@avi_collis @Jabaluck We’ll hopefully be integrating this into our forthcoming suite of AI-related measurement work!
English
Connacher Murphy
501 posts
Connacher Murphy
@connacher_
Stanford Digital Economy Lab https://t.co/eU3kfDdGdd
Katılım Haziran 2020
543 Takip Edilen272 Takipçiler
In worlds where you get full automation and a production explosion, GDP ceases to be a decent proxy for welfare. We should really be forecasting other metrics in those worlds. (1/n)
English
@Jabaluck Because I’m growing more curious about this: are there any existing measures we should’ve used instead of GDP? Any measures that should be created for this purpose?
English
So really, forecasts about GDP in a world with complete automation are irrelevant. What matters is forecasts of welfare. Makes me think that next time I co-author a paper asking partly about such futures, we should use alternative measures of growth!
English
@evavivalt I find the ATUS-related work in the Aguiar et al. paper above exciting, but I haven't seen any alternatives
English
@evavivalt Is there a good way to disentangle the "good" and "bad" cases of exiting the labor force?
*Not speaking for my coauthors on the linked study*: one quibble I had with the paper is our framing of a forecasted increase in LFRP due to a policy ==> support for the policy
English
I perhaps optimistically think the way in which non-employment is likely to grow is from substitution into new and better kinds of leisure, rather than AI taking all the jobs.
Gregor Schubert@gregorschub
A minor quibble - but matters for interpretation: lower LFPR is not "fewer jobs" but "fewer people interested in a job". The latter is not always bad, might come from leisure technologies etc. - workers lost their job and are unemployed would still be in the labor force!
English
@AndreyFradkin Do you have any ideas on promising alternative measures given (2)?
English
I am more optimistic than my fellow economist forecasters about annual GDP growth by 2050. That said, even my forecast is not as high as the 10%+ forecasts that some are predicting. A few reasons why:
1) The spread of AI capabilities is greatly increasing variance (which includes substantial weight on the negative scenarios including war and revolution).
2) It's likely that AI will cause a greater decoupling of apparent progress and GDP, so that focusing on GDP will miss something.
3) A lot of economic bottlenecks in the US are regulatory, and in many states of the world, it will take more than 20 years for governance to adjust to let AI subsume these.
4) Heathcare is a large share of GDP, and even though AI will drastically increase medical progress, we may still need RCTs before medicines are widely used and affect GDP.
Nonetheless, the possibility of extremely fast growth cannot be dismissed, and I place about 10% weight on these scenarios.
tom cunningham@testingham
This is a great paper but contains a puzzle: forecasters expect even if we automate most labor and wait 20 years, GDP will only increase by 45%. I would love to hear how people are thinking about this.
English
an important determinant of the labor market impacts of AI!
Martha Gimbel@marthagimbel
Consumer demand - how does consumer demand respond to a technological shock? 8/
English
where’s the market for people to sell their Claude Code usage limits?
English
@Research_FRI Open question: does accuracy on capabilities questions translate into accuracy on adoption and impacts questions?
English
@Research_FRI On the usefulness of resolution, I like this post following-up on results from the XPT: substack.com/home/post/p-17…
The conclusion there: "Respondents underestimated AI progress, especially superforecasters"
English
The reaction to this project highlights one thing that excites me about judgmental forecasting: debates that culminate in (falsifiable) forecasts are more transparent and productive. This is especially true with conditional forecasting. One such example:
tom cunningham@testingham
This is a great paper but contains a puzzle: forecasters expect even if we automate most labor and wait 20 years, GDP will only increase by 45%. I would love to hear how people are thinking about this.
English
@bqbrady I guess there's no reason to think there is any obfuscation going on in this context
English
Benchmarking Frontier LLMs on Chess
Over the weekend I built a series of evals to understand how language models reason about endgames, tactics, and full chess games against strong opponents. Turns out they are getting pretty good!
benedict.dev/chess-bench
English
@Jsevillamol This is why I found the forecasts for the 90th percentile growth outcomes so surprising! I would have expected many more respondents to allow for at least a 10% chance of much more rapid growth.
English
This was exactly my reaction when I saw the poll results. One thing is to be uncertain, but not even acknowledging that we might be headed for very rapid growth is a mistake!
Basil Halperin@BasilHalperin
1. Economists are AI-pilled (surprising!)... - Both economists & AI folks give ~15% chance that AI surpasses humans by 2030 on most cognitive and physical tasks 2. ...but no one is singularity-pilled - Expecting basically normal GDP growth (1950s levels at best)!! 🧵
English
@BasilHalperin I would have expected that at least a quarter of economists would assign a >10% chance to much more rapid growth
English
@BasilHalperin This was great!
I don't find the forecasts the 50th percentile growth outcome by 2050 in the rapid scenario surprising as much as I find the 90th percentile forecasts surprising!
The *top quartile* of *90th percentile forecasts* is 10% per year.
English
1. Economists are AI-pilled (surprising!)...
- Both economists & AI folks give ~15% chance that AI surpasses humans by 2030 on most cognitive and physical tasks
2. ...but no one is singularity-pilled
- Expecting basically normal GDP growth (1950s levels at best)!! 🧵
Forecasting Research Institute@Research_FRI
We completed the most comprehensive study of how economists and AI experts think AI will affect the U.S. economy. They predict major AI progress—but no dramatic break from economic trends: GDP growth rates similar to today's and a moderate decline in labor force participation. However, when asked to consider what would happen in a world with extremely rapid progress in AI capabilities by 2030, they predict significant economic impacts by 2050: • Annualized GDP growth of 3.5% (compared to 2.4% in 2025) • A labor force participation rate of 55% (roughly 10 million fewer jobs) • 80% of wealth held by the top 10% (highest since 1939) 🧵 Here's what we found:
English
New post out, in which I use the phrase “figgy abundance” and rail against “keeping your options open”
English