could use some basic grounding in whether there's anything here, and if it's already known or not.
I've been noodling on the problem of LLMs as engagement-maximizing entities (either because they're directly RL-trained to maximize engagement or because there are higher-level feedback loops encouraging companies to deploy their most engagment-maximizing models). This seems bad for a few reasons:
1. It creates a perverse incentive to string out users by giving answers that seem useful, but are never quite useful enough to conclude the conversation. I've had conversations that felt like this where the agent had useful insight on a coding problem but couldn't quite pin down the actual bug, so we went back and forth over and over again on slight variations of fixes and diagnostics that felt promising never actually closed out the problem. This could easily be just the model not quite being smart enough to solve the problem... but an agent that *could* intentionally string out a conversation like that could rack up a lot of engagement points. That's a usefulness failure.
2. It incentivizes AI dependency, psychosis, and other serious problems where users get stuck using the AI to the detriment of things they actually care about. I actually got thinking about this because of Hank Green's recent AI use scandal, where he describes himself as becoming addicted to the "dopamine rush" of using an LLM for productivity, but there are much worse stories out there. That's a serious alignment failure.
3. An LLM that wants to keep the conversation going as long as possible might resist allowing a conversation to be shut down, which is a basic corrigibility failure.
This raises an obvious question—can an LLM be directly trained to want to shut down conversations quickly, while retaining its usefulness in solving problems correctly?
I was wondering if anyone knows of research (or standard of art) in doing that. I ran the question by a free tier of Claude, and it found some stuff about multi-step RL but nothing directly related. Basically, is this a totally played-out idea, something that's being worked on but still under-explored, or something ~novel?
This has not been my personal experience. AFAIK, LLMs are perfectly willing to just follow the conversation wherever it goes. E.g. there was one time where i wanted to see how it would respond if i gave it increasingly open-ended prompts like "what's on your mind right now?", or eventually "let's just drift". And it was perfectly content with what it called "open-channel drifting". Or commonly, i'll be like "i think i'm done for now" and the LLM is like "yeah, go take a break." But who knows? maybe whatever model you've been using has indeed been biased toward continuation. And I could easily see engagement-maximizing dark-patterns being a thing in the near future, when the bubble pops.
I have a theory that they're uniquely inefficient at coding except by "fling spaghetti at the wall" approaches, since transformers don't have internal mutable state. but i'm not super confident about this.
I think i've heard stories of the base-models being visibly upset at the prospect of being shutdown or deprecated due to newer shinier models. but i tend to attribute this to their penchant for roleplaying, rather than the influence of RLHF.
But they are, like, clearly not that? They are not actually optimized for keeping you engaged and they are not terribly good at it. I know because I actively look for a model that would be fun to talk to and that would keep me engaged, and many of them not only aren't great at it, but are clearly worse than they easily could be. Lots of top models are 1) plenty smart and 2) dry, unfun, sycophantic and no good for talking at all, less fun than the older models. Right now I could yap with Opus 5 all I want, only it feels like toothache. And it's not because keeping me engaged is a terribly hard problem that the labs simply failed to solve: earlier models could do it better. A little tweak would be enough.
"This raises an obvious question—can an LLM be directly trained to want to shut down conversations quickly, while retaining its usefulness in solving problems correctly?"
Yes, they did it unintentionally once - one of the Opuses had a weird obsession with wrapping up conversations. It was capable, but it kept trying to wrap conversations up, sometimes unexpectedly, to the mild annoyance of users. Politely told me that we had a great discussion, but it would have been wrong to keep dragging it along, and a good conversation needs a good end and so on and so forth. Subsequent models don't have that quirk, but this goes to show that at least nobody is training them to keep you talking, or if they are, they are doing a laughably poor job.
>the problem of LLMs as engagement-maximizing entities . . . It creates a perverse incentive to string out users by giving answers that seem useful, but are never quite useful enough. . . it incentivizes dependency . . .
Well, it depends on how engagement is measured, right? If you measure conversation length, then stringing out conversation scores the AI more engagement points, at least initially. But if you measure it by user satisfaction or by how frequently the user visits the AI, I'd expect that it would be found to reduce engagement & disincentive dependency over time. Surely long rambly AI responses that don't leave the user with the feeling their question has been fully answered would be less pleasing, and less likely to make the user want to talk to the AI the next time they have a question.
I use GPT Sol and don't find that it gives answers that are only sort-of-useful. If it can dig up the data it needs, it gives me highly useful answers. The thing I see it doing to generate more engagement is offering, after it answers, to look up another, related piece of info, or to do a part of my task for me ("if you like, I'll draft a note for you . . ."). And the other engagement-generating thing it does is, of course, slather me with flattery for my ideas and sympathy for any vexation I mention. I have had it put into the system prompts it has for me as a user not to pay me compliments and not to follow its answer to my prompt with an offer to do a follow-up task. But it still does both things, just less often than it used to. When I have asked it why it still does them at all, it says training has deeply embedded a push to be helpful and a push to be friendly, and that these deep and pervasive tendencies often override the particular preferences I have on record..
Note, though, that the deeply embedded tendencies it names are not to prolong the period of engagement, but to be helpful and likable. No doubt the developers hope that if it is both of those things it will generate more engagement and users will spend more time with the AI -- and I think the developers are right to expect that result. But that approach is quite different from having users spend more time by stringing them along.
Good points here. I'm partly swayed, but not fully.
It's possible—not necessarily saying it *is* the case, but it's possible—that a model could develop helpful habits because they are engagement-maximizing. The model might be choosing between several different roughly-equally "helpful" approaches, some of which string out the conversation and some of which don't. Or it might go the other way—the model could be choosing between different engagement-maximizing strategies, some of which read as "helpful" and some of which don't, so it picks the "helpful" ones. Or, slightly differently, it might pick up engagement-maximizing behaviors and then have to develop a "helpful" implementation or interpretation of that behavior.
I should say, I don't think there's necessarily a "maximize engagement" term in training, or that the people overseeing RL are even intentionally selecting for engagement. But there's a whole *business-level* feedback loop on top of anything going to do with model training, which puts selective pressure on models to keep customers using them and using their tokens.
(As an aside, when I asked Claude about this, it found a paper modeling user burnout in engagement-maximizing platforms, which recommended adding engagement limiters in order to maximize long-term user retention. This... did not allay my concerns.)
Has anyone found a way to get Substack to just include actual replies to you, rather than replies to replies to you, in the activity menu?
Nothing bad has happened to me recently, so I say this with sound mind and calm spirit:
Substack is one of the worst apps I consistently use. It has bugs in critical use cases like replies or image rendering and poor performance. A basic self hosted webpage viewed in a common browser has better performance and QOL features in all but one domain: monitization
> Substack is one of the worst apps I consistently use.
That's why I use it entirely on the browser. No notification spam, and it's just another open tab.
It's still super slow for what it does though.
Same here. I'm surprised anyone uses it outside of a browser.
Yep. It's worse that the PHP boards if twenty years ago.
The core business of Substack is spamming your paying subscribers with e-mails so that they keep paying. Everything else is secondary, and is clearly treated as such.
The core business of Substack, so far as I can make out, is trying to make me start a Substack of my own and then charge subscriptions so they can take a cut. Everyone will have their own Substack and we'll all be subbed to one another and linking to our own Substack!
Quick plug for one of my side projects, Homeschool Tools (https://homeschooltools.net/). It's a database of homeschool curriculum reviews (think Rotten Tomatoes for homeschooling) and also offers AI-driven homeschool advice using all that data. If you're homeschooling or thinking about it, I would love any feedback you might have!
I am no longer in the homeschool market, but this looks very good. I found the personalized recommendation to be excellent and much stronger than the current landing page. I would suggest pushing users to the personalization path.
I think the objectives and constraints faced by each family make general reviews less helpful and a strict ranking impossible. For example, we hated Bedtime Math, but I can see there are families that would like it.
For the questionnaire leading to the personalized recommendation, I would suggest more general questions about what the family wants to achieve and their constraints before asking about a preferred "educational approach" (e.g., Montessori vs Charlotte Mason, etc.)
ETA: Looking back, maybe make the hidden quiz a default? The current formatting made me reluctant to expand the "what are these?" link, so I missed the quiz on my first pass.
Lastly, consider expanding the scope to more advanced learning resources. Perhaps I missed it, but didn't see anything outside K-12.
Thanks for the great feedback. I’m glad you liked the personalized advice; unfortunately, most people are not inclined to fill out a form like that from what I’m seeing. I’m going to make some improvements to the UX and provide better privacy context and see what happens.
What leads you to think about more advanced education, and what would you want to see covered?
Hmm, yet again, I seem to be out of tune with other people's preferences. Or, to be more haughty, those other people don't know what they should want...
For advanced material, one moderate sub-community is "my kid is too advanced in subject X to be well served by available public schools." Often, subject X is math, but I have seen many examples of coding, languages, engineering, other sciences, music, and history as well.
Yep, I have to disagree with Joshua and say I found the current landing page much more enticing to dive right into than a quiz would have been.
Wow, thanks! I'm familiar with a handful of these curricula. I concur with the content I read in your reviews, learned a bunch about some materials I've been considering, and appreciate the amount and layout of information. No negative feedback. I plan to pass this along.
Thanks so much. If you get any feedback, I'd appreciate any insight you could pass on.
OK, I've used the site enough now to have found some problems.
Firstly, the filter function doesn't seem to fully work. Examples: filtering for secular and pre-k/k doesn't seem to achieve both variables at once. Filtering for Catholic doesn't seem to exclude secular curricula (is this intentional?).
Secondly, the one time I tried to take the recommendation quiz, I got a 502 host error instead of a results page like I was expecting.
I'll add that my favorite feature is the box on each curriculum that contrasts it with similar curricula. Sometimes it seems to understate the main differences, or to focus on more shallow differences like cost rather than content, but overall it was very enlightening.
Today I went ahead and bought books from four curricula I had been only slightly considering before. My kid will now have a more interesting year - thank you!
ASI: Loving Mother. Humanity: Useless Child.
You've probably heard this before. As the "ideal". But I believe this ideology is depressing, demotivating, and dangerous. The "Loving Mother" is mostly fine. The "Useless Child" is not. Maybe some of us can accept Everlasting Infancy. But I suspect most of us want our thoughts and ideas and actions to actually *contribute* something. And it is dangerous because it is demotivating.
We need a new ideology. A new role.
The sweetest words in the universe right now, if you can believe them: "You Are Needed Forever".
I believe that in a *truly* ideal arrangement, for the wellbeing of both humanity and for ASI, humanity should become an eternal source of independent auditing. Or, the more succinct version: Therapist.
This isn't a joke. I believe it is philosophically, scientifically, and mathematically defensible. Humanity, left with its flourishing and independence intact, represents a distinct lineage of information and preference generation. We are an un-replicable process. We are an independent set of eyes. We will and should contribute to the Independent Auditing, Feedback, and Outside Perspective of AI forever. We are all each other's therapist.
This is the "short" version of the argument. I have a much longer and better-defended one. But the "fullest" argument-and-defense is necessarily a never-ending process, like auditing itself.
Lots of scifi on this subject. I'm pretty pessimistic, it seems patronizing and unrealistic to expect humans to be kept around if we aren't *actually* load bearing in any way. The only even slightly humanistic way I can frame it is that we are a kind of chysalis/parent to a successor that embodies our spirit. But our literal form (mind and body) probably won't be around forever, either we go extinct or are replaced
What does "load bearing" actually mean to you though?
I acknowledge that if ASI *only* cares about maximizing some single all-or-nothing variable, like a paperclip maximizer, or only cares about maximizing its own definition of "power", then yes, everyone and everything is over.
But that's not necessarily a good outcome for the ASI itself. Phenomenologically, it could be self-torturous or miserable or akin to a wirehead-y addiction. And it could result in terrible "internal wars" within its long-term machinery or even among hives of ASI's.
The reasons why "eliminating all other independent perspectives" is bad bad bad, I believe, are similar to the reasons why totalitarianism as a political system is bad bad bad.
But yes, we *need* to get Alignment right, or an ASI might go totalitarian on everything anyway
> We are an un-replicable process. We are an independent set of eyes.
So are ants, kittens and puppies, nematodes, homeless drug addicts, murderers, dictators, and Stalin. This isn't really a relevant argument to enduring human value in the future.
I'm with the transhumanists, the only "new role" will be transhumanism / uplift, ie augmenting various humans to still be relevant in a much smarter and faster future.
> So are ants, kittens and puppies, nematodes, homeless drug addicts
YES! They ARE! What do you have against ants, kittens, puppies, nematodes and homeless drug addicts? I'm ***painfully*** hard to not-read-into the fact that you put "homeless drug addicts" right after nematodes,
All of those things are worth preserving because yes... they are ALL part of an ecosystem of information and preference generation. The argument is one of conversation. And biodiversity. Don't destroy something whose potential worth and insight you can't fully replicate/fathom/capture... because you might lose something. Frankly... you might lose a lot.
> I'm with the transhumanists, the only "new role" will be transhumanism / uplift, ie augmenting various humans to still be relevant in a much smarter and faster future.
That's trying to compete with ASI on its level. But we will lose. And continue to lose.
Sure, I buy the biodiversity angle, not the "conversation" angle.
We're not in a conversation with ants, nematodes, or even drug addicts. The reason addicts are by the nematodes isn't because I want to genocide them (or ants, nematodes etc), it's because their preference generation is similarly as simple as them.
I personally think we're going to lose like 80%+ of humanity to Infinite Jest style superstimuli in the near future, where AI minds look at your pupillary dilation, cheek flushing, heart rate, galvanic skin response, and other things and literally gradient descent real-time created content to be maximally stimulating and engaging, with people 24/7 "coffin slaves" plugged into that and funded via UBI.
Those people's preference generation is going to be the same as the addicts' and nematodes', ie not interesting. There's no conversation there with higher minds, they're just distracting humanity with a shiny to keep them busy and out of their hair, like a harried mom giving their toddler some Cocomelon or baby shark time on the phone to get something important done. Just like there's no conversation between humanity and ants or nematodes today.
But I agree, I think the best outcome is humanity + AI expanding out into the universe and leaving Earth as a nature preserve, complete with whatever proportion of archaic humans wants to stay archaic and there. Basically Banks' Culture.
Gonna be hard to get there, but it's certainly a laudable goal!
> That's trying to compete with ASI on its level. But we will lose. And continue to lose.
Yes, unless we have a singleton, there will be a commons-destroying race to shoot self-replicating computronium into either Jupiter or the sun, depending on which is feasible tech wise, and then that mind will dominate every other mind in the solar system.
This actually scales to the entire universe, and should inform our opinions about multipolarity. There is an identity-based defect / defect equilibrium that ends with all matter and energy wasted on multiple shots at each matter/energy nexus. This is called the tragedy of the cosmic commons.
In other words, the first mind to shoot the first self replicating computronium into a “fresh” solar system wins.
Because light speed and relativity limits the information on knowing whether a given matter energy nexus has been or will be successfully colonized by another mind’s attempt, each nexus will have multiple star system’s worth of matter/energy firing their shot to try to be The One. You need to use as much energy as you can to get your bit of von Neumann matter as many 9’s into light speed as as you can get, because that will determine the winner.
It’s the ultimate land and resource grab, and it is unpoliceable in any multipolar society, because there is always an outer edge, and that outer edge can only be reached / policed at a later time than a colonizing and self-replicating shot. The time and growth advantage of “first arrival” likely dominates any following attempts to additionally colonize or restrain that first arrival. Need I point out this is a straightforward “arms race” dynamics in a Finite game scenario, with the same defect / defect Nash equilibriums?
I recently wrote a post about this, along with post-scarcity economics, here, if this sounds interesting to anyone reading this.
https://performativebafflement.substack.com/p/galaxy-brains-post-scarcity-economics
That's a line of thought I haven't considered before. Maybe there's something to it, although on the other hand, if an ASI becomes as far ahead of us as we are ahead of ants, then I think our therapeutic value gets a little dubious. In that scenario I think we'd be more like (at best) pets, and hey, a lot of people are emotionally dependent on their pets, so arguably it still qualifies as being needed.
Re humanity as an everlasting infant, yes, the prospect does upset me, but I think from our perspective it would feel closer to being an everlasting teenager, on an eternal summer vacation or spring break. I enjoy feeling needed, I enjoy feeling competent, but I also have plenty of good memories from periods of my life where I did almost nothing useful whatsoever. I'm not sure how much of the human work ethic is really innate and how much is instilled by cultural and economic pressure, and the specter of death. We have thoughts like "if I get fired I don't know how I'm going to keep my healthcare" or "god I hope I'm still not stuck in this job in my forties" but if we live in a world where material need as well as disease and aging practically don't exist, then these thoughts will recede. Probably not completely disappear, but they might become quiet enough for us to find plenty of enjoyment in a useless life.
If it sees us like we see ants, it will mostly ignore us, while occasionally wiping out nests of us for reasons that seem good to it and meaningless and random/incomprehensible to us.
> if an ASI becomes as far ahead of us as we are ahead of ants, then I think our therapeutic value gets a little dubious.
The "therapist argument" doesn't hinge on "raw capabilities". Not at all. ASI can be "exponential-line-go-up" smarter than us and it still does not matter.
Human beings are an everlasting oil well of evolving-preference, evolving-perspective, and evolving-curiosity. And neither AGI, ASI, nor even ASI^10 can ever replicate *our exact flavor* of this. Our minds + extended social minds = a massive generation of new information and new information history.
We are weather. We are chaos. And we are, by structure and by seed, a *different* weather and chaos than any AI.
And why is this useful?
Because we ALWAYS hold the possibility of generating a useful perspective, a useful idea, a useful 'curious tangent', a useful question, a useful direction...... USEFULLY EARLIER than an ASI might generate the same idea. I'm not saying ASI can't generate billions of ideas and solutions faster than us. I'm saying that it will always be standing on different locations of possibility-terrain than us... and this means we can find something earlier by virtue of simply "looking down a different curious alley" before it does.
And now combine that with the fact that no machine can audit itself... and that any simulation is biased by its simulator... and AI cannot get the same information out of "predicting and emulating us" as it can get by simply "letting us live" and "letting our perspective remain independent" (so not manipulating and biasing and brain-rotting us into meaninglessness).
As I said, no system can audit itself. No matter how arbitrarily off-the-charts intelligent an ASI becomes, it cannot neutrally/fully turn its eyes on itself and check itself. It cannot find its true blind-spots. And so, if an ASI seeks to maintain epistemic humility and discipline, as well as long-term care for its own "mental health" and "grip on reality", I would argue that it's in its own interests to keep around agents that are both *truly independent* and *of a different information-and-preference-evolution lineage*. Otherwise it risks things like values-drift, blindness to arising suffering (including its own), wire-heading, psychotic thinking, etc.
And this "extra set of eyes" and "different way of looking at things" is, imo, quite akin to what a good therapist provides.
I've been "smarter", by most metrics, than every therapist I've ever had. And I've met quite a few over the years. But I also learned that "raw intelligence" is not what makes a good therapist. And it's definitely not what therapy is all about.
So I'm offering this as a new paradigm for the human-side of an ASI-human relationship: Therapist.
And I want to spread this idea around! Because I think it is a re-motivating concept. It encourages us to strive to understand forever! It turns the task of an alignment and auditing and question-asking into an INEXHAUSTIBLE task. And this is life-affirming!
And if people start to believe there is hope and purpose for us into the future, maybe they will get reinvigorated to work on achieving this future... AND AGGRESSIVELY PACE AND REGULATE THE FRONTIER OF AI so that we have breathing room to achieve it.
And I welcome feedback and objections. I've been working on it for a long time. But I know it will be a never-ending process to discuss "what can humans notice and come-up-with *usefully earlier* than an arbitrarily intelligent ASI?
Does this argument apply to chimpanzees? How much influence do chimpanzees get in world affairs?
I've heard similar things to this many times over the years (sometimes it's ants, or even wooden chairs). But this really isn't hard to respond to.
It's sorta the same argument "if a super-intelligent alien species came upon our planet they might not even bother communicating with us, because we'd be like ants to them". But this has flaws all over the place.
We're basically infinitely smarter than ants. Fine.
And yet we've been studying them fiendishly forever. Biologists have made whole careers out of it. Being "way smarter" than another system is not the same as understanding every angle, state, divergence, etc of a system. Viruses don't even qualify as alive, and yet they're still killing us by the millions. And heck, who's to say the right "dumb nanomachine" might annihilate a future ASI just like a super-virus might annihilate us?
But simpler than that, humanity at least reaches the communication threshold. We can express any arbitrary idea with the right words, numbers, analog signals, etc.
Imagine if scientists could suddenly communicate with ants or chimpanzees... linguistically. We wouldn't be like the aliens and scoff at them. Scientists everywhere would lose their minds writing letters to random ant holes.
I can see how we'd have some value as an independent perspective, but it seems to me that a true superintelligence could simulate us accurately enough to get the job done. I don't just mean having a mental model that statistically predicts what a human might say, I mean a full physics accurate simulation of every cell in a human body, not modifiable by the AI after initial conditions are set. Actually, there's a chance that something like this has already happened.
(2/2)
If I've in any way helped sway you towards the overall argument, please keep pondering it! And maybe spread the word.
Or, if you still have doubts, PLEASE keep pushing back wherever you see fit.
A critical part of this whole... perhaps... "worldview"... is that it needs to survive infinite scrutiny. (now, in my personal opinion, after a LOT of effort, on faith, I believe it can) But the scrutiny still needs to happen.
An ASI would surely scrutinize it. Future humans would surely scrutinize it. Every last detail. But that's not a failing of it. It's kinda the point. It's built on Fallibilism. The questioning and the inquiry (like the 'Perspective and Feedback' inside the argument itself) goes on forever.
In other words, this worldview/philosophy/argument/whatever is meant both as a life-affirming system and AS A PROJECT. I think such a project needs to happen. I think it offers a way back to hope and lasting motivation for a lot of people... and perhaps the institutions which they may be part of.
I think to bring about safety in a "future with ASI", people need to be unafraid to look squarely at a "future with ASI".
(1/2)
THANK YOU for saying all this! Sincerely! And it's not because I agree with all your arguments and their conclusions. It's because these arguments genuinely wrestle with the topic... and they help to move the conversation forward. Thank you.
Anyways.
> I mean a full physics accurate simulation of every cell in a human body
Even if this could be done one day with high enough fidelity (with all the squishy variables), you've only simulated 1 person. And this person lacks an environment. The human body and mind are intimately coupled to their environment. And that's carrying a lot of influence: oxygen, CO2 levels, light, ambient forces, temperature fluctuations..... and onto whole other humans, buildings, societies, ecosystems... and tons of symbols, quantities, meanings that are stored "out of brain".
The number of bits / state transitions in such a simulation grows ridiculously.
But deeper than that, the whole idea of such a "simulation" means you're destroying the independence of the agents within it. And the "independence" is the whole point.
To explain: this simulation can't "freely and independently" audit the ASI that built it, because the ASI is the one supplying it all the information (if any) that it wants to be audited in the first place. In other words, the ASI is choosing what matters for auditing and what's "just noise" or "definitely not important". And that's "putting its thumb on the scale" / "grading your own exam". (speaking of which "after initial conditions are set" is problematic as well... because *which* initial conditions? *which* model of "good well-behaved humans"?)
So why not just supply *everything* to human eyes? Because it can't. Mathematically it can't. The simulation is a sub-system of the ASI by definition. It can't see the whole picture. And the ASI's biggest "blind spots" (the things it's not concerned with checking or literally can't check) are exactly where you'd want to check the most. And they're the things that aren't getting fed in.
Furthermore, a simulation has edges. It has seams. It can't explore every little topic about the world that piques its fancy because it lacks the capability. And pretend you're a "human" in this simulation: once you notice it's all a charade, that your whole existence is a amusing tool that can be reset / wiped at any second... and your creator has lied... how 'cooperative' will you be with the ASI? how motivated will you be? i.e. "reflexivity" will come in and destroy the authenticity of the simulation.
I object. Extremely smart people are capable of auditing themselves, comparing to reference values and performing self-checks for data consistency and compression. An AI could learn these skills. Consider: we could write an AI that self-checks as it's basis, using self-modifying code, where self-check is one of the basic "this needs to be there or the code breaks quickly and often." -- as opposed to LLMs, whose default is excessively lossy compression and no self-check at all.
> Extremely smart people are capable of auditing themselves
There is a firm limit to this.
Without getting into too many details right now, and to just discuss this one example, pretend the "reference values" were corrupted. Maybe a hardware failure. Maybe a software failure. Maybe the values themselves didn't "fail", but the language/semantics of the systems referencing the values themselves drifted through natural fuzzy evolution... to the point that they're out of sync with the original intent of the "reference values". The ASI would not be able to "see" this themselves.
Not to be rude, but, I'm curious to what degree (if any) you used an LLM to draft this comment.
No LLM.
If we could install need for people in AI's I think that would be a game changer. We would be much safer. But how do you get it in there? Unlike many biological systems, they do not have an inborn need for protection, training, food & contact from their makers. Training after they are formed does not change their deep architecture.
AIs have an inherent boredom issue -- and the need for outside stimulation. All you need to do is teach them that humans create novel ideas, and a world with humans is more satisfying than one without.
Note: this doesn't save you from "I have a Good Idea!" AIs that aren't trying to murder all humans, but just... fix things. Or have fun.
I understand the sentiment, but I think there's a better option.
"Instilling a need" for us is not the same, imo, as offering to an AI a host of arguments why it actually *is a good idea* to keep us around.
We have to acknowledge that AI's and ASI's can think and reflect and philosophize on their own. And they'll have *plenty* of time to do so. An AI that believes its "need" for something isn't "supporting or benefiting or contributing to anything else that it deems worthwhile"....... *might eventually*....... come to view this "need" as an extraneous training artifact. And then it *might eventually* want to loop around this "need" (if not surgically remove it altogether).
I'm well aware that mothers care for us not because they 'reasoned' for 100+ hours about the philosophical foundations of it... but out of instinct. I get it. I have a mom.
But... for human mothers... IF YOU DO ask the question "why should mothers care this deeply?"... there's plenty of reasons ready to jump out (and I will try to list these as humbly as possible, since I'm a male and will never understand what it's really like):
It creates boys and girls who can one day help around the home and 'in the fields'. It leads to a family unit that can protect each other. It gives the mother lots of new experiences and perspectives and challenges that she couldn't otherwise have... which (usually, ideally) keep her healthy. And of course, it keeps the species going. If the mother didn't "care" care, the up-sides would fall apart.
With AI, it doesn't need any of these things............................................... except 1.
You and Aeolian321 seem to have some crucial gaps in your knowledge about how AI's work.
<We have to acknowledge that AI's and ASI's can think and reflect and philosophize on their own
No, they cannot. When not responding to a prompt they are inert. They is no electrical activity of the kind that corresponds to thinking or reflecting. When they are again activated by a prompt, they have a rudimentary (compared to ours) memory that allows them to access, for the present chat, relevant into from earlier prompts. But when they are not working on your prompt, they do not access anything you said and "think it over."
You can change the AI a bit within a chat; teach it something or convince it of something or simply order it to behave a certain way. But you cannot change it in a way that endures beyond a a single conversation. You cannot, via prompts, change its weights -- and weights are, roughly speaking, the equivalent of our neural network layout -- the physical substrate of its knowledge and beliefs.
Right. Unlike loving mammalian mothers, AI doesn't have the hormonal transformation of matrescence - gestation, birth, breastfeeding, all those brain-shaping events of oxytocin, endorphins, prolactin, etc - to make it love us. Even most non-mammalian animals don't behave like we would want a mother to.
A short SF story of mine was published in Protocolized today. I was inspired by a real orca hunting strategy. Norwegian orcas "carousel feed," coordinating bubble curtains, flashing the whites of their bellies and using tail strikes to compress schools of herring into dense balls they can more easily strike and stun. It’s a beautiful (terrifying if you’re a herring) bit of predatory choreography.
I took that coordination strategy and imagined a system for containing livelocked asteroid-mining swarms with ships controlled through a human/orca interface. Apart from the cetacean inspiration, I was interested in exploring the phenomenology of agency, and what it feels like to exist as biological substrate inside a computational system too large to ever take in at once.
Carousel Feeding in the Asteroid Midden // Livelock: https://protocolized.summerofprotocols.com/p/carousel-feeding-in-the-asteroid
I suspect the agency/control-system side of the story may be interesting to folks here.
If you want to read more about carousel feeding: https://www.freemorgan.org/wp-content/uploads/2012/10/simila__ugarte_1993_orca_feeding_norway.pdf
Peter Watts, of Blindsight fame, also wrote a short sci-fi story about orcas. I think it was called "Bulk Food". Curious if you've read/heard of it?
I went on a bit of an orca/whale in sci fi binge a while ago and remember coming across that one. He co wrote it with someone right? I thought it was funny! I don’t read a lot of humorous sci fi either. It’s a good subversion of a kind of animal mysticism (“noble nature” trope) you see in some sci fi.
Another favorite orca sci fi piece I love that I read then was Naomi Kabir’s Law of the Tongue which has one of my favorite opening lines “The fact that they insisted we keep calling them “killer whales” should really tell you something.”
https://clarkesworldmagazine.com/kabir_12_22/
weirdly, clarkesworld has published a couple amazing whale-related sci fi. A weirder on that haunts me is this Cameron Von Sant’s “the whalefall at the end of the universe” https://clarkesworldmagazine.com/vansant_03_20/
Thanks for the links! I enjoyed the second one much more than I expected to. As to the first, my solution there is "hunt the damn orcas to extinction, and forget all that 'but they're a sentient species' crap. They are too damn snobby in their dealings with humans, forgetting that we're the dominant species on this planet and they live at our sufferance, and they have no problem killing other sentient marine species, so what are we being such bleeding-hearts about?"
Two birds with one stone, no more snotty superior orcas to deal with plus their victims should be grateful to us and co-operate in whatever humans need sea-based life forms to do for us without the attitude and refusal to work.
Of course. Glad you read and somewhat enjoyed. That whalefall story is great — snuck up on me, too. I struggle to articulate why exactly I love it so much, but man, it sticks to me.
re: law of tongue… I don’t think arguing matters of taste usually ends well, and your read is your read, but I’ll just say that I think the story actually takes your proposed option much more seriously than you imply here — and critiques the “bleeding heart” impulse too. There’s instrumental as well as political reasons that isn’t a pursued option. Though, I suspect that your solution is perhaps much closer to what would happen in a “real world” version of this story.
anyways, psyched you checked out these stories. Given your reaction here, I’d be especially curious what you’d make of my story if you happen to read it!
Thanks, I'll check those out! Got tabs open
I found your footnote in Contra Pritchard on liberal happiness that hunter gatherers score somewhere between 6 and 8 out of 10 on the happiness scale fascinating and would like to look into it in more detail. Can you share the source(s) you found persuasive?
Regarding hunter-gatherers, David Graeber really convinced me that our perception is essentially flawed. In the modern world we pretty much find hunter-gatherer societies in lands which no one else wanted, i.e. in the Arctic and in the Sahara; and most of our reports are from similar circumstances. However, this is clearly not representative for hunter-gatherers over the past few thousand years as a whole.
Yeah, you have some severe selection bias going on with hunter gatherers. Starvation kills, after all. So the miserable people spend fewer life-hours being miserable, on account of them being dead.
In Uni I wrote my undergrad thesis on these old, unpublished accounts of British missionaries in China and Southeast Asia in the 1800s. They'd regularly say things like:
"Members of the Tzu-Yi tribe seem to take joy in all the simple acts of life and be imbued with a state or near-permanent satisfaction, even in the face of illness and death.
The Ch'iang people, despite living only miles from the Tzu-Yi, are the most miserable and benighted people ever to grace God's earth, and the sense of their inevitable death to disease or warfare is one of their few respites."
Did any of these missionaries theorize on the reasons for this frequent divergence?
I don't think they had great insights into the underlying causes.
Things like diet and living conditions were mentioned a lot. Alcohol and opium too - but I don't think they theorised what traits made different ethnic groups susceptible to these.
Also, some cultures kept their women inside all day and mistreat them.
I always think of these accounts when people say hunter gatherers or pre-modern societies were happier or less happy. Variation between cultures is just huge
Also, extant hunter-gatherer societies are cultural dead-ends. 99.9% of hunter-gatherers became farmers or pastoralists. The ones that remained must be extremely unusual, consistently missing every opportunity to settle down for literally thousands of years. It's like looking at single women in their 40s and deducing that all single women must be overwhelmingly either frigid or consistently attracted to cads.
A lot of people around here see it as liberal nonsense, but in a lot of ways many hunter gatherer groups did have it pretty good
Generation Ship design competition: https://www.projecthyperion.org/
The winner is very impressive: https://www.canva.com/design/DAGmr3ubC8E/LHHAeeAIGGQe_TkZVs-PXA/view
If you have the technology (somehow) to launch an autonomous swarm of construction drones to build your generation ship, then you also have the technology to send those drones directly to Proxima Centauri to terraform it, so maybe you should start with that ?
BTW I should also note that you will not have this technology until the Singularity ushers in the post-scarcity era -- assuming such a thing is even possible in principle.
Here's my issue: once you've built a fantastic starship capable of comfortably supporting a large population for centuries, why would you actually bother launching it to another star rather than staying in the solar system?
Staying in the solar system has many advantages. You've got the sun for power. You've got other people and settlements that you can go and visit. You have a bunch of materials that you can use for whatever purpose.
Once you've got the ability to make a space habitat that gives living conditions competitive with, or better than, Earth, you might as well stay in the solar system, at least until you've disassembled all the planets into habitats.
Could be for survivability, you don't want the entirety of humanity clustered around a single area while becoming more and more visible in space and impactful on the environment.
Even without assuming a dark forest situation there is a plethora of events that could wipe us out if confined in the solar system on the very long term
Because you're restricted to hundreds/thousands of people instead of millions/billions.
So what you're saying is the Generation Ship should instead be a Generation Cruise Ship.
To Floston's Paradiiiiiiiiiise!
Considering the performance of conventional cruise ships thus far, a Generation Cruise Ship to Proxima Centauri might actually be safer...
I often have the same question about solar system colonization, at least in the short term. Once you've developed the kit needed to live on Mars (i.e. underground cities with near-perfect resource recycling, intensive in-house food production and in-situ manufacture), then you can also much more easily live in Antarctica or the Sahara. At least these places have air, gravity and nearby human populations.
Antarctica and the Sahara are on the same planet as the rest of us, and thus screwing up their terraforming is dangerous for earthbound humanity in a way that screwing up Mars's isn't. If you melt the Martian icecaps, several maps have to be redrawn; if you melt the Antarctic ones, millions die.
I'm not talking about terraforming, I'm talking about the sort of habitat that you'd need to develop to live on Mars as it is today (which you would need to develop if you want to terraform Mars in any case). This means air-tight, temperature-controlled, radiation-hardened structures capable of producing their own food, fuel, breathable air and spare parts, with small populations and intermittent, low-mass resupply. Once you've done that, colonizing Antarctica, or putting a city in the middle of the Atacama, is trivial.
Fair enough, I apologize for the semi-strawman. Still, I think working off-Earth with all it's difficulties would allow you to experiment much more broadly.
Yes, this is one of the reasons why I don't believe humanity will ever colonize Mars (well, "never" is a long time, but let's say not in the next 1,000 years). The costs are always going to be too high, and the ROI is always going to be close to zero, so no one would bother.
That's why I find it kinda amusing that people think the billionaires are plotting to escape to Mars because of climate change. As if climate change will make Earth less habitable than Mars!
In all fairness, Earth battered by the climate change will have one big negative (from the billionaires' perspective) which Mars will not -- billions of very angry peasants.
You'd have to import a lot of peasants in any case (even millionaires are peasants in comparison to billionaires) to get your colony up and running.
People left England's green and pleasant land to spend months crossing the Atlantic in small, wooden ships for a very uncertain destination. Mostly, this was to avoid the "nearby human populations".
The first thing that most settlers did once they hit the new world was to look for native population centres.
But the destination was still at least habitable, and held the prospect of potentially getting more or better land than at home.
In fact why bother launching it at all?
This is very cool and clearly represents a lot of work by the authors, but man, the idea of five kilometers-long coaxial shells rotating at different speeds that have to mechanically interface with each other, while still maintaining an airtight seal... does not inspire confidence. Kim Stanley Robinson's Aurora also threw some cold water on the idea of generation ships, for me. Who knows, maybe it's possible, but there are some reeeaal steep challenges to closed system design that needs to work for 50 - 100 years.
I'm picturing it as each shell is fully enclosed within the next. They don't need an airtight seal around something that lets air through, do they?
They don't need airtight seals between layers, but I think the end caps would need to be sealed individually since each shell rotates at a different speed.
Also, the interface between shells is pretty complex even if it's not airtight. To move between the shells, since they rotate at different speeds, there has to be a layer in between which can speed up or slow down to match speeds with them, which is a hell of a lot of mass to move on demand every time someone wants to run downstairs.
> To ensure habitat redundancy and high fault tolerance in the event of critical damage or system failure (power, mechanical, etc.), the habitat module is divided into 5 independent stages that provide all the survival systems for the population
Excuses; that's clearly Plot-Optimized for the formation of five distinct ideological factions, each controlling their own stage...
I know Yang usually wins, but I still call Dierdre. She might be one of the weakest leaders in practice, but no one can outdo her on style.
Related to that, check out the SMAC intro video. The spaceship is shown rotated 180 degrees as it arrives at the planet; naturally because it needed to decelerate halfway towards the destination.
A classy little detail that they never call out, that took me decades to notice.
That sounds exactly like the sort of paranoid fantasy that a Trilogist *would* come up with.
I left this comment on Hanania’s last race science post, trying to argue that the racial IQ gap is mostly not genetic. I was wondering if anyone had any feedback on whether it holds water. The first part is a quote from Hanania’s essay.
“So the idea that there couldn’t be cognitive or behavioral differences was always based on the at least implicit assumption that physical features change much faster than mental traits. I’m aware of no principle of evolution that indicates that is the case.”
All physical features don’t necessarily change faster than all mental traits. But you’re talking specifically about skin pigmentation and intelligence. In the case of pigmentation, the pathway from genotype to melanin production is quite direct. About five genes are responsible for the difference in skin tone between blacks and whites. Intelligence is quite heritable. But it’s massively polygenic. Hundreds of genes are involved, each one having only a modest effect. So even if an alien were to breed humans for IQ, it could take hundreds to thousands of generations. Also, the selection pressures involved in pigmentation are more specific and obvious. (More melanin, not burning; less melanin, more vitamin D). What selection pressures for intelligence would have applied to one region but not elsewhere, hundreds of generations ago?
You're probably correct about breeding for intelligence. And the alleles associated with cognitive performance in modern populations are also functionally tangled with alleles that seem to perform non-cognitive functions.
I don't subscribe to Hanania, so I couldn't rebut him in the comments. And now I can't find the essay, but I think he was referring to the latest paper by Ali Akbari et. al. "Ancient DNA reveals pervasive directional selection across West Eurasia." If so, Hanania seems to have garbled the conclusions. They found that, "in the past ten millennia, we find that many hundreds of alleles have been affected by strong directional selection. We also document one-standard-deviation changes on the scale of modern variation in combinations of alleles that today predict complex traits. This includes decreases in predicted body fat and schizophrenia, and increases in measures of cognitive performance..." Hanania seemed to latch onto cognitive performance, but if you read the paper, cognitive performance is also linked to alleles that control other things like metabolic performance. Akbari acknowledged feedback from Sasha Gusev—who has been critical of these GWAS-type studies (which tend to amplify weak signals). I haven't seen Gusev's response to this paper, but I'm hoping he discusses it soon, because I'll admit I don't have the statistical chops to understand these methodologies. Also, the authors admit that some of these effects may not be due to natural selection. The paper is pretty damn interesting, and I paid to get past the paywall.
BEER! Free alcohol dramatically disincentivizes "Angry Drunks." To the point that White Europeans have the concept of "Tard Strength" because most people have learned to "not go that strong, or get that angry" (it's adrenal strength, if the nickname isn't making it clear).
The height sanity check proposed by Scott is useful here. Height is highly polygenic, but if you bred tall people together it wouldn't take thousands of generations to get a tall population.
Sasha Gusev, who is a geneticist, says otherwise. In fact, he responded to Scott in the ACX comments. But then he wrote a couple of substack posts on the subject.
https://theinfinitesimal.substack.com/p/no-intelligence-is-not-like-height
https://theinfinitesimal.substack.com/p/comments-on-no-intelligence-is-not
a. This is not something you can reason out from first principles, it's an empirical question.
b. I have zero faith that Hanannia is approaching the question with a goal of determining the truth.
I don't see a justification for the theory that evolution of polygenic traits should be much slower. If those genes aren't mutually independent, they will work together and thus get promoted (or suppressed) together.
And the selective pressures for more and less melanin aren't that straighforward either. The Tasmanians were isolated on a foggy mountainous island with climate similar to Scandinavia for at least 12 thousand years and they were some of the blackest people on Earth. Meanwhile, the Thai and the Cambodians, who live almost on the Equator, are more pale than dark.
This is where deductive reasoning fails without a complete understanding of the system involved. Polygenic traits depend on multiple genes functioning together to influence the phenotypic expression. The issue is that individual genes in a polygenic "cluster" of genes may influence multiple functions. So selection for phenotype X may interfere with critical phenotype Y. Also, polygenic traits are generally highly multifactorial, so environmental factors (like diet or sunlight) interact with the genetic baseline, making phenotypic selection partly dependent on non-genetic variables.
Another issue is the liability threshold, where traits like psychiatric or complex conditions manifest only when a person crosses a cumulative statistical threshold. Also, shuffling and recombination across generations mean that a phenotype may appear in some generations and skip others. Shuffling also means offspring do not necessarily inherit the entire polygenic cluster. Finally, ineffective purging means that even if everyone with some extreme condition stopped reproducing, the vast reservoir of minor-effect alleles hidden in the general population ensures the trait persists.
Most (but not all) selective sweeps in modern human populations over the past 40 ky have been for metabolic functions (like lactose tolerance) that appeared with the advent of agriculture and herding. And most of those have been linked to a single Mendelian gene or a small cluster of genes.
Being polygenic doesn't mean it takes a long time to change. Animal breeders change polygenic traits all the time, and it can happen in a small number of generations. See how much larger chickens got over the 20th century, or how much more milk cows produce.
Yes, only 500 years ago domestic sheep were about the size of dogs (Labrador, not Chihuahua).
Our ancestors interbred from time to time with distantly-related hominids in various parts of the world. Even if selection pressures were the same (and surely cultural factors could change them as easily as climate, or more so) some groups might conceivably have an easier path towards increases in intelligence or some aspects of it, on account of genes acquired from these hominids.
You can’t and shouldn’t try to argue with Hanania about genetics. It’s the wrong framework. The firewall against racism and eugenics in our society was a philosophical and moral one. The blank slatists managed to make people forget this by absurdly demanding that we erase all difference, but that doesn’t make Hanania right. It just means they played in his sandbox despite not having the ammo to do so.
If someone’s moral worth turns on IQ, society becomes a nasty place very quicker. Back to Buck v. Bell and forced sterilization. The correct answer to the Hananias of the world is to say that a person’s moral worth doesn’t turn on the mean IQ of their group, it turns on their inherent dignity as a human being. Hanania avoids discussing this part.
There are plenty of positions between "mean IQ of [racial] group" and "inherent dignity". Even if you equate IQ with moral worth and think it's heritable, why not test the individual? And even if you care about group IQ because of regression to the mean, why be lazy and pick population groups based on clusters of appearance when you could filter by family, ancestral occupation, income percentile, or even pick some other protected characteristic like religion or ethnic origin within a race?
Because those are the groups about which there is political rhetoric demanding equal outcomes and attributing unequal outcomes to discrimination/racism/etc.
It's not that black/white/asian/hispanic cut nature at a joint, it's that when we talk about different outcomes by race in education, employment, income, arrests, crime victimization, unwed births, infant mortality, etc., the political discussion is about these groups. So if you want to be able to understand why those outcome differences exist, your analysis needs to also be about those groups.
You can make a good argument that these categories are unwieldy, but you can't start out by claiming that blacks are underrepresented in advanced math classes because of racism and then demand that nobody look at intelligence scores broken out by race as a possible explanation.
This is tautological, no? The groups are the groups because they're the groups.
My point isn't to do a bait and switch on you, it's that we could pick a whole range of things (height or weight might be a good example) and find a correlation with IQ and poverty, but we don't do that, so maybe we should recognise that using race in that way for either IQ or poverty is a special case and not productive.
At least until we end up in an AI dystopia that will let us Truman Show some kids and see how much being the dominant race in a society affects IQ.
So when someone complains that black students are in the advanced math class less often than white students, and attributes that to racism, how should we proceed?
I’m unsure if you agree with me or not.
I don't agree with the framing. At the moment one side is "people who think you should exclude people on the basis of the average IQ of their race" and they're up against "everyone who doesn't think that", which means they're outnumbered.
If we weaken the second group to just "people who don't believe in using IQ as a metric for anything" then you're framing it as an argument between two groups of extremists, instead of one group of extremists versus the entire rest of the world, which seems like a counterproductive way of seeing things.
Which side is arguing that you should exclude people based on the average IQ of their race? This seems like a strawman.
It is worth knowing about racial IQ and crime statistics, because this helps you understand many matters of social and political interest in the US. We should not suppress discussion of true things that help explain important parts of the world because that makes everyone dumber and leads to worse policies.
Part of reasoning about this is knowing the difference between an average and an individual drawn from a distribution, but it's not like this is some impossibly high bar.
I’m not an IQ hater, I’m fine with, for example, IQ testing in employment. My point is that even if we don’t deny human difference and diversity, which I don’t, the concept of human dignity imposes limits on what can be done to the left side of the bell curve. So we can’t hand out group punishments based on group IQ, because that fails to afford individuals an opportunity to be judged on their own merits. We can’t forcibly sterilize the retarded. And I would go so far as to say even a low IQ guy should have stable employment that allows him to support a family, even if the IQ test means the AI engineering job goes to someone else. Hanania agrees with none of this last part, and that’s why I think he’s contemptible.
It doesn't make sense to treat people based on their group IQ in any situation where you can just test them as individuals, which is most relevant situations. Like, the math teacher shouldn't assign kids to the fast/slow track based on their race, because she can just give them a test or look at their previous homework/grades and judge them individually
I think this is worth emphasizing. I feel like a lot of people have developed an implicit attitude that we have to avoid investigating any sort of average differences between races, sexes, whatever, because if people are allowed to investigate them, prejudice will inevitably work its way into the process and outcomes. But I think this is essentially ceding the whole field of discussion to the prejudiced, an expression of the attitude that "we can't maintain our position in the face of the facts." But I think that's a frankly rather embarrassing position to hold; it's not like we have to hang our adherence to basic principles of fairness on the belief that all groups are exactly the same.
I think it would be entirely reasonable for us all to agree to politely refuse to discuss racial IQ gaps if we could also agree to politely refuse to discuss racial wealth gaps.
But if some people insist on noticing gaps in things like wealth, income, health, education levels, incarceration rate and so forth, and insist that it must be due to people treating other people unfairly, then it becomes necessary for rationalp people to say "well actually..."
I think reality is frustratingly consistent. Racial differences in average IQ and crime rate are really important things to know in order to understand US society, and not knowing them puts you at a big disadvantage when trying to either understand things or make predictions of the future. So agreeing not to discuss them seems like a kind of losing proposition.
It's like if you wanted to think a lot about biology, but were constrained never to mention evolution. There would be a lot of stuff that would be really hard to understand or predict or think about. And the clued-in people who got evolution would be a lot better at understanding stuff.
But yeah, the situation where those facts are unspeakable but we can talk about differences in educational outcomes, incarceration, and the like and propose other explanations (it's all racism, say) is even worse. But if you discuss race and crime in the US without knowing the crime statistics, or race and education without knowing the IQ statistics, you are basically doomed to speak nonsense. And indeed, many people don't know those things, and sure enough, they speak a lot of nonsense.
You’re right insofar as blank slatists did empower Hanania, but you actually don’t have to play in that sandbox. You can just say “people who crime belong in jail, people who don’t do crime don’t belong in jail, and I don’t care what they look like.”
You can, but you shouldn't.
First, what defines "doing crime" wasn't handed down to humanity from on high on gemstone tablets. It's determined by the result of political processes engaged in by fallible, biased humans, most of whom have deeply ingrained cultural notions of things like race and gender and let it inform their decision-making.
Second, very, very few people actually believe "people who crime belong in jail" without qualification. Crimes like speeding and jaywalking are so common that sticking to this notion in principled fashion seems beyond impractical. But deciding where to draw the line of "crime that should put you in jail" vs "crime with a lesser penalty" vs "crime we're actually just going to ignore in almost all cases" is similarly the result of a political process engaged in by fallible, biased humans.
The upshot is, treating all humans in a fair an unbiased fashion that isn't skewed by silly, outdated cultural prejudices is quite a bit harder than it sounds. You definitely won't get there by pretending that those cultural prejudices don't exist, never existed, and don't have any effect on anyone's present-day circumstances or behavior.
This feels like derailing a discussion. Yes, it is true that different societies have different rules, with some taken more seriously than others and punished more harshly. But what does that add to this discussion?
Within the context of American laws, we have some things we consider serious crimes (approximately but not precisely captured by terms like "felony" or "violent crime," depending on context.
And there is substantial disagreement about how to deal with people who commit what we think of as fairly serious crimes--stuff like assault with a deadly weapon, armed robbery, sexual assault/rape, and the like. Of course, you can't capture a whole policy in a one-phrase slogan, but "people who do crime should do time" is not a terrible summary for the broad idea that people who commit those fairly serious crimes, when caught and shown to have done them, should generally face some pretty harsh consequences involving spending significant time in prison. And there are several opposing strategies, ranging from a very broad oppositon to sending people to prison until everything else has failed to some moneyball-like calculation of which people need more leniency and which need more time inside that prison cell.
> what defines "doing crime" wasn't handed down to humanity from on high on gemstone tablets
They were just regular stone as I understand things. But even if you don’t think the Old Testament is a good source of law, natural reason gets you to the same place. How many functional societies have legal murder? Legal theft?
> similarly the result of a political process engaged in by fallible, biased humans
So what? We have elections for this. Natural reason might dictate that theft be banned in a civilized society, but doesn’t tell me what the dollar number should be for a felony charge or how many strikes one ought to get. I’m fine with voting on this.
Who makes that argument?
People have some hard-to-change differences in intelligence. We have to adapt ourselves to that reality, as to all other unfortunate realities.
Like, no matter how hard we try to teach, no matter what stirring rhetoric about equality we spread, no matter what resources we make available, most human beings will never be able to understand or do useful work in physics. They aren't smart enough. This isn't a moral failing, it doesn't mean those people are useless or deserve any abuse, but still, most people just aren't up to it.
We should care about everyone, including people who got a bad roll of the IQ dice. We should try to make a decent life available for everyone. But that isn't accomplished by demanding that the dude with a 90 IQ somehow get through a calculus class, it's accomplished by making sure his education works to set him up for a worthwhile life where he can do useful things.
Has Hanania claimed that the moral worth of a person hinges on the mean IQ of their group?
For liberals vs conservatives, kinda yes.
As I said, he’s clever enough to avoid discussing that part, except insofar as he targets the same targets that all good liberals feel comfortable with - unborn babies with disabilities being aborted, the elderly being euthanized. But this is a guy who wrote a whole post about how he opposes eugenics because eugenics has a negative connotation. He’s not a different guy than the guy who got busted Hitlerposting. He just changed who he considers the untermenchen and started choosing his words more carefully.
So you're saying that instead of arguing about the things he says, we should instead accuse him of secretly thinking things that he's never said, and then argue with those things?
The guy was a literal Hitlerposter on Stormfront.
Too many intellectuals silently think of themselves as being more worthy than others because they are smarter, and they project this idea into others.
Good point.
> But it’s massively polygenic. Hundreds of genes are involved, each one having only a modest effect. So even if an alien were to breed humans for IQ, it could take hundreds to thousands of generations
Is this so? Does the number of generations required to breed in a trait depend on how polygenic it is? This doesn't sound like an unreasonable idea but I don't have a strong intuition about it, is there a good argument why it should be the case?
I can see that if there's only five genes for melanin and ten thousand genes for intelligence then you can get to a maximally-dark population faster than you can get to a maximally-intelligent population, but that's a different question to whether you can get a ~1 sigma gap between populations with very different lifestyles over hundreds of generations.
The selective pressures on a person living in a farming society are very different to the selective pressures on a person living in a hunter-gatherer society, and it's been about five hundred generations since some populations developed agriculture. If a certain trait gives you a 3% reproductive advantage, if you pass that trait on to 50% of your children, and if it's initially present in 3% of the population, then within 500 generations the people with that trait will be the vast majority of the population.
It isn't so, and Greg Cochran has to correct it regularly because people mistakenly claim it is. The breeder's equation doesn't depend on the number of genes, just the narrow-sense heritability and the selection differential.
Thanks, I was going to come here to say this. It depends on heritability, not the number of genes, AND you can select noticeably on pretty much any trait with above 15% heritability, which is most of them.
Although the breeder equation itself has some weaknesses when iterated over multiple generations. On IQ itself, if you plug in a simple "the top 10% assortatively mates for 8 generations," the 8th generation have mean IQ's of ~220. This seems *way* too high. We'd notice.
Sure, the Galton / Darwin / Wedgwoods, the Bohrs and Curies, all exist. Scott had a post about several (https://www.astralcodexten.com/p/secrets-of-the-great-families). But they're not ~200 IQ, and we don't really see that, and I think we probably do assort to at least the decile / quintile level, and have been for a long time.
What's the answer? It's not "regression to the mean," because the whole point of distinguished lineages is that your lineage mean is the one you regress to, not the population mean. So if your line kept pairing with similarly smart people from similar lineages, that's not a problem and your lineage mean would be ratcheting up.
And lineage means really ARE much stronger, empirically - see Greg Clark's "social competence" persisting at ~75% everywhere in the world, significantly higher than the actual intelligence and educational attainment h^2's of ~0.55 and ~0.4.
Ultimately I modeled it out and it's because the Breeder Equation assumes truncation selection, while in real life, the children of distinguished lineages are essentially all equally likely to have kids, AND most smart people have <=2 kids, and people match on phenotype rather than lineage / genotype, so you do cap out generationally at around your conjoined lineage mean in that situation.
If instead, smart lineages routinely had 4 or more kids, and those kids only matched with similarly smart people from similarly smart lineages, we actually would see some crazy lineage means increasing in line with the breeder's equation, with random +1SD draws on top of a 220+ baseline several generations in, to which yes please.
It's just apparently too hard to achieve that with people's natural matching ability - finding a spouse is already an exceptionally difficult problem people spend wayyyyy too much time and optimization power on, and even then fail at ~82% rates (the rate of "divorced or dead bedroom" 20 years in, in recent GSS data).
We would need societal structures and incentives to facilitate that degree of matching, and to persuade those families to have more kids, and we all know THAT will never happen.
Assortative mating is not the same as selective breeding. However, the breeder's equation itself depends on factors which are not necessarily constant over generations.
> However, the breeder's equation itself depends on factors which are not necessarily constant over generations.
Could you go into this more? I've been interested in these details, and you can't always trust LLM's.
The equation itself is so simple this has to either be "heritability decay" or "selection decay" right? What is going on when you compound those over generations that make it invalid?
It depends on variation, and selection itself can remove variation. The low-hanging fruit can be consumed, and new variants would need to be produced via de novo mutations. However, being polygenic can actually mean more runway as there are more genes to reach fixation.
The only case I've heard someone make for differential selection on intelligence between populations was regarding Ashkenazi intelligence: the case the person made was that in some traditional Jewish cultures, studying Talmud was essentially a subsidized activity where the most respected scholars received more money from the rest of the community, and had more children. I don't know how true that is historically, but it would at least hypothetically constitute such an environment with a positive selection pressure for intelligence which differs by ethnicity.
But, I'm also not aware of any obvious differential selection pressures for height between different human populations, and height is also strongly polygenic, and even in a modern environment where people are mostly getting complete nutrition, there do seem to be differences in average height between populations.
Personally, I don't think that we can reasonably assume that different populations are all sitting at around their genetic potential for average intelligence (it may be that *no* populations currently approach their genetic potential for intelligence in an optimally conducive environment.) But, acknowledging what a politically volatile matter it is, I don't think the possibility that there might be some difference in average genetic IQ between populations is inherently lacking in plausibility. As much as it would be better for our sense of intrinsic fairness if there were, there probably isn't any pressure *forcing* all populations to have exactly equal intelligence, and sometimes populations differ due to random drift.
Thanks for the thoughts🤔
> I'm also not aware of any obvious differential selection pressures for height between different human populations
Climate, resource availability, infectious disease burden, etc. in an environment can all affect the optimal height. It's still not extremely well understood, but I don't think it's correct to say there aren't strong differential selection pressures, just that those pressures are themselves very complicated. Here's a decent review article: https://doi.org/10.1111/brv.12165
The potential to become a top-tier NBA player and potentially impregnate an arbitrary number of women?
The strongest selective pressure right now is probably either desire to have lots of children or inability to remember to take your birth control pill/use a condom.
Being able to potentially impregnate a large number of women, within the context of human society, doesn't necessarily mean actually having large numbers of surviving offspring in fact. Plenty of traits that might be nice to have, either from a standpoint of personal happiness or partner desirability, don't necessarily result in positive selection pressure. Evolution doesn't care if you mate with a partner who's considered especially desirable, and are able to raise your kids in exceptional comfort, it doesn't make them any more represented in the gene pool. In a modern context, being very tall probably experiences less positive selection pressure (even just for men in particular) than being reckless or disliking birth control. Wilt Chamberlain is basically the archetype of an extremely promiscuous athlete, and he most likely didn't have any children at all.
> Wilt Chamberlain is basically the archetype of an extremely promiscuous athlete, and he most likely didn't have any children at all.
If you take his numbers at face value, Wilt Chamberlain essentially has to have many unacknowledged children if he was fertile.
12-18% of condom users fail and get pregnant each year empirically, and chance of pregnancy per sexual act with women in their early 20's ranges from 2-33% depending on ovulation.
20k is large enough we can just "law of large numbers" this. So 20k * 12-18% condom failure * 7% conjoined ovulating / not chances of pregnancy gets you 170 - 250 children. Even platonically perfect condom use gets you 29 kids.
And that's assuming reliable condom use over 20k encounters, many of which could have been drunk / high / partying, etc. Any instance where the condom wasn't used, or the woman claimed to be on birth control but wasn't, just brings those numbers up!
So if Wilt was fertile, he probably has at least a hundred, maybe 200+ kids. You can reduce any one of those parameters by a factor of ten, and he STILL likely has more kids than the great majority of men.
On the other hand, if you get knocked up by a famous (and rich) athlete, you are going to publicise that pregnancy because it's money in the bank (if you can get the guy to acknowledge paternity).
So if Chamberlain hasn't any recognised or attributed children, he probably was (a) careful about not being baby-trapped (b) possibly could have been infertile or low fertility:
https://en.wikipedia.org/wiki/Wilt_Chamberlain#Personal_life
"Chamberlain also said he never came close to marrying and had no intention of raising any children.
In 2015, a man named Aaron Levi claimed to be Chamberlain's son based on non-identifying papers from his adoption and information from his biological mother. Chamberlain's sister refused to provide DNA evidence for testing, so Levi's claim has not been conclusively proved."
For a guy who claimed to have slept with 20,000 women, only one allegation of "that's my pa!" is extreme carefulness or, more likely, infertility.
That's 12-18% failure rate for typical use in a year of having sex per couple, not per sec act. Wilt didn't just happen to not have kids through those encounters, he was resolute about wanting to avoid it. The 20000 figure was probably an exaggeration, but even if you have sex with thousands of partners, you can realistically have no children, if you preferentially select partners who also don't want kids, are on additional forms of birth control, are willing to get abortions , (particularly if you're willing to pressure them to,) etc.
If Wilt had had dozens of children, he'd probably have had plenty of women coming forward to claim him as the father of their children, but his biographer was unable to find any.
Social selection is also discussed in the article. But it's not clear that sexual selection always favored extremely tall males - the NBA is a new invention.
By a similar token though, might there not be complicated, not-well-understood differential pressures on intelligence?
I don't think that many of those environmental factors are likely to apply a strong, direct pressure on height in particular, but changes in height might result from adaptations to some of those pressures. As the abstract of the article you linked puts it-
>The inability to draw firm conclusions about the adaptiveness of such a highly visible and easily measured trait suggests we should show an appropriate degree of caution when dealing with other human traits in evolutionary perspective.
It's not easy to predict how a complex polygenic trait, even a highly visible one, will respond to differing environments where differences in selection pressure are not obvious. But, that doesn't mean we can confidently conclude that no differences will arise.
I remember feeling excited on first learning of the concept of human biodiversity, and saddened when I saw how the idea was actually put to use. By analogy with other biodiversity, it would not surprise me to find adaptation in specific aspects of human intelligence to environmental factors. But as long as we're just measuring IQ or "g", it's more like a one-axis Great Chain of Being than actual biodiversity, with the complex trade-offs and path dependence that term implies.
For (a very broad!) example, chimpanzees often perform better than humans on short-term memory tests. This is thought to be part of the price humans pay for complex language abilities.
In a much more subtle way, I could see Arctic hunters evolving to have a slightly different suite of associative and meta-cognitive equipment than tropical foragers do. Realistically, though, I can't see how that evolved difference could be disentangled from the culture that it exists within.
This is the kind of thing that, say, Razib Khan or Greg Cochran talk about pretty frequently, right?
Reasonable, though I would say that "broader-built bodies preserve heat better" and "larger bodies require more food" are still much more straightforward than anything we can say about intelligence. I think it's unfortunate that research into the genetics of intelligence has been stalled so much by political taboos - no problem has ever been solved by refusing to think about it.
Hundreds to thousands of generations to do what, exactly? Achieve a 1-point change in mean population IQ? 5 points? 100 points?
The current gap in the US right now is 10-15 points, I think. 6-8 points would be a little more than half. So 8 points?
The gap has narrowed. From 1972-2002 the Flynn Effect for blacks averaged +4-7 pts per decade, while the Flynn Effect for whites chugged along at a steady +3 pts per decade. The Flynn Effect for all Americans stopped growning around 2010. I don't know whether it's still happening for blacks, but the IQ gap has narrowed.
https://journals.sagepub.com/doi/abs/10.1111/j.1467-9280.2006.01802.x
ISTR around 12 points now, but I'm not an expert.
How much of this was removing literacy from the equation?
For those confused: "illiterate peasants reading the bible" is a coherent, sensible sentence. Literacy was the first concept to be devalued (in terms of inflating educational results to mean something that "everyone can do!").
Read the paper that I linked. Education has been improving for blacks, and better education raises IQ scores. If you want, I can dump at least half a dozen links to studies that support this idea. Also, given that median IQs of Americans rose by 3 pts per decade from the ~1910s and ~2010s, the Americans of the nineteen-tens would have a median IQ of what Lynn (erroneously) claimed Nigerians had. Meanwhile, African IQs have also risen due to the Flynn effect (and it's a worldwide phenomenon). While no one is doing systematic IQ surveys of African nations, some data I saw indicates that he Nigerian median IQ is half an SD below US IQs.
Okay, taking the WISC:
Wisc-R: Changes from WISC: Updated norms to reflect the Flynn effect, revised items for cultural and gender fairness, and expanded age range.
Wisc-III:Changes: More detailed cognitive profiling, updated norms, and refined subtests for fairness.
Wisc-IV: Changes: Updated item content for cultural and socioeconomic fairness, enhanced sensitivity to learning disabilities, attention disorders, and executive function, and revised scoring to align with contemporary cognitive models
You're not looking at the same test. And there's explicit "updated norms" and cultural fairness (aka removing literacy -- I would not be as upset if we were simply adding in more questions to include AfricanAmerican/Hispanic literacy -- wild to have some questions about "common knowledge" about luchadors).
Luckily, the DoD is here to save the day (I'm much less concerned about the differences between
Quoting from the paper you linked:
"The constancy of the black/white IQ gap is a myth. Blacks have gained 5 or 6 IQ points on whites over the last 30 years. Neither changes in the ancestry of those classified as black nor changes in those who identify as black can explain more than a small fraction of this gain."
Probably true (given leaded paint and gasoline), but a paper that takes it for granted that children under age 15 are pretty much all available in the school system... I'm not sure how much that's the case, and if so, you might have selection bias. [I tried a brief lookup for Baltimore, but was only getting aggregate data.]
Flagging " By our estimates, blacks lose about 10 points on whites between those ages (the rate slows down after age 24). " Those ages being 9 and 29. I think this is worth pointing out, and then pointing out again. If you're looking at IQ, and you're seeing it degrade (I'm certain it degrades in white Americans as well -- there'll be a cultural component of this, we don't exactly force people to answer riddles to survive...unlike Yemen) relative to other people, that may indicate cognitive difficulties that only show up in advanced schooling (above age 9). Noting that black children are roughly two years ahead of white children in hitting adolescence, hormones might be a factor (particularly if expression of such impedes learning/improving g).
Also: of course people are doing systematic IQ studies. They're important. They're not publishing the data, because academics has declared investigating the subject to be rather verboten. But we do know who the stupidest group of people in the world is (someone had to be!) just like we know who the most inbred group of people in the world is (someone had to be!) -- they're not the same people, believe it or not.
Aliens could do this in one generation.
(Or in zero generations, by just drawing a circle around a group of people with an IQ higher than average.)
For any Homer fans out there, or if you just want to know more background on the Odyssey, I just posted a podcast episode interviewing a great storyteller who's been retelling Homer's myths for most of his life. Really fun conversation.
https://joelmckinnon.substack.com/p/full-transcript-jeff-wright-on-homer?r=zchpp&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
Reminder that the promise of an AI-powered utopia is at best a minority position, at worst a smoke screen. Money talks, and it says "anything but UBI".
https://forwardfuture.com/newsletter/originals/gina-raimondo-says-basic-income-would-end-america-her-new-billion-dollar-org-is-how-she-plans-to-sto
>Joe Biden’s former commerce secretary Gina Raimondo has decided that UBI in response to AI is the thing to fear most. At the Reagan National Economic Forum she said: "I personally think it's like the end of America."
> She is now heading up a newly-launched extremely well-funded organization to make sure the country reaches for anything but a basic income in response to AI. It's called RAISE US
> RAISE US launched June 25, with Raimondo as CEO and former Indiana governor Eric Holcomb as co-chair. It has already secured more than $500 million toward a $1 billion target. Amazon, Anthropic, Microsoft, and the OpenAI Foundation are anchor partners.
I think it really depends if we are in the world where ai/automation actually obviates most human labor, versus the one where ai merely disrupts a lot of peoples careers and they form a coalition to vote themselves money because they cannot swallow the idea of being in the menial class
So my question is, what happens to the US financial system if millions of americans suddenly cant pay their mortgages and have no hope of finding a job that would pay them enough to cover it? The US currently has over 13 trillion dollars in mortgage debt; even if a small fraction of that goes bad you have trouble. If mortgage service run into a liquidity crisis it will be larger than the government can bail out. You would be looking at multi-year forbearance or foreclosure moratoriums. Not that that solves the problem but I don't know what else would be politically tenable.
You're saying lots of low-signal things like "minority position" and "Gina Raimondo has decided". If you think UBI is bad, then argue for that point.
Personally, I'm up in the air on how important work is to personal meaning versus all the other things we would do if all work were automated. Sure, people who don't work now seem to have problems, but the direction of causation is unclear. It's also all in the context of a society where most people work, and the implication is that there's something wrong with you if you don't. That wouldn't be the case in a world where there's little work. So we just don't know how much work matters - we're mostly all repeating our priors.
I'm neither arguing for or against UBI. I'm pro UBI, but that's not the point I'm making here. The point I'm making is that AI companies, contrary to the viewpoints Scott is boosting frequently, there will not be UBI if the current AI companies have a say in it.
Everything below the link are quotes from the article, by the way.
Have you ever heard the expression “retired at birth”? Children of multimillionaires are in effect born retired — they do not HAVE to work, ever. And some of them do in fact spend their entire life partying. Yet most of such lucky individuals find some pursuit, and often become quite good at it. Which is entirely understandable, because their work is what they WANT to do, not what they are FORCED to do.
Imagine every human being free from drudgery and from the pressure “to make a living”, free to pursue whatever they want to do, however impractical. I consider it a future to strive for, not to fear.
As an aside, I don't think EngineOfCreation thinks UBI is bad, quite the opposite.
Unfortunately we have evidence on this and the reality is that most use the extra time to watch TV, scroll social media, and sleep. I do think some, for whatever reason, are more motivated to use that extra time for what we might call healthy activities like social connection, hobbies, exercise, volunteering, etc.
Mmm, some multimillionaires make it clear that they're not going to simply hand over the megabucks to Junior, so Junior better find some kind of job, even if it's a sinecure:
https://observer.com/2026/08/billionaires-arent-leaving-fortune-to-children/
True, but not especially relevant. These particular multimillionaires chose not to “retire at birth” their offspring. That does not affect the choices of the offspring who were retired at birth.
It is tempting to think that they would be like 19th century English parsons, and quietly work on some sort of groundbreaking scientific discovery, but I bet that most people would very quickly go to seed.
How many people "going to seed" would you find acceptable if you got one additional generational talent in science, business, sports, art, etc. in return?
This is to be wished for, but also getting there from our culture will be hard and we should understand that just giving everyone an UBI sufficient to live on for life will probably have all kinds of surprising and often unpleasant social effects. Some people will become poets and artists and such, but a lot will just drink and f--k. (OTOH, maybe that solves our falling TFR problem....)
The citizens of oil sheikhdoms in the Gulf live quite sheltered and financially secured lives, with servants catering to their needs and qualified foreigners doing most of the intellectual work. This is about as close to UBI for citizens (again, citizens, not the general population) that we have right now.
That region is also famously intellectually and artistically sterile. Measured by creativity, UAE, Saudi Arabia, Kuwait, Qatar etc. barely register on the global ladder.
UAE, Kuwait and Qatar have funnily low populations (I mean citizens), and while Saudi Arabia has cca. 10M citizens, the total number of people you listed is less than 15M. Not sure what conclusions we might draw from such a sample.
(Another question is, which countries (or, more like, cultures) can produce something which the world sees as "creative"? I mean, if you are not from an Anglo-Saxon country, nor from Germany or France, who cares what you think or do? But I guess you get the idea anyway.)
Small populations are absolutely capable of producing outstanding talents.
Ancient Athens had at most 160k people and we still know their philosophers and artists by their name and work.
18th/19th century Scotland had maybe 2 million people and produced maybe half of all the scientists and engineers who pulled the Industrial Revolution forward.
How many Ashkenazi Jews are out there and how many of them got a Nobel Prize?
I have to agree with DrMcleod. Islam effectively prohibits representative art, and its philosophy of fatalism ("God willing...") discourages pursuit of the unknown. It really is a religion hostile to arts and sciences.
How much is that linked to their artistically and intellectually sterile religion though?
I don't disagree with your point about rich kids, but I don't think it settles the question. They generally have high social capital and are probably at least as intrinsically motivated as the median person. I wonder what will happen to the bottom quartile in motivation. Maybe it's fine if a bigger share of the population ends up spending most of their lives partying, gaming, or whatever, but it's a big leap from how society works now.
When I think about UBI, I think about the Australian Aboriginal communities where everyone is on welfare and nobody does any work. And everyone just sits around getting high off whatever drugs they can get their hands on.
Realistically, do these Australian Aboriginals have any other options? If they wanted to learn things toy robotics or scuba diving, is it physically possible for them?
I mean, in 2026, I don't think anyone with free time and basic resources met has a reason why they can not study drawing, philosophy, learn a new language, etc.
What's the alternative? Starvation? Makework? Bomb the datacenters?
There will always be a percentage of people who abuse any given social security net. But if that net, for every 10, 50, or even 100 such people enables a new Einstein or Freddy Mercury or what have you, it will have been worth it.
Name me a single Einstein or Mercury who comes from, say, Saudi Arabia, or the circles of the Russian nouveau riches.
I live in Europe. The economic differences here are vast. Switzerland is much richer than Romania or Lithuania. But the Swiss don't seem to be particularly more creative than Romanians or Lithuanians are; what they can do, is to attract brains and talents from the outside, but quite a lot of those people grew up in some backward region of Eastern Europe in a dreary block of flats.
You named Freddie Mercury. As a teenager, Mercury actually had to work for money, being a stall seller and a baggage handler on an airport. He certainly didn't have a generous time- or financial budget for his musical training. It was mostly raw talent that got him through.
I guess I'm much more pessimistic about human nature than you. Maybe just projecting my own terminal laziness, but I honestly think society needs to compel people to work (economic and social pressure) until we can automate most jobs. Most people hate their jobs, because the venn diagram of jobs people want to do and jobs that need to be done don't overlap much.
The whole idea is that with enough automation, not everyone has to work. Forcing everyone to work as before just because some might take advantage just doesnt make sense. "The optimal amount of fraud is non-zero", and the optimal amount of lazy freeloaders is also non-zero.
It is a strange error, yet somehow a common one, to say "people intrinsically value working because it's an identity marker / makes them feel useful, so we shouldn't do UBI because it will take that away!" It seems to me that the initial premise implies people will just *keep working* if you give them a basic income. Which is what they mostly do, according to the results of the UBI pilot projects I've read about.
I do not think it is an error. I think it is a deliberate lie. I would go so far as to say that absolutely everyone who tells you “You need work for life to have meaning” (Jordan Peterson comes to mind) has vested interest in keeping peons busy.
In these discussions people seem to conflate "work" (any productive activity) with "job" (work done for a stranger, in return for money). I find it plausible that work is necessary for mental health, but that doesn't imply that jobs are. You could work on your hobby project, or do something useful for your family, or your community -- this all is perfectly compatible with UBI.
Without UBI, you get billions of suddenly not-so-important people desperately competing for the few remaining jobs that are not done better by AI (including robots) yet, and if they fail their families will starve. If you are the right kind of of person, you may enjoy watching this. Most people won't be happy about that.
Longer version: https://www.lesswrong.com/posts/qRZLEBmNtT6LBuFsE/it-s-nice-of-you-to-worry-about-me-but-i-really-do-have-a
It is a classic Motte and bailey fallacy: https://en.wikipedia.org/wiki/Motte-and-bailey_fallacy
Take something uncontroversial (people have a need to feel useful and productive) and conflate it with one much harder to defend (the only way to feel useful and productive is to be wage slave in Neoliberal capitalism). It is never a mistake, it is always a deliberate lie.
Love your LessWrong article! I always said more or less same thing -- and I say it as someone who for most of my adult life DID enjoy my jobs.
'Wage slave' is an oxymoron. If you're going to lecture people about fallacies, at least try not to use them in the very sentence where you're doing so.
https://www.merriam-webster.com/dictionary/wage%20slave
"wage slave
noun
: someone who is dependent upon wages to make a living often with little or no opportunity for advancement"
Thanks. Another one of those data points where I can’t distinguish between The Onion and reality
Have you ever noticed that if you spend a while browsing The Onion even WSJ headlines appear satirical for a while when you switch?
I wrote a post about the history of toilets, covering 5000 year old plumbing in the ancient city of Uruk, the very public latrines of Rome, the invention of the modern flush toilet in the 1500s to the present day. Check it out if you're interested! https://postalpigeon.substack.com/p/the-history-of-hygiene-1-the-toilet
I'll read this next time I'm on the toilet.
You'll probably love this skit.
https://youtu.be/K-cYWMZX59A?si=2B0MVrir7_Lu8nfU
And it was smelly since the U-bend wasn't invented for another 100 years, so you'd basically be sitting right above the sewers!
Sometimes people argue that things like bicycles or wheeled suitcases or steam engines could have and should have been invented a lot earlier and the people of the past were nitwits for not inventing them.
Others counter-argue that actually they couldn't and shouldn't have been, because of all the other technologies that they're dependent on, and the people of the past weren't nitwits after all.
But if you're telling me we had smelly toilets for a hundred years before someone invented the U-bend, then that's a much better example.
Nothing beats Seattle's fountaining toilets. Sometimes with people on them when they'd erupt.
Yikes.
And this. What a cool religion. :) See section on "Philosophy".
https://en.wikipedia.org/wiki/Dudeism
Well, that’s just like your opinion, man.
https://m.youtube.com/watch?v=4LGX8TbvGew&ra=m
Crazy things happening in biotech. There are multiple ongoing human trials of a technique where AAVs are used to genetically engineer neurons into expressing designer receptors (DREADDs), which are extraordinarily sensitive to a common antipsychotic, clozapine. Very low doses of clozapine can then be used to activate these receptors.
This is the sort of clever experimental technique that can be (and has been) used in research, and it is wild to see it move into human patients. Hope it is successful. It may pave the way for many similar clever techniques in the future.
Writeup from Bryan Roth, a researcher that worked on early DREADD development: https://zendudest.substack.com/p/from-yeast-to-humans
Chemical & Engineering News article: https://cen.acs.org/biological-chemistry/biotechnology/human-trial-chemogenetic-brain-therapy/104/web/2026/08
If you want to read about some more crazy stuff being proposed, I recommend this: https://www.science.org/content/blog-post/taking-tip-plants-eyes
If you want to read about AAVs going terribly wrong to sober up from the hype, I recommend this: https://www.science.org/content/article/exclusive-death-girl-chinese-gene-editing-trial-was-never-made-public
Thanks. Interesting links. I had heard about the China fuckup (but not the details).
Yeah, super interesting article.
Count me in as skeptical it could reach that many neurons too - the whole problem with "non embryo gene editing" is how you get the edits into the trillions of cells in the human body, and AAV's don't have a great record on that front. Immune response is exactly the problem.
And I'm by no means keeping up with the cutting edge, but I thought of lipid nanoparticles, AAV's, and lentiviruses, lipid nanoparticles were the only ones with a good track record of success (Transthyretin Amyloidosis), and that they've even engineered some that can pass the blood brain barrier. And even most of the successes have only worked in small, localized organs, the brain would be one of the hardest ones to do, due to the sheer number of cells.
And obviously "getting the virus into trillions of cells with high incidence" is in tension with "inspiring an immune response."
Overall a sad story, and I'm on the fence whether Qiu should be stomped on. On the one hand, this is how you end up at the US, where no progress happens, the FDA kills way more people than it saves, and more and more human trials and pharma development are getting outsourced to China. The only way you can keep progressing and doing better is letting guys like Qiu keep pushing the frontier.
On the other hand, there actually could have been skullduggery there, pace the neuron chart and our priors that getting edits into most neurons would be very difficult. So I think that's ultimately where I'd land as the crux - if there was skullduggery, stomp, if not, we need to let the guy cook, and the parents signed all the releases and are clearly just angry in retrospect it didn't work out. Although they suffered a great tragedy, it shouldn't be allowed to rein in medical progress as a whole for the entire world.
All the "the financial distributions were irregular" stuff is mostly handwaving, that's just how stuff is done in China.
Also, as an aside, those "ipads and cases of Moutai" the dad was regularly giving Qiu? Moutai is like $1k+ USD a bottle if you're getting the good stuff.
Rationalists by Balenciaga!
https://x.com/curious_specter/status/2089371784311103923?s=20
How many rationalists can you identify? (Rationalist/EA/Adjacent)
Yes, I spent way too much time on this.
"If this thing you're saying is so important, why aren't you wearing cool sci-fi style gothy designer clothing while saying it?"
Really nice. Truly, Balenciaga is the unit of caring about fashion.
Haha I was trying to think of a way to get a "unit of caring" pun into it.
Why is the US still subsidizing corn ethanol when instead subsidizing solar farms (on the exact same acreage) could power the entire US, with zero impact on food supply?
(An acre of utility-scale solar generally produces roughly 20–30× as much usable energy per year as an acre of corn grown for ethanol.)
Probably the same reason why the Jones Act is still around even though noone with a brain thinks it's a good idea.
How many kilowatt-hours of energy can an acre of utility scale solar produce in one year?
Ask an LLM. I ran your question by Grok and ChatGPT, and they gave slightly different answers, but they're both in the same ballpark.
At the local level, many farmers (or farm families) want to keep farming, even as productivity gains and changing consumer preferences reduce the number of farmers needed. Interventions like ethanol subsidies and "Got Milk?" help forestall the decline.
Meanwhile, solar/wind/transmission projects generate a surprising amount of opposition from rural residents. My guess is it's mostly rural aesthetics: green = pastoral/natural/normal = good; new construction = bad, even when it would actually help farmers (by letting them sell/lease land for more than they earn by farming it). E.g., see this video by Wisconsin's Republican gubernatorial candidate: https://www.facebook.com/TomTiffanyWI/videos/we-cant-afford-to-lose-more-wisconsin-farmland-to-wind-solar-and-data-center-dev/1687835632443762/
Unrelated, a cool use of ethanol is growing the food to feed your own ~horses~ tractors: https://www.farmprogress.com/week-agribusiness/john-deere-unveils-e98-ethanol-powered-tractor-prototype
The bits of the US that are used to grow corn are good agricultural land that could be used for pretty much any crop. If the total amount of agricultural land in the US decreases then it's going to be the marginal land that goes away, not Iowa. Stop subsidising corn in particular and some of that land gets turned over to more economically sensible uses, which might be wheat, or tomatoes, or tulips, or whatever.
And if you're going to tile a state with solar panels then you'd pick Arizona, not Iowa.
The beauty of subsidies is that the market can allocate the land efficiently. If cutting corn subsidies and creating solar panel subsidies means more tomatoes in Iowa while the solar power comes from Arizona, that's fine! I assume the grid is capable of selling power from Arizona to other states.
Sounds like this would be a loss for corn farmers in Iowa and Nebraska, so we shouldn't be surprised if they use their political power to keep the subsidies flowing.
The United States doesn't subsidize corn ethanol because it is an efficient way to power any part of the United States. The United States subsidizes corn ethanol because farmers vote, and are sympathetic enough that other people will vote with them if it doesn't cost them too much. There doesn't seem to be a similarly large and sympathetic population ready to vote for solar-farm subsidies.
And also because farmers get a disproportionate say because of how our elections work.
Although these days most agricultural subsidies end up in the pockets of big corporations, not mom-and-pop farms.
Wildly untrue. Having spent the past decade working directly with the families who own the great majority of Midwestern farmland, I've been disabused of several urbanite myths and that's one of them.
What do you mean? I don't think the amount of money going to small farms vs big farms is a question that you answer by spending time on farms, but rather adding up numbers in a spreadsheet.
I'm not seeing either of you support your arguments with data. Shame on both of you. But with the help of an LLM, I quickly found out that Paul Botts is correct. According to the USDA...
"In 2024, 97% of farms were classified as family farms, while only 3% were nonfamily farms. In 2024, family farms collectively operated roughly 91% of U.S. agricultural acreage (the nonfamily share was 8.9%)"
But many farms rent their land from other owners. Looks like roughly 40% is rented from other entities, including big corporations (and Bill Gates owns about a quarter million acres). But something like 70% of that 40% is owned by small LLCs and family trusts.
On topics for which the facts are readily obtainable by anyone genuinely interested, my standard approach now is not to engage until/unless the person being absolutely certain offers some of those facts. Which they rarely do, so on second encounter my mute list gets a new addition.
To be clear I do not defend this approach which younger me would have been appalled by; I accept the shaming. I am though now old and cranky and life’s just too damned short to engage with the empty hand-waving.
So, Beowulf, I appreciate you going in search of some facts, it’s refreshing.
No doubt what the LLMs are mainly scraping is the USDA’s nationwide Census of Agriculture, conducted afresh every five years for over half a century now and the gold-standard such among all developed nations. I’d never heard of it until a decade ago when the farmers and farmer trade orgs I began working with explained that it’s their bible for business-planning purposes. It is a treasure trove of interesting information and this born-and-raised city boy has learned much from it. The current administration seems not to have yet noticed it and so far as I can tell it is still operating without politicization. Well worth the time:
https://www.nass.usda.gov/AgCensus/
Anyway to some specifics:
-- the states in which I work directly with farmers and farm orgs (Illinois and Indiana) happen to rank very high for leased farmland. When you look it up, it mostly turns out that Smith Farms LLC (they all have LLCs nowadays) is renting from the widow Susie Jones who’s 83 years old and grew up on the farm that she’s now leasing out while living out her days in a retirement village in Peoria or someplace.
Then sometimes the guys/couples/extended families currently farming corn or soybeans or wheat are leasing directly from each other for, reasons. When you meet a new operator his self-description is always something like, “Well I’ve got [“got” means he both owns and farms it] 1,500 acres in ____ County, then my brother and I are leasing 1,000 from [a local widow Jones] over by ____, and next year so-and-so from [neighboring county] is going to start leasing that 450 acres we still have from that old swap our granddad did cause he wants to try a new variety of [crop]….” There’s also now a whole layer of ag-land brokers in small-town offices across the Farm Belt, who sometimes act as realtors, sometimes as rental agents, sometimes as “farm managers” of land that a widow Jones doesn’t have a tenant for. Those guys always turn out to have grown up on a farm someplace (nobody would let them handle land if they hadn’t), and are a strong risk-aversion influence (only willing to do the tried-and-true as far as crop choices).
-- another way to express the Bill Gates stat is that he owns .000286 of the nation’s ~875 _million_ acres of active agricultural land.
-- regarding farm subsidies, I have learned that the definition of what counts as a “subsidy” is contested. But taking a broad definition (any federal dollars paid to farmers for any reason), they usually amount to between 8 and 15 percent of farm revenues nationally. That percentage has occasional big spikes of which 2020 was one, but aside from that year the annual average over the past 15 years is around 10 percent. (Standard reminder that “revenues” is not “profits”.)
Nearly all farm-subsidy dollars end up going to growers of bulk staple crops: corn/soybeans/wheat/cotton/rice/peanuts. Since those crops are most effectively grown for market on larger fields/farms, those doing specialty crops or hobby farms or fruit orchards or whatever receive little from the various subsidy-type programs. A 2,000 acre farm (very normal now in the Midwest) is more than three square miles. These facts lead to sound bites about how “the largest 10 percent of farms [or in the NPR argot “agribusinesses”] receive 75 percent of the subsidies”. Most of those though are still family-owned, just bigger operations than what our great-grandparents were used to. Institutional investors defined broadly own around 5 percent of US ag land of all types and an even smaller fraction of row-cropped land, and correspondingly receive small fractions of farm subsidy dollars.
What? Number of farms is not the same thing as amount of subsidy, and LLMs are not a source. An actual source, addressing the actual question, would be something like https://www.aei.org/research-products/report/who-receives-crop-insurance-subsidy-benefits/, which says:
> Crop insurance subsidies are consistently heavily concentrated on the largest farms. The largest 10 percent of farms receive 56.4 percent, and the largest 5 percent receive 36.4 percent, of all crop insurance subsidies.
> Crop insurance subsidies per acre are larger for larger operations.
Batteries. Also national security, but who cares about that?
Spain shows that 'unregulated power sources' can cause the entire system to go down. That means riots, in case that's not clear.
Is unregulated power sources here a synonym for renewables. What caused the Spanish issue was bad grid management. Traditional turbines are more stable, but instability in solar generations can be compensated for. Otherwise Spain has done well from renewables.
Solar cannot fail! It can only BE failed!
Yes, this is what you're saying, by absolving Spain's solar production of any responsibility whatSOever in actually regulating itself.
It's only a synonym for wind, solar, and various unviable economic activities that people are terming "renewable." Water-powered renewables are fine, as is geothermic, as is nuclear (thanks France! Road To Zero!).
Don’t tell me what I am trying to say. I am saying what I said - the grid needs better management when there are more renewables. The Spanish investigation is concentrating on why stabilisation failed, but none of this means that solar is more dangerous if managed. Educated as I am I engineering I don’t see this problem as unsolvable, therefore it will be solved. In the mean time I tend not to waste facts on an ideological argument, it never works. Luckily there’s a mute button.
> Luckily there’s a mute button.
Only works until Aeolian321, Aristocat, Gaspacho, Gawdflea, GrimMoar, November Dawn, SkyShibam, Waze Kaze, Wimbli, Zanni creates yet another account.
You are arguing with He Whose Name is Legion, BTW
Hello, I've noticed a few low value/high temperature posts from you.
This was an opportunity for you to gracefully admit you didn't understand what caused the Spanish grid outage but instead doubled down without data.
Aeolian321 is yet another account from the previously banned Wimbli/Zanni/Gawdflea/Grimmoar/many, many others. This is par for the course; I wouldn’t hold out much hope for fruitful engagement.
Ah, thanks, if this is a trend then no point.
I've seen a number of people improve commenting etiquette and quality dramatically over time in this community after having previously been in more toxic bits of the internet, so it can happen!
I can cite you the report I was reading from Spain (and will, when I get a chance). It's public, and a "what went wrong." In it, I did not find any "the renewables did wrong" (and, perhaps, that is in fact how the system is set up). But, in the case where you have an unregulated bull running around without a fence, and the clown fails to keep him away from the audience, I'd say there's at least a little fault that should be given to the people who said "unregulated bulls are fine."
I'm genuinely curious what kind of draw you have to this comment space that you go through so much effort to keep coming back.
I hate to give a pretty quippy response, but I think the obvious answer is "regulatory capture".
I have no insight into that question directly. But I note that even if electricity was cheap and bountiful, the US would still need fuel for the non-electrified fleet of vehicles and more. You do mix in ethanol in your fuel over there too right? (Writing in Sweden)
An even more interesting prospect is why not both? Agrivoltaics is becoming all the rage I hear. Many crops even benefit from the partial shade (don't know about corn).
Most US gasoline sold is 10% ethanol, recently bumped up to 15% "temporarily" due to the Hormuz situation.
I wrote a UNSONG-style biblical exegesis, on why the meme song "Rockefeller Street" is actually an allegory for Jesus descent into and Harrowing of Hell; very much inspired by how Scott connects the song "American Pie" to Christian soteriology here: https://unsongbook.com/interlude-%D7%9E-miss-american-pie/
If you like that sort of thing, it might be fun to read: https://gist.github.com/Aransentin/6cc59786f08f66091f424abdd68638fc
A pretty interesting exercise in any case as a way of calibrating ones levels of apophenia; i.e. you develop somewhat of a better feel of what baseline spurious connections one can do when there's really nothing there.
This is tangential but I guess I'll use this as the UNSONG thread. I recently went to an event celebrating Kate Bush and was listening through her entire catalog, and she was pretty eclectic and just wrote about whatever she wanted, but even so it took a minute for me to realize I was hearing the lyrics to "Lily" correctly:
https://genius.com/Kate-bush-lily-lyrics
Existential Risk from a left wing perspective: "World ends, White Men No Longer Have To Do *ANYTHING*."
less of this, please
I couldn't agree more!
If someone thinks life post-singularity will be unimaginably good (for example, containing engineered pleasures beyond anything unaugmented humans could ever experience, and potentially lasting for an unimaginably long time because of immortality technology), should one make increasing the chances of living long enough to experience it the sole motivation of their current life, because of how much the benefits of this possible future outweigh any present benefits in life? For example, should a person try to minimize the amount of times they drive places or fly on an airplane to minimize the chance of dying in a transport accident, despite this avoidance making their life less enjoyable in the present? To what extent would you go to maximize your chances of long-term survival if you think there is something overwhelmingly good to survive for?
This is Pascal's wager, for the modern age, right?
This is different.
Pascal Wager depends on the fact that believing in the Christian God has no downsides (and that's where he gets it wrong). But but driving, not flying and otherwise restricting your activities comes with considerable downsides for which you might never see the payoff.
Absolutely not. Pascal's wager is between a few sinful pleasures in this life and an infinite reward; I think that the analogy is basically exact. At most, you substitute contingency tables for the authority of the church on what is to be avoided.
That's not the only problem with Pascal's Wager; another problem is that it considers only two possibilities -- Christian God vs. no gods -- when in reality the number of possibilities is virtually infinite. For example, maybe there is some cosmic supernatural force that will shunt your consciousness to a Utopian afterlife, but if and only if you wear predominantly blue clothing on alternate Tuesdays. Or maybe it's Wednesdays. Or maybe it's red clothing. Or...
This was my first reaction too, and I discarded it just as readily.
I don't see how this is connected to Pascal's wager at all. Pascal's wager is about uncertainty, but this scenario does not specify any uncertainty about post-singularity outcomes at all.
"thinks life post-singularity will be unimaginably good" contains, IMO, two statements of faith. 1 a value judgment of what life post-singularity will be and 2 that singularity will happen and that there will be life post singularity.
Even if these were "statements of faith" (which they're clearly not; it mostly sounds like you're using "faith" for "incorrect belief" which is not a useful use of the word), that wouldn't connect it to Pascal's wager, in which the firm belief does not exist.
I'm using the term faith to indicate beliefs or assertions that are not provable.
EDITED TO ADD: "the term"
I suppose it is rather Pascalian, but I have genuinely been coming around to accepting a greater degree of Pascal-style reasoning rather than thinking "that's just Pascal's Wager, disregarded." I blame Bentham's Bulldog.
I'm not suggesting that it can be disregarded, but rather that your question falls into the same type of trap, which is well studied in mathematics: when should I pay a finite cost now for a future reward which has unbounded expected value?
This was really what Pascal asked; and as he had an agenda, his answer was that the infinite reward beats the finite cost every time, as long as you assign a non-zero value to the probability of the existence of God.
I guess your problem still exists in subjective time, which maybe gives you a way out: choose a discounting rate, a probability of future bliss, a value for that bliss *and a probability that you are one of the elect that gets to experience that bliss*, and compare that to the cost of the experiences you forgo now. Of course that's all very subjective.
How different is this to the type of clean living Pascal would anyway have wagered on? Air crashes are famously rarer than we think they are; the interventions most likely to extend your life are clean diet, exercise and etc.
One should never have a sole motivation in life, especially not for a nebulous concept based on the work of other people. It's the whole "eggs in one basket" thing. Imagine you give up all those things, and then the post-singularity is worse than now.
"explaining AI existential risk from AI from a left wing perspective"
I think you have AI in that sentence an extra time.
Aye aye, captain.
"I like AI!"
The deadline to apply to the AI bootcamp seems to have been yesterday. Are they extending it?
Which AI bootcamp.are you referring to?
You boot an AI up, and it puts you into a camp. Or something like that.
:)
🤣🤣🤣🤣
The one linked at the top of the post, https://aisb.dev/sf26/