quick takes 3: productivity edition

some revisions!!!

In my last edition of Quick Takes I talked about large language models (LLMs) and productivity. I want to make some revisions to my predictions there, because I thought about the situation a little bit more. I’m going to edit that post to note that the QTs have been updated and are no longer valid. Note that these updates make the QTs spicier 🌶️🌶️🌶️.

To recap, here’s productivity growth since 1947 according to BLS. Also note that Q2 2023 productivity growth was an impressive 3.7%, the unemployment rate is low, and anecdotally employers complain a lot that they can’t find employees.

Basically, I think this scarce-worker situation is exactly the kind of situation where firms have an incentive to invest in labor-saving productivity improvements. So after mulling it over for a few days I’m actually more bullish about productivity growth in the next decade than I was a few days ago.

At the same time, I do not think LLMs will be a primary factor in that productivity growth. So here are my updated QTs (note these apply to the USA only!!)

• Productivity growth will be above the 1.4% average of the the 2007-2023 period in the 2024-2029 period (90%)

• Productivity growth will be above 2% over the 2024-2029 period (65%)

• LLMs will not be a consensus driver of this higher productivity growth (70%)

• Generative AI in general, such as image generation models or text-to-image models, will not be a consensus driver of this higher productivity growth (65%)

Where will productivity growth come from? Idk, probably just like doing shit better. We could do construction a lot more efficiently, for instance. You don’t need generative AI to do that, you need regulatory reform.