Causal effect of LLMs on labor-market outcomes

Establish the causal effect of large language models (LLMs) on labor-market outcomes, rather than correlations, by identifying the impact of LLM adoption on employment, unemployment risk, job entry rates, wages, and job-search durations.

Background

The paper triangulates evidence from unemployment insurance records, LinkedIn career histories, and university syllabi to analyze timing and patterns of labor-market changes around ChatGPT’s launch. While documenting deterioration in AI-exposed occupations beginning prior to ChatGPT and continued value of LLM-relevant education, the authors explicitly note that they did not identify causal effects of LLMs on labor-market outcomes.

They further suggest that future work with direct measures of LLM adoption and linked worker–firm data will be needed to separate displacement from productivity gains and to understand distributional impacts, underscoring the need for causal identification.