Could Savannah be the Next San Jose? The Downstream Effects of Large Language Models

Abstract

The anticipated production shock created by large language models (LLMs) may significantly alter the geographic structure of U.S. labor demand. We use established estimates of LLM occupational exposure to develop a map of LLM impacts and find that urban, highly educated coastal metro areas are the most affected. We then consider the downstream effects as places and people adjust to changing employment opportunities. We identify three avenues of adjustment that emerged during the post-1980 decline in manufacturing employment, which we apply to the present situation. Combining our mapping with these avenues, we predict that displaced college graduates will migrate towards smaller, lower exposure urban centers including Rochester, New York and Savannah, Georgia, that demand for a four-year college education will fall, and that the migration patterns and politics of affected persons will dampen rather than exacerbate political polarization-provided that government can successfully moderate the pace of change.

Keywords: production shocks, manufacturing employment, geographic polarization, large language models, automation, downstream effects

JEL Classification: N12, O14, O33, O51