Joshua Mask
@joshuafmask
Labor Economist @ Temple University I think a lot about how the economic conditions you face when you first enter the labor market affect long run outcomes.
Broke: A is technological development Woke: A is institutions Bespoke: A is AGI
Thanks to everyone that commented on last week's post about my new paper, Same Occupations, Different Clocks: Separating Remote Work from Generative AI for Labor-Market Entrants week. Your feedback was very helpful. I've now posted a revision to SSRN: papers.ssrn.com/sol3/papers....
Lol, Grok Build thinks my figures are "Frisch-Waugh theatre". First off, it's Frisch-Waugh-Lovell, Grok.
It's live! Very happy to see competition for that very ridiculous Claude Fable price point.
The clearest result I find is on wages. A standard deviation of remote-work exposure lowers entry wages by about 1.9 percent, including for cohorts that entered before ChatGPT. I do not find a detectable AI wage effect once I control for remote work.
I'm happy to share a new working paper on remote work and AI: Same Occupations, Different Clocks: Separating Remote Work from Generative AI for Labor-Market Entrants www.joshmask.com/s/Mask2026c....
Well if I now have to pay a toll for frontier intelligence, I might as well get the best possible version: openrouter.ai/fusion.
Hit the limit in the middle of a job. 10 minutes later I already have two receipts.
I just posted a new working paper to my website: www.joshmask.com/s/Mask2025.pdf
I pulled up an article in Elsevier and got this notice. I first thought the paper I was reading had been retracted, but it’s one of the papers it cites. This is helpful information, but I think I could be notified in a manner that didn’t make me worry about the paper citing it. They didn’t know.
If you use Chrome with the Pangram plugin enabled, it will auto tag posts as AI or Human written based on Pangram’s rating. I wonder how the AI labs will respond if Pangram becomes a default.
How my local AI model experiments fare against the current cloud frontier. There’s still a noticeable gap, but these local models would’ve been frontier this time last year. And I was comfortable using AI for a variety of tasks then.