Let me talk today about sponge cities.
When I discuss fertility decline, I often get the following comment:
“As the population falls, housing prices will also fall, which will help with fertility and the system will self-correct.”
Perhaps not.
As the population falls, we are observing a phenomenon called sponge cities (see the map from Japan: do you know which city is the green spot?). There are even more incentives for the population to concentrate in large cities (e.g., Tokyo or Seoul) for three reasons:
1⃣ Jobs. As the population shrinks in many regions, jobs disappear with it. Yes, you can telecommute for some jobs, but there are fewer of those than you’d think. A plumber cannot telecommute.
2⃣ Services. As the population shrinks in many regions, services like grocery stores, hospitals, schools, etc., also disappear. I’ve seen this in many villages in Europe: population falls below a threshold, and the local supermarket closes. This creates a negative spiral that’s hard to break.
3⃣ Amenities. As the population shrinks, amenities like bars, theaters, and restaurants vanish too. And it turns out people, especially younger cohorts, care more about amenities than about jobs. You might be telecommuting, but you cannot telebar.
So it might well be the case that housing prices won’t fall in sponge cities, and that this won’t help fertility.
Self-correcting mechanisms often don’t work.
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
When I discuss fertility decline, I often get the following comment:
“As the population falls, housing prices will also fall, which will help with fertility and the system will self-correct.”
Perhaps not.
As the population falls, we are observing a phenomenon called sponge cities (see the map from Japan: do you know which city is the green spot?). There are even more incentives for the population to concentrate in large cities (e.g., Tokyo or Seoul) for three reasons:
1⃣ Jobs. As the population shrinks in many regions, jobs disappear with it. Yes, you can telecommute for some jobs, but there are fewer of those than you’d think. A plumber cannot telecommute.
2⃣ Services. As the population shrinks in many regions, services like grocery stores, hospitals, schools, etc., also disappear. I’ve seen this in many villages in Europe: population falls below a threshold, and the local supermarket closes. This creates a negative spiral that’s hard to break.
3⃣ Amenities. As the population shrinks, amenities like bars, theaters, and restaurants vanish too. And it turns out people, especially younger cohorts, care more about amenities than about jobs. You might be telecommuting, but you cannot telebar.
So it might well be the case that housing prices won’t fall in sponge cities, and that this won’t help fertility.
Self-correcting mechanisms often don’t work.
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
😱4
This media is not supported in your browser
VIEW IN TELEGRAM
Morgan Stanley's Adam Jonas in new note: "We estimate 1 humanoid robot at $5/hour can do the work of 2 humans at $25/hour, generating an NPV of ~$200k/humanoid. 1 robot car can potentially drive down cost/mile of a ride share vehicle to <$0.20/mile (1/10th human-driven ride share). The marginal cost of your personal C-3PO robot should, over time, approach the marginal cost of power - you can't have GPUs without BTUs."
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
👀3😱1
Guys, ai progress just isn't slowing down
gpt-5 completes tasks that take 52% longer
trust the exponential
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
gpt-5 completes tasks that take 52% longer
trust the exponential
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
💯3😱1
DoomPosting
Guys, ai progress just isn't slowing down gpt-5 completes tasks that take 52% longer trust the exponential 🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
Measuring AI Ability to Complete Long Tasks
Summary: We propose measuring AI performance in terms of the length of tasks AI agents can complete. We show that this metric has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months. Extrapolating this trend predicts that, in under a decade, we will see AI agents that can independently complete a large fraction of software tasks that currently take humans days or weeks.
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
Summary: We propose measuring AI performance in terms of the length of tasks AI agents can complete. We show that this metric has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months. Extrapolating this trend predicts that, in under a decade, we will see AI agents that can independently complete a large fraction of software tasks that currently take humans days or weeks.
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
🔥3💯1
NEW: Trump to host leaders of sworn enemies Armenia and Azerbaijan today for historic ‘Peace Signing’
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
😁3🙏1🕊1
GLORIOUS LEADER ANNOUNCES HISTORIC VICTORY: EUROPEAN WINE & CHEESE BECOME RARE TREASURES, PEASANTS SALUTE THE RISE OF AMERICAN FLAVOR!
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
😁4🏆3🫡1
DoomPosting
Measuring AI Ability to Complete Long Tasks Summary: We propose measuring AI performance in terms of the length of tasks AI agents can complete. We show that this metric has been consistently exponentially increasing over the past 6 years, with a doubling…
Typical experience with GPT-5 so far:
+ Ask GPT-5 for ideas on how to solve some very non-trivial problem, that I carefully specify
+ GPT-5 casually suggests a solution approach that would be nice, but would be wildly difficult to code
+ I say, Oh really? Prove it bro
+ GPT-5 does it no problem, and shows that its solution works
So, GPT-5 kinda totally nailing it so far
ALTHOUGH — this my tests so far are still largely the “truth/factual” domain, rather than the “values” domain — i.e. synthesizing a solution that factually fits my detailed specification, rather than having the sense of values to know what are valuable specifications to come up with in the first place
Will have to run more tests on my collection of hard problems to see how this holds up
Will say I still don’t trust any AI code at all, unless very thoroughly verified, and AI code still has a huge list of problems, probably
We’ll see
(Image is just random piece of code, too lazy to pull together best examples rn)
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
+ Ask GPT-5 for ideas on how to solve some very non-trivial problem, that I carefully specify
+ GPT-5 casually suggests a solution approach that would be nice, but would be wildly difficult to code
+ I say, Oh really? Prove it bro
+ GPT-5 does it no problem, and shows that its solution works
So, GPT-5 kinda totally nailing it so far
ALTHOUGH — this my tests so far are still largely the “truth/factual” domain, rather than the “values” domain — i.e. synthesizing a solution that factually fits my detailed specification, rather than having the sense of values to know what are valuable specifications to come up with in the first place
Will have to run more tests on my collection of hard problems to see how this holds up
Will say I still don’t trust any AI code at all, unless very thoroughly verified, and AI code still has a huge list of problems, probably
We’ll see
(Image is just random piece of code, too lazy to pull together best examples rn)
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
👀5🔥1
NEW - U.S. removes online versions of past National Climate Assessments, saying they are being reviewed and updated
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶
⚡3😁1