Thanks Erik!!!

3. There is substantial multi-homing, with users of the same app employing a mix of models.

This is still very preliminary work, so comments are welcome!

2. New model releases differ in the degree to which they cause substitution from existing models or expand the market.

Through the use of three case studies of model releases (Sonnet 3.7, Gemini 2.0 Flash, Gemini 2.5 Pro), I document three stylized facts:
1. New models are adopted quickly, with increased demand stabilizing within a few weeks.

I'm just using their GUI. The papers submitted to MS are mostly available online already, so am not worried about leakage.

Write a referee report for Management Science based on this paper. Take the perspective of an economist who cares about causal methods.

But what if we're interested in studying AIs?

I think your example of the rent curve is great, but doesn't represent much of social science research (based on things I referee). A lot of (most?) empirical papers are driven by hammer and nail thinking (something changed, run a DID), and not by careful descriptive work followed by modeling.

Thanks, right back at ya!

Write down a vague idea of a model and it will write the math. Tell it to translate the math into code. Tell it to make it run quickly and to try other variations. I think the key speedups are actually in steps 2 and 3, which prevented most from using them.

Will be a huge complement to most of it, but the efficiency improvements for tasks in IO for me have been immense. So maybe IO / anything with structural modeling. But also, it might now be cheap to include a structural model in any type of paper, making econ papers even longer.

I should mention that I'm a fan of one of the authors, Byrne Hobart, and his newsletter The Diff, which provides interesting analysis at the intersection of tech and finance.