Strategic Model Releases in the Generative AI Industry
This project studies how leading AI firms time model releases in response to rival
product improvements. I combine a simple dynamic model of leader–follower
competition with empirical evidence from public model release dates and LMArena
performance scores. The project asks whether frontier AI firms “follow”
each other’s releases, how quality improvements affect competitive responses,
and how public benchmarks shape strategic incentives.
Methods: innovation, dynamic games,
benchmark data, Python, MatLab.
Benchmarking and Innovation in Generative AI
This project studies whether generative AI firms selectively report benchmark scores to compete,
and whether existing AI benchmarks shape the direction of firms’ innovation. I examine the relationship
between technical benchmark scores across different categories and human evaluations to assess whether
frontier AI firms respond strategically to benchmark incentives by competing on, adopting, and selectively reporting
benchmark results. More broadly, this research asks what an optimal benchmarking structure should look like in the generative AI industry.
Methods: empirical IO, innovation, AI benchmarking, Python.