Pinned Tweet
Has AI hit the labor market yet? @alexolegimas and my verdict after ~20 papers: not yet in aggregate. Unemployment and layoffs show almost nothing. But AI may already be cutting junior hiring in exposed white-collar jobs. Remote work may explain part of that decline.
A 🧵
A 🧵
17
55
190
87,328
The UK’s new AI Economics Institute is hiring up to 15 research economists and analysts. The roles combine research on the economic effects of AI with real policy impact.
2
24
85
7,811
They’re recruiting people from two specialisms:
➡️ Research Economists - lead the most technical primary research projects
➡️ Analysts - bridge the gap between technical research and policy
➡️ Research Economists - lead the most technical primary research projects
➡️ Analysts - bridge the gap between technical research and policy
1
0
8
571
If you are interested, here are the job adverts:
Head of Research Economics: civilservicejobs.service.gov.uk/csr/jobs.cgi?j…
Head of Analysis: civilservicejobs.service.gov.uk/csr/jobs.cgi?j…
Research Economist: civilservicejobs.service.gov.uk/csr/jobs.cgi?j…
Analyst: civilservicejobs.service.gov.uk/csr/jobs.cgi?v…
Head of Research Economics: civilservicejobs.service.gov.uk/csr/jobs.cgi?j…
Head of Analysis: civilservicejobs.service.gov.uk/csr/jobs.cgi?j…
Research Economist: civilservicejobs.service.gov.uk/csr/jobs.cgi?j…
Analyst: civilservicejobs.service.gov.uk/csr/jobs.cgi?v…
0
2
5
689
Jacob Schaal retweeted
1/6 METR’s report mentioning the existence of a specific unredacted report could actually be quite important. This is because it gives government bodies something concrete to ask developers for, even if sharing isn’t legally required. 🧵
2
13
123
7,114
Useful context from Epoch’s RLI review by @xeophon and @GregHBurnham: tasks are self-contained multimedia freelance projects. Furthermore, the scaffold problems they found understate the model's ability.
epoch.ai/publications/w…
epoch.ai/publications/w…
Fable 5.2 scored 41.3% on the Remote Labor Index, more than three times Ryan's forecast for 2026
0
0
7
654
Over lunch I said AI-assisted AI research should speed up capabilities before it closes the RSI loop.
@METR_Evals finds Anthropic’s benchmark progress up 2x since Oct 2025, mostly from AI. Fable 5.2’ fits that. First AI jolt I’ve felt since o1.
@METR_Evals finds Anthropic’s benchmark progress up 2x since Oct 2025, mostly from AI. Fable 5.2’ fits that. First AI jolt I’ve felt since o1.
Claude Fable 5.2 sets a new record on the Epoch Capabilities Index (ECI) at 174. The previous high was GPT-6 Astra’s 169. Fable 5.2’s score is above what the pre-Astra trend for reasoning models would have predicted.
This is notable given recent discussion of AI contributions to release more work in this area soon.
This is based on testing we did prior to launch using only our own benchmarks. The ECI is continually updated as more benchmark results become available, and may change in the future.
0
1
9
516
Chinese AI labor market policy uses strong words in its 18 point plan, but mostly uses recommendations.
More binding are court rulings limiting AI-based worker replacement. But increasing firing costs could also inadvertently reduce hiring.
mee.gov.cn/zcwj/gwywj/202…
More binding are court rulings limiting AI-based worker replacement. But increasing firing costs could also inadvertently reduce hiring.
mee.gov.cn/zcwj/gwywj/202…
1
1
4
269
Still in line with this great read: mattsheehan.substack.com/p/china-is-get…
0
0
1
38
Here is the vibecoded version of the semi-elasticity. How to interpret index scores remains an issue. Still, each additional ECI point may even represent a larger capability gain at higher index levels.
Even if the R&D returns per dollar are falling the R&D returns per doubling can be stable which matters more for feedback loops cc @cherylwoooo Additionally its unclear whether we care about AECI units (or they are getting harder)
0
2
9
530
Jacob Schaal retweeted
Grateful to our authors and my co-editors @TimothyTTaylor @ProfJAParker and Jeff Kling for all of the work that went into this issue of the Journal of Economic Perspectives.
Open access as always: aeaweb.org/issues/868
Open access as always: aeaweb.org/issues/868
1
9
41
5,830
Jacob Schaal retweeted
Biggest question in AI right now: does the acceleration only apply to verifiable tasks and so quickly peter out, or are non-verifiable tasks also improving rapidly, just a bit behind? My money's on the latter.
14
22
206
9,674
Jacob Schaal retweeted
Ought to add—this chart would doubtless look the same in the rest of Europe, but there the relevant data is so lagged that we can't even say for sure yet.
That, at least, is one problem Britain doesn't have quite so much...
That, at least, is one problem Britain doesn't have quite so much...
AI will be the first tech capex boom that Britain misses out on in the modern era
0
1
15
1,023
Total private victory for @ElasticityInst and @testingham: They got the data, and the results are striking. The redactions limit what we can learn and what I can comment on.
Anthropic shared extensive evidence across its AI R&D pipeline, from pretraining to RL, including key details like estimates of algorithmic progress over time. We found this nonpublic evidence highly informative about AI R&D acceleration.
0
2
10
428
Jacob Schaal retweeted
1. Benchmarks provide one piece of evidence for acceleration. Since Oct 2025, AECI scores for Anthropic’s models have grown at 1.9x [95%: 1.5–2.8] the earlier rate. Analysis choices affect this estimate, and it remains unclear how benchmark scores map to real-world capabilities.
19
96
1,101
629,519
Jacob Schaal retweeted
2. We believe that a majority of this acceleration in capability improvement is likely now driven by AI being used in AI R&D. We aren’t sure of this, though, and did not find consensus about this among Anthropic’s staff. Non-AI factors likely played a large role in late 2025.
9
50
812
148,665
Jacob Schaal retweeted
Amazing new paper by Chetty and the Opp Insights team — digital twins can recover causal effects
They validate the digital twins (ie rich ML applied to generate counterfactuals from lots of covariates) with RCTs, allowing for scalable program evaluation
opportunityinsights.org/wp-content/upl…
They validate the digital twins (ie rich ML applied to generate counterfactuals from lots of covariates) with RCTs, allowing for scalable program evaluation
opportunityinsights.org/wp-content/upl…
11
86
372
107,050
Jacob Schaal retweeted
Is the UK Government literally #1 in the world in terms of how well it's preparing for AGI, given the resources it has available?
After last week's budget, I think so.
Puts some other countries to shame really.
After last week's budget, I think so.
Puts some other countries to shame really.
30
71
620
89,949
Skeptical that AI fear is suppressing pay: so far, AI exposure and adoption coincide with wage gains. But the Atlanta Fed finding is worth watching: higher perceived job-finding risk predicts fewer job switches. economist.com/finance-and-ec…
1
1
7
382