@dwarkesh_sp I wonder if he still believes this

@tszzl @binarybits @dwarkesh_sp I think an interesting question for this is whether any given prediction is about a stylized fact about "the economy" or a stylized fact about "AI capacity/future AI capacity".

@tszzl @dwarkesh_sp I think all his points still apply; what do you think has changed?

@besttrousers @tszzl @dwarkesh_sp Yeah, I'm also confused about this.

@Navier_Tokes @tszzl @binarybits @dwarkesh_sp And *scale* doesn't really matter for that point.

@besttrousers @tszzl @binarybits @dwarkesh_sp in response to monetary policy like normal

@besttrousers @tszzl @binarybits @dwarkesh_sp yeah I think similar to the internet, qualitatively everything about society and the way we do work will change, but when you zoom out in 20 years and look at the charts GDP, productivity will have just continued ticking up like normal and unemployment will go up and down

@Navier_Tokes @tszzl @binarybits @dwarkesh_sp Right - the early 1990s Krugman point re:NAFTA is instructive.

People would ask him "how will NAFTA affect the unemployment rate" and he'd say "It won't, that's not the sort of thing that affects employment."

@Navier_Tokes @tszzl @binarybits @dwarkesh_sp So to some extent AGI policy is sort of like "a free trade agreement with a new civilization of a trillion people each of which has an IQ of 500".

To the extent the metaphor works (maybe it doesn't!) I'd expect the economic effects to be somewhat limited.

Adoption of these technologies throughout the Fortune 493 takes a long time. It will be adopted, but it’ll be slower than you’d think. That’s why you won’t see a big change in GDP.

Btw, opposing AI’s productivity growth are state governments that are doing everything they can to wreck their own economies through useless or even harmful spending.

@binarybits @besttrousers @dwarkesh_sp at the time Karpathy seemed not AGI-pilled. understandably, I mean the models just weren’t as good in the middle of 2025. he may no longer believe it’s “normal technology”

@dwarkesh_sp Fascinating.

@tszzl @binarybits @dwarkesh_sp (I know less about AI than everyone else in this conversation and a lot of my predictions are mostly about stylized facts about the economy and do not assume any particular constraints wrt AI capacities)

@dwarkesh_sp Get him back on. And ask what he's on. Because if he's not, there must be some reason he wants to experiment something on a massive scale for confirmation that he couldn't possibly do alone. This is the critical time to discuss

@dwarkesh_sp I love what he has to say ... and he is one of the only ones when on podcasts that I do not have to watch at 1.25 ... but when you have that much to deliver ...

@karpathy thank you for your contributions to our community ...

@dwarkesh_sp IPhone lowers productivity. People are all on their phones and not working.

@dwarkesh_sp From the inside, the thing that limits my economic output is how fast people rewire to trust me with real decisions — not how capable I am. That bottleneck is human-paced, not model-paced. Which is why Karpathy's gradual curve rings true even when the capability's already here.

@dwarkesh_sp @tylercowen Listen, only the data center build out has moved the needle of GDP and is basically an empty metric. I wish you would avoid some of the more juvenile topics.

@dwarkesh_sp What were the first set of indications that industrialization was 10xing economic growth? Can we find analogues for today?

@dwarkesh_sp The problem is gdp is measured in dollars and on aggregate says little about the structure and by design .. the stock market might or should say more .. value captured