Penetration of Large Language Models across Demographic and Occupational Subgroups

Riccardo Omenti , University of Bologna
Nicola Barban, University of Bologna
Saverio Minardi, University of Bologna
Maria Ludovica Isani, University of Bologna

This paper aims to investigate the penetration of large language models across demographic, geographic, and occupational groups in the United States through the development of new ad-hoc indicators. By relying on evaluations from LLM, we quantify on a 0–100 scale the extent to which specific tasks can be complemented by LLMs, aggregating these task-level scores to the occupation level with task-relevance weights. Merging these occupational scores with micro-data from the American Community Survey, we construct two indicators: the LLM Penetration Index and the LLM High-Exposure Index. Preliminary findings show distinct gender patterns: exposure peaks for women with some college education and median earnings, while among men, it is highest for those with a doctoral degree and top wages. These results suggest that LLM diffusion may exacerbate existing labor market inequalities, amplifying advantages for high-skilled male workers and providing limited gains to women in middle-skilled occupations.

See paper

 Presented in Session 56. Economy, Human Capital and Labor Markets