Parking space requirements are the enemy of urban density whenever a new multi family structure is built. If the future of self driving cars is shared taxis, rather than personally-owned vehicles, then I would expect parking-space requirements to shrink more and more over time and space currently given over to parking be instead given over to things people want (parks, yards, apartments) - all of which argue for increased urban populations as a result of self-driving.
This is my thought as well, but also people often opposes increased density on their street because they are worried about the in street parking. That would no longer be a concern with self driving cars. We have lived in an age where parking was considered more important than walk ability, and now parking will be irrelevant.
The group being pulled towards both Christianity and Nazism feels culturally insecure, that's the common thread. Very entertaining armchair psychoanalysis of such personality types (from 1941!): https://harpers.org/archive/1941/08/who-goes-nazi/
I think there is another group of people who aren't necessarily opposing a slow down, but also aren't viewing AI via the same apocalyptic (or Utopian) lens. Folks like Gary Marcus or Cal Newport. They understand the technology, but tend to point to examples like the Hugging Face attack as more a case of negligence by OpenAI and its ilk, rather than evidence that a malevolent machine god is nigh, nor that a vibe coded super virus will kill us all.
To be clear, this world view is not incompatible with negligent, criminal, or otherwise hostile uses of LLMs. I work in the energy industry, and the prospect of AI enhanced hackers working for hostile governments (e.g.,. GRU) disrupting critical infrastructure feels very salient. Similarly, one can imagine chaos being caused among financial institutions.
This raises a few thoughts for me.
First, negligent use of LLMs that result in crimes being committed or financial harm should be prosecuted / exposed to civil liability. Hugging Face style attacks don't just happen on their own. A team at OpenAI, presumably with the approval of management, started the process that ran an LLM that was trained on hacking examples, had constraints removed, and was left running on a loop apparently without human supervision. And they committed a crime in the process!
Second, given that lighter-weight models appear to only be about 6 months behind the frontier models in functionality, we're going to need to find a way to find and patch vulnerabilities ahead of new hacking capabilities becoming broadly available.
Third, I wonder if the gap between AI frontier lab employee expectations of harms LLMs can cause and the folks mentioned above (Newport and Marcus), is similar to the gap between those same lab employees and professional economists. 15% annual growth in GDP is plainly ridiculous to anyone who uses these tools in the real world. Similarly, a machine god killing us all is probably also the result of those folks irrational exuberance for/ fear of the technology they work on day to day. I appreciate that they have a closer view as to what is actually happening with the progression of LLMs and coding harnesses. But it is also becoming pretty clear that intellectual lineage of these labs are in some ways quite odd (e.g., a foundational text is a piece of Harry Potter fan fiction that is approximately the same length as War and Peace).
That's a reasonable point, although I haven't actually read War and Peace specifically, so I don't know whether I can recommend it to the kind of person who likes old school science fiction (Asimov, etc.) and post-Tolkein epic fantasy and are not necessarily fans of modern lit-fic.
Also, just because something is fan fiction doesn't mean it's bad. After all, the Pulitzer Prize winning novel "James" by Percival Everett is, quite literally, fanfiction of Mark Twain's "The Adventures of Huckleberry Finn".
And if you want a genuine recommendation for something to read, I'm absolutely convinced that Terry Pratchett is the best writer in the English language since Shakespeare and his books kept getting better and better as he got older. (The book to start reading Discworld with is Mort.)
Thanks for the recommendation! I'm definitely a fan of Pratchett, Asimov and Tolkein.
As someone who grew up on sci-fi and fantasy, I've recently branched out to literary fiction. This includes classics like War and Peace, as well as 20th century literature (e.g., Cormac McCarthy, Don Delillo, Isabelle Allende, Tony Morrison, and many more). It has been a very positive experience.
My recommendation in turn would be to consider reading Lonesome Dove. A modern classic, but timeless as well and very readable.
As for fan fic, I know that was a bit of cheap shot. But I was using it as a rhetorical tip of the iceberg for a broader set of oddities associated with the Rationalist/ Effective Altruist intellectual cul-de-sacs that these labs come out of.
I have not read the new Zidar and Zwick book, but it does not make a novel claim.
The Millionaire Next Door (1996) by Thomas Stanley and William Danko already argued that the stereotypical image of the American millionaire was badly misleading. Their typical wealthy person was often a self-employed business owner living in an ordinary community, operating something like a small factory, retail chain, or service business.
Everyone I know who rages against the tech bros would firmly agree with that picture as of the 1980s/90s. Indeed that's half the point. That raging is strongly driven by a perception that the nature of people getting rich has _changed_ for the worse starting with the Bill Gates/Paul Allen cohort.
So the new research that Noah quotes sounds like quite a novel argument about American wealth distribution _now_.
No, I guarantee you that everyone who rages against the tech bros now were also raging against the wealthy in the 1980s/90s as well.
This rage is nothing new, nor are the facts. The rage just has a new target.
I think that it is more accurate to say that The Millionaire Next Door (1996) by Thomas Stanley and William Danko challenged false perceptions about the wealthy, but it had little impact on public opinion.
Raging against the wealthy is more driven by ideological assumptions rather than accurate perceptions of how people acquire wealth.
I am not straw-manning your comment at all. I never said anything about your beliefs.
My reply is directly to your comment.
I am saying that rage against the wealthy was not invented recently. And the rage today is still based on the same false information that it has been in previous decades and centuries.
I have no idea whether you feel that rage or not, nor did I claim that you did.
On urban self-driving cars, I see how they would be popular, because of the universal benefits (convenience, safety, personal space, comfort) and also economic (much less need for parking at residential or commercial areas). There is an issue of road capacity, but one way to solve that is to have autonomous minibuses which operate like modern office building elevators where you enter a floor number and it tells you which elevator car to get in which stops at your floor. On your phone you say where you want to go, and the service finds the autobus near your area that is going toward your destination and has seats available - which would result in fewer vehicles on the street, quick pickup, no unnecessary stops, no vagrants on board and cheaper energy costs. I guess UberPool is almost like this, but with a human driver, but I’ve never used it out here in the exurbs.
Most people don’t understand the basic mechanism of LLL AI which means it is easy to generate anxiety about it. In the ideal world the political system would make the first goal understanding. For instance the Hugging Face hack, which Nate Silver wrote about in a piece on persistence, was really just classic hacking at scale. Many fears I’ve read, like the fact that agents created a “new” language to communicate, are easily understood if you understand the technology. The AI doesn’t know words it knows tokens so to AI it’s not a new language, it’s just the mixture of tokens that are most probable at that point in time.
Like our politics the extremes tend to dominate. With AI we see this clearly, with Bernie Sanders and Trump taking expected positions, neither of which reflects a balanced view of the technology and potential risks. AI is a powerful new tool different from most tools in that we don’t know, and can’t know, the output from a given input. There is a lot of work in prompt engineering, and RAG, and harness ontology, to make AI repeatable enough to a useful tool. A pixel in the wrong place in an AI image is different from prescribing a drug that will kill the patient. Without a knowledge metric AGI isn’t possible to even consider, and with some domain exceptions, like coding, RSI isn’t possible.
I’m solidly with the economists on this. It is very powerful for some things that make humans more productive. It will continue to change and improve. It could result in large job losses across some sectors but job gains in others. It will take time to diffuse through the entire economy.
But I was thinking that if pre-gps enabled phone with internet service available around much of the world did not result in explosive growth why would AI? I would think that in any economic model suddenly having that combination of location, communication, and transaction, would lead to a large productivity burst. It is clearly a huge step forward in potential productivity.
As far as the danger of the OpenAI Hugging Face hack and similar incidents goes, an interesting counter argument is that the AI companies are too focused on model alignment and training instead of just basic cybersecurity and focus on the actual software instead of just training. This is explained at length at Normal AI Technology here on substack https://www.normaltech.ai/p/ai-as-normal-technology
Noah, I see that you follow Loyal, which aims to increase healthy lifespan in dogs. A pension and financial expert recently wrote a report on the short-term difficulties of increased healthy lifespan in humans, though he mentioned the long-term benefits. I would be awfully curious about your takes on increasing healthy lifespan in humans through biomedical research targeting the biology of aging.
Self-driving cars are going to add to congestion until all cars are self-driving and are able to be managed as a public transportation service that ensures maximum flow rate. The reason is simple: self-driving cars must run empty to re-position to the next pick up. You can reduce the fleet while increasing the journeys because an owned car goes from A to B occupied. It also needs parking at both ends.
I've always thought that a big part of densifying in the world of self-driving cars would be the elimination of the garage. If you don't need a car, because you can easily and cheaply rent one for your trip, then you don't need a garage. And if you don't need a garage, you can cut a massive portion of the size of your home.
Self-driving cars will also eliminate a lot of need for parking in dense cities, allowing more amenities to pack closer together and enhancing the urban landscape in exactly the ways that urbanists like the most.
I haven’t read the comments, so I don’t know if someone else already made this point.
Glaeser seems to believe that if you replaced public transit and walking in dense cities with self-driving cars, everyone would get to their destination more quickly. But he’s ignoring the problem of geometry, I.e. two objects cannot occupy the same space at the same time.
Picture 100 urban dwellers walking to the nearest bus stop and getting on a bus. Now picture each of these 100 people in their own (self-driving) car instead.
Replacing one bus with 100 cars, self-driving or not, is going to create a ginormous traffic jam and it’s going to increase, not decrease, travel times!
This seems like a really obvious point for someone as intelligent as Glaeser to overlook. What am I missing?
I’m neither progressive nor do I think that AI doesn’t work, but I don’t think AI is anything more than a new tool. Where does that fit in the groups of people who think the AI doom stuff is over the top?
Tahra is IMV a super underrated writer on Substack because she is a rigorous and thorough journalist who also isn’t chronically online (or at least oneshotted by its discursive norms). She’s also willing to get on the ground in a way fewer people tend towards these days. Anyway hype to see you elevate her work:
I love that in the recent 'Confucian chart' Mongolia and Hungary are right next to each other.
Attila would be proud.
Attila the Hungarian, or Goulash Khan?
Parking space requirements are the enemy of urban density whenever a new multi family structure is built. If the future of self driving cars is shared taxis, rather than personally-owned vehicles, then I would expect parking-space requirements to shrink more and more over time and space currently given over to parking be instead given over to things people want (parks, yards, apartments) - all of which argue for increased urban populations as a result of self-driving.
This is my thought as well, but also people often opposes increased density on their street because they are worried about the in street parking. That would no longer be a concern with self driving cars. We have lived in an age where parking was considered more important than walk ability, and now parking will be irrelevant.
"Christians and Nazis should be natural enemies"
The group being pulled towards both Christianity and Nazism feels culturally insecure, that's the common thread. Very entertaining armchair psychoanalysis of such personality types (from 1941!): https://harpers.org/archive/1941/08/who-goes-nazi/
I think there is another group of people who aren't necessarily opposing a slow down, but also aren't viewing AI via the same apocalyptic (or Utopian) lens. Folks like Gary Marcus or Cal Newport. They understand the technology, but tend to point to examples like the Hugging Face attack as more a case of negligence by OpenAI and its ilk, rather than evidence that a malevolent machine god is nigh, nor that a vibe coded super virus will kill us all.
To be clear, this world view is not incompatible with negligent, criminal, or otherwise hostile uses of LLMs. I work in the energy industry, and the prospect of AI enhanced hackers working for hostile governments (e.g.,. GRU) disrupting critical infrastructure feels very salient. Similarly, one can imagine chaos being caused among financial institutions.
This raises a few thoughts for me.
First, negligent use of LLMs that result in crimes being committed or financial harm should be prosecuted / exposed to civil liability. Hugging Face style attacks don't just happen on their own. A team at OpenAI, presumably with the approval of management, started the process that ran an LLM that was trained on hacking examples, had constraints removed, and was left running on a loop apparently without human supervision. And they committed a crime in the process!
Second, given that lighter-weight models appear to only be about 6 months behind the frontier models in functionality, we're going to need to find a way to find and patch vulnerabilities ahead of new hacking capabilities becoming broadly available.
Third, I wonder if the gap between AI frontier lab employee expectations of harms LLMs can cause and the folks mentioned above (Newport and Marcus), is similar to the gap between those same lab employees and professional economists. 15% annual growth in GDP is plainly ridiculous to anyone who uses these tools in the real world. Similarly, a machine god killing us all is probably also the result of those folks irrational exuberance for/ fear of the technology they work on day to day. I appreciate that they have a closer view as to what is actually happening with the progression of LLMs and coding harnesses. But it is also becoming pretty clear that intellectual lineage of these labs are in some ways quite odd (e.g., a foundational text is a piece of Harry Potter fan fiction that is approximately the same length as War and Peace).
I'll take "Harry Potter and the Methods of Rationality" over "Atlas Shrugged" any day...
Why read either War and Peace length tome, when one could just read War and Peace (etc.) instead?
That's a reasonable point, although I haven't actually read War and Peace specifically, so I don't know whether I can recommend it to the kind of person who likes old school science fiction (Asimov, etc.) and post-Tolkein epic fantasy and are not necessarily fans of modern lit-fic.
Also, just because something is fan fiction doesn't mean it's bad. After all, the Pulitzer Prize winning novel "James" by Percival Everett is, quite literally, fanfiction of Mark Twain's "The Adventures of Huckleberry Finn".
And if you want a genuine recommendation for something to read, I'm absolutely convinced that Terry Pratchett is the best writer in the English language since Shakespeare and his books kept getting better and better as he got older. (The book to start reading Discworld with is Mort.)
Thanks for the recommendation! I'm definitely a fan of Pratchett, Asimov and Tolkein.
As someone who grew up on sci-fi and fantasy, I've recently branched out to literary fiction. This includes classics like War and Peace, as well as 20th century literature (e.g., Cormac McCarthy, Don Delillo, Isabelle Allende, Tony Morrison, and many more). It has been a very positive experience.
My recommendation in turn would be to consider reading Lonesome Dove. A modern classic, but timeless as well and very readable.
As for fan fic, I know that was a bit of cheap shot. But I was using it as a rhetorical tip of the iceberg for a broader set of oddities associated with the Rationalist/ Effective Altruist intellectual cul-de-sacs that these labs come out of.
You already have a way to find and patch the vulnerabilities. Just use the AI to hack yourself.
I have not read the new Zidar and Zwick book, but it does not make a novel claim.
The Millionaire Next Door (1996) by Thomas Stanley and William Danko already argued that the stereotypical image of the American millionaire was badly misleading. Their typical wealthy person was often a self-employed business owner living in an ordinary community, operating something like a small factory, retail chain, or service business.
Everyone I know who rages against the tech bros would firmly agree with that picture as of the 1980s/90s. Indeed that's half the point. That raging is strongly driven by a perception that the nature of people getting rich has _changed_ for the worse starting with the Bill Gates/Paul Allen cohort.
So the new research that Noah quotes sounds like quite a novel argument about American wealth distribution _now_.
No, I guarantee you that everyone who rages against the tech bros now were also raging against the wealthy in the 1980s/90s as well.
This rage is nothing new, nor are the facts. The rage just has a new target.
I think that it is more accurate to say that The Millionaire Next Door (1996) by Thomas Stanley and William Danko challenged false perceptions about the wealthy, but it had little impact on public opinion.
Raging against the wealthy is more driven by ideological assumptions rather than accurate perceptions of how people acquire wealth.
??
Oh I think I see...well your assumptions are showing.
Your reply has no connection to mine. Straw-manning someone else's comment is rude, kindly don't do it to me again.
I am not straw-manning your comment at all. I never said anything about your beliefs.
My reply is directly to your comment.
I am saying that rage against the wealthy was not invented recently. And the rage today is still based on the same false information that it has been in previous decades and centuries.
I have no idea whether you feel that rage or not, nor did I claim that you did.
On urban self-driving cars, I see how they would be popular, because of the universal benefits (convenience, safety, personal space, comfort) and also economic (much less need for parking at residential or commercial areas). There is an issue of road capacity, but one way to solve that is to have autonomous minibuses which operate like modern office building elevators where you enter a floor number and it tells you which elevator car to get in which stops at your floor. On your phone you say where you want to go, and the service finds the autobus near your area that is going toward your destination and has seats available - which would result in fewer vehicles on the street, quick pickup, no unnecessary stops, no vagrants on board and cheaper energy costs. I guess UberPool is almost like this, but with a human driver, but I’ve never used it out here in the exurbs.
Most people don’t understand the basic mechanism of LLL AI which means it is easy to generate anxiety about it. In the ideal world the political system would make the first goal understanding. For instance the Hugging Face hack, which Nate Silver wrote about in a piece on persistence, was really just classic hacking at scale. Many fears I’ve read, like the fact that agents created a “new” language to communicate, are easily understood if you understand the technology. The AI doesn’t know words it knows tokens so to AI it’s not a new language, it’s just the mixture of tokens that are most probable at that point in time.
Like our politics the extremes tend to dominate. With AI we see this clearly, with Bernie Sanders and Trump taking expected positions, neither of which reflects a balanced view of the technology and potential risks. AI is a powerful new tool different from most tools in that we don’t know, and can’t know, the output from a given input. There is a lot of work in prompt engineering, and RAG, and harness ontology, to make AI repeatable enough to a useful tool. A pixel in the wrong place in an AI image is different from prescribing a drug that will kill the patient. Without a knowledge metric AGI isn’t possible to even consider, and with some domain exceptions, like coding, RSI isn’t possible.
I’m solidly with the economists on this. It is very powerful for some things that make humans more productive. It will continue to change and improve. It could result in large job losses across some sectors but job gains in others. It will take time to diffuse through the entire economy.
But I was thinking that if pre-gps enabled phone with internet service available around much of the world did not result in explosive growth why would AI? I would think that in any economic model suddenly having that combination of location, communication, and transaction, would lead to a large productivity burst. It is clearly a huge step forward in potential productivity.
As far as the danger of the OpenAI Hugging Face hack and similar incidents goes, an interesting counter argument is that the AI companies are too focused on model alignment and training instead of just basic cybersecurity and focus on the actual software instead of just training. This is explained at length at Normal AI Technology here on substack https://www.normaltech.ai/p/ai-as-normal-technology
Noah, I see that you follow Loyal, which aims to increase healthy lifespan in dogs. A pension and financial expert recently wrote a report on the short-term difficulties of increased healthy lifespan in humans, though he mentioned the long-term benefits. I would be awfully curious about your takes on increasing healthy lifespan in humans through biomedical research targeting the biology of aging.
Paper on short-term risks:
https://static1.squarespace.com/static/6a64bc7e630f854226e5ef77/t/6a9aeb9d1a9dba267d64c18f/1788537757908/Extreme+Longevity+Coughlan-Faragher+2026-07-20.pdf
Paper on overall economic benefit:
https://www.nature.com/articles/s43587-021-00080-0
Self-driving cars are going to add to congestion until all cars are self-driving and are able to be managed as a public transportation service that ensures maximum flow rate. The reason is simple: self-driving cars must run empty to re-position to the next pick up. You can reduce the fleet while increasing the journeys because an owned car goes from A to B occupied. It also needs parking at both ends.
I've always thought that a big part of densifying in the world of self-driving cars would be the elimination of the garage. If you don't need a car, because you can easily and cheaply rent one for your trip, then you don't need a garage. And if you don't need a garage, you can cut a massive portion of the size of your home.
Self-driving cars will also eliminate a lot of need for parking in dense cities, allowing more amenities to pack closer together and enhancing the urban landscape in exactly the ways that urbanists like the most.
I think that's true, but what do you envision for overnight storage? I'm thinking maybe one of those elevator garages like Mitt Romney has.
I haven’t read the comments, so I don’t know if someone else already made this point.
Glaeser seems to believe that if you replaced public transit and walking in dense cities with self-driving cars, everyone would get to their destination more quickly. But he’s ignoring the problem of geometry, I.e. two objects cannot occupy the same space at the same time.
Picture 100 urban dwellers walking to the nearest bus stop and getting on a bus. Now picture each of these 100 people in their own (self-driving) car instead.
Replacing one bus with 100 cars, self-driving or not, is going to create a ginormous traffic jam and it’s going to increase, not decrease, travel times!
This seems like a really obvious point for someone as intelligent as Glaeser to overlook. What am I missing?
I’m neither progressive nor do I think that AI doesn’t work, but I don’t think AI is anything more than a new tool. Where does that fit in the groups of people who think the AI doom stuff is over the top?
Tahra is IMV a super underrated writer on Substack because she is a rigorous and thorough journalist who also isn’t chronically online (or at least oneshotted by its discursive norms). She’s also willing to get on the ground in a way fewer people tend towards these days. Anyway hype to see you elevate her work:
Glaeser is probably one of the best / most prominent living urban economists, but who are the best dead ones?
Baumol’s cost disease sounds like a D&D spell.