This is a great roundup - thanks for putting it together. I would recommend that you write new posts when new studies come out in addition to keeping it updated, just because I find that Substack posts don't really work well as living resources. Plus these studies become irrelevant so quickly that the new stuff is disproportionately important. I really wonder what results we'd see if they re-ran that 2023 Kenyan study with current SOTA models - I can't help but feel like doing it with an Opus 4.6 based model would yield better results.
Fascinating research on the adoption margin. The finding that AI usage is concentrated among higher-skilled workers echoes what I explore in my article about the precision of machine-instruction vs. human briefing. We craft detailed prompts for AI but rarely brief humans with that same clarity. Worth examining: https://creatism.substack.com/p/we-prompt-machines-better-than-we?r=177ve
This is a great roundup - thanks for putting it together. I would recommend that you write new posts when new studies come out in addition to keeping it updated, just because I find that Substack posts don't really work well as living resources. Plus these studies become irrelevant so quickly that the new stuff is disproportionately important. I really wonder what results we'd see if they re-ran that 2023 Kenyan study with current SOTA models - I can't help but feel like doing it with an Opus 4.6 based model would yield better results.
Yeah that’s a good idea. I might write up short posts with a link as the conclusions evolve.
Are you only interested in genAI? Because we have a few contributions on earlier types of AI, first with adoption: https://onlinelibrary.wiley.com/doi/10.1111/jems.12576 and then both adoption and productivity: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5036270. Lots of detailed breakdowns on use, including use in production, at the establishment and firm level.
I thought this was a legitimately good breakdown of adoption and productivity gains. Thank you for putting this together!
Thanks!
Thanks for writing the definitive article on this topic! Finally someone reporting on all we know, not just one study
🙏
Fascinating research on the adoption margin. The finding that AI usage is concentrated among higher-skilled workers echoes what I explore in my article about the precision of machine-instruction vs. human briefing. We craft detailed prompts for AI but rarely brief humans with that same clarity. Worth examining: https://creatism.substack.com/p/we-prompt-machines-better-than-we?r=177ve
Calvino and Fontanelli also producing nice non-US insights: https://www.sciencedirect.com/science/article/pii/S0927537125000788 and https://www.oecd.org/en/publications/a-portrait-of-ai-adopters-across-countries_0fb79bb9-en.html and https://www.ifo.de/en/cesifo/publications/2024/working-paper/ai-users-are-not-all-alike-characteristics-french-firms-buying-and
Thank you we will add this!!
Thanks! Raff told me at Summer Institute (or possibly Winter Orgs meeting?) to send things to you, so it's my bad for not following up!
Strong breakdown of the mechanics and the discipline behind it. The emphasis on volume as practice, and comments as leverage, is spot on.
Where I think most people stumble isn’t growth, but adoption. Posting daily for 90 days builds skill. It doesn’t automatically build signal.
If your positioning isn’t tight and your perspective isn’t differentiated, you just get louder, not clearer.
Curious how you think about the trade-off between quantity and distinctiveness once someone passes the beginner stage?
It would be interesting to better understand what the ai skill are, whether they are stable and confer advantage over time. https://blog.nappisite.com/p/what-are-ai-skills