Another possibility is that AI is just priced in. What I mean is that German GDP has been *falling* between 2023 and 2024, and only grew by 0.2% in 2025. We take the idea of infinite GDP growth for granted but maybe GDP growth is not automatic and happens because of constant development. Thus AI is required to merely sustain GDP growth, and the baseline to compare the impact of AI to is a timeline where GDP growth halts entirely as it did in Germany
I agree, I think this is why the internet doesnt show up on the graph. It fueled 20 years of continuous exponential growth and this is just seern as "business as usual."
I could be wrong but I don't think 2027 was ever Scott Alexander's median prediction, 2027 was their mode as a group and Daniel K was always the most optimistic in the squad, I tnink he was more like 2030-2031 though I need to find an exact source.
To clarify to those not in the know, AI 2027 was an early 2025 project by Scott Alexander, Daniel Kokotaljo, and others predicting AGI by 2027 and then takeover in 2028.
2027 was their modal prediction but they had variance among members of the group and their median was a little longer.
I would say end of 2029 AIs will be making frontier-level discoveries in science with regularity, aiding ML research to a significant degree to the point that it starts to feed back into itself, and code projects that humans would take a week to two weeks to complete in one session. Does that sound reasonable?
I think we would need to dig deeper into those terms.
The second point you make about coding projects that would take humans weeks to finish is a tricky one. Arguably they already do that, see the C compiler made by Claude recently. They are good at making prototypes of projects, or redoing things that have already been done before, but with some tweaks.
What I don't think they're good at, is maintaining code, tracking down hard to find bugs, creating novel ideas, executing on novel ideas...
I would prefer to make a bet on something concrete like unemployment statistics. Do you think it's fair to believe that by 2035 we would have unemployment at 50% or higher? I think if AI really achieved AGI we would expect to see stuff like this.
I had the same basic take as you. I don't disagree with Freddie's fundamental premise, but a couple of his metrics for proving things were still basically "normal" were wildly off base. Worse than the 18% total unemployment was the 35% professional/business services unemployment. If we got even 20% unemployment among that specific class, a second great depression is inevitable. That class of people (roughly the top 20% in income) do 60% of all consumer spending in our economy, pay the vast majority of all income taxes, and hold 70% of the nation's assets. If we got anything remotely approaching 35% unemployment, all their spending will screech to a halt (even by the remaining ones with jobs, who will all fear for them), the tax base goes poof, and they will crash all asset classes in a sell off. When their spending stops, everyone else loses their job too. There is zero way to avoid a Great Depression (a worse one, actually) if a third of the high income professional class that do all the spending and pay all the taxes are out of a job, other than massive redistribution and socialism. That would be a massive change. So I agree with him that the kind of stuff he's thinking of is unlikely to happen, but that metric is crazy. We could have 34% white collar unemployment and he would still win the bet and say things are normal. A second (even worse) great depression and/or straight socialism are NOT "normal" for the US!
Without accounting for climate degradation all measures of the economy are severely flawed. For data centers to exist at the scale needed, where do the metals and energy come from? GPUs have a limited useful life. How do we keep building at the rate needed and still have materials for anything else?
I think The Great Smpification is more likely than continuing growth. AGI is a dead end.
I do not how how exactly they could look like, but an entirely new set of econometric indicators of the state of an economy based on psychometric and biophysical properties would be cool. Maybe a gdp-gwas coupled with energy usage per capita statistics?
Another possibility is that AI is just priced in. What I mean is that German GDP has been *falling* between 2023 and 2024, and only grew by 0.2% in 2025. We take the idea of infinite GDP growth for granted but maybe GDP growth is not automatic and happens because of constant development. Thus AI is required to merely sustain GDP growth, and the baseline to compare the impact of AI to is a timeline where GDP growth halts entirely as it did in Germany
I agree, I think this is why the internet doesnt show up on the graph. It fueled 20 years of continuous exponential growth and this is just seern as "business as usual."
I could be wrong but I don't think 2027 was ever Scott Alexander's median prediction, 2027 was their mode as a group and Daniel K was always the most optimistic in the squad, I tnink he was more like 2030-2031 though I need to find an exact source.
To clarify to those not in the know, AI 2027 was an early 2025 project by Scott Alexander, Daniel Kokotaljo, and others predicting AGI by 2027 and then takeover in 2028.
2027 was their modal prediction but they had variance among members of the group and their median was a little longer.
He still put his name on a piece called AI 2027. And 3 years out is not that big of a difference.
I think one should be as correct as possible. And 2027 vs 2030 is a big difference as AGI by 2030 is not crazy at all.
That's fair.
In regards to your second point I would happily place a bet of 2,000 dollars on it NOT happening if we rigorously defined the terms.
I would say end of 2029 AIs will be making frontier-level discoveries in science with regularity, aiding ML research to a significant degree to the point that it starts to feed back into itself, and code projects that humans would take a week to two weeks to complete in one session. Does that sound reasonable?
I think we would need to dig deeper into those terms.
The second point you make about coding projects that would take humans weeks to finish is a tricky one. Arguably they already do that, see the C compiler made by Claude recently. They are good at making prototypes of projects, or redoing things that have already been done before, but with some tweaks.
What I don't think they're good at, is maintaining code, tracking down hard to find bugs, creating novel ideas, executing on novel ideas...
I would prefer to make a bet on something concrete like unemployment statistics. Do you think it's fair to believe that by 2035 we would have unemployment at 50% or higher? I think if AI really achieved AGI we would expect to see stuff like this.
I had the same basic take as you. I don't disagree with Freddie's fundamental premise, but a couple of his metrics for proving things were still basically "normal" were wildly off base. Worse than the 18% total unemployment was the 35% professional/business services unemployment. If we got even 20% unemployment among that specific class, a second great depression is inevitable. That class of people (roughly the top 20% in income) do 60% of all consumer spending in our economy, pay the vast majority of all income taxes, and hold 70% of the nation's assets. If we got anything remotely approaching 35% unemployment, all their spending will screech to a halt (even by the remaining ones with jobs, who will all fear for them), the tax base goes poof, and they will crash all asset classes in a sell off. When their spending stops, everyone else loses their job too. There is zero way to avoid a Great Depression (a worse one, actually) if a third of the high income professional class that do all the spending and pay all the taxes are out of a job, other than massive redistribution and socialism. That would be a massive change. So I agree with him that the kind of stuff he's thinking of is unlikely to happen, but that metric is crazy. We could have 34% white collar unemployment and he would still win the bet and say things are normal. A second (even worse) great depression and/or straight socialism are NOT "normal" for the US!
Without accounting for climate degradation all measures of the economy are severely flawed. For data centers to exist at the scale needed, where do the metals and energy come from? GPUs have a limited useful life. How do we keep building at the rate needed and still have materials for anything else?
I think The Great Smpification is more likely than continuing growth. AGI is a dead end.
I do not how how exactly they could look like, but an entirely new set of econometric indicators of the state of an economy based on psychometric and biophysical properties would be cool. Maybe a gdp-gwas coupled with energy usage per capita statistics?