Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Language Models at Work in China's Labor Market

Qin Chen, Jinfeng Ge, Huaqing Xie, Xingcheng Xu, Yanqing Yang

TL;DR

This paper quantifies how large language models (LLMs) may reshape China's labor market by mapping occupation- and industry-level exposure using GPT-4, GLM, and InternLM as classifiers, and validating against expert judgments. It finds pronounced heterogeneity: higher exposure in educated, high-wage, and non-routine cognitive tasks, with exposure correlating to wage and experience premia and a potential shift in labor demand through rising vacancy shares. To explain these patterns, the authors develop an entropy-based information-theoretic extension of a task-based framework, distinguishing LLMs from traditional AI via KL-divergence and neural scaling laws, and showing that LLMs privilege data-intensive, complex tasks, potentially reducing wage inequality under certain conditions. The work combines rich Chinese occupational and vacancy data with a stylized model to derive policy-relevant implications, notably the need for demand-tracking systems and workforce training to harness AI benefits while mitigating disruption.

Abstract

This paper explores the potential impacts of large language models (LLMs) on the Chinese labor market. We analyze occupational exposure to LLM capabilities by incorporating human expertise and LLM classifications, following the methodology of Eloundou et al. (2023). The results indicate a positive correlation between occupational exposure and both wage levels and experience premiums at the occupation level. This suggests that higher-paying and experience-intensive jobs may face greater exposure risks from LLM-powered software. We then aggregate occupational exposure at the industry level to obtain industrial exposure scores. Both occupational and industrial exposure scores align with expert assessments. Our empirical analysis also demonstrates a distinct impact of LLMs, which deviates from the routinization hypothesis. We present a stylized theoretical framework to better understand this deviation from previous digital technologies. By incorporating entropy-based information theory into the task-based framework, we propose an AI learning theory that reveals a different pattern of LLM impacts compared to the routinization hypothesis.

Large Language Models at Work in China's Labor Market

TL;DR

This paper quantifies how large language models (LLMs) may reshape China's labor market by mapping occupation- and industry-level exposure using GPT-4, GLM, and InternLM as classifiers, and validating against expert judgments. It finds pronounced heterogeneity: higher exposure in educated, high-wage, and non-routine cognitive tasks, with exposure correlating to wage and experience premia and a potential shift in labor demand through rising vacancy shares. To explain these patterns, the authors develop an entropy-based information-theoretic extension of a task-based framework, distinguishing LLMs from traditional AI via KL-divergence and neural scaling laws, and showing that LLMs privilege data-intensive, complex tasks, potentially reducing wage inequality under certain conditions. The work combines rich Chinese occupational and vacancy data with a stylized model to derive policy-relevant implications, notably the need for demand-tracking systems and workforce training to harness AI benefits while mitigating disruption.

Abstract

This paper explores the potential impacts of large language models (LLMs) on the Chinese labor market. We analyze occupational exposure to LLM capabilities by incorporating human expertise and LLM classifications, following the methodology of Eloundou et al. (2023). The results indicate a positive correlation between occupational exposure and both wage levels and experience premiums at the occupation level. This suggests that higher-paying and experience-intensive jobs may face greater exposure risks from LLM-powered software. We then aggregate occupational exposure at the industry level to obtain industrial exposure scores. Both occupational and industrial exposure scores align with expert assessments. Our empirical analysis also demonstrates a distinct impact of LLMs, which deviates from the routinization hypothesis. We present a stylized theoretical framework to better understand this deviation from previous digital technologies. By incorporating entropy-based information theory into the task-based framework, we propose an AI learning theory that reveals a different pattern of LLM impacts compared to the routinization hypothesis.
Paper Structure (36 sections, 41 equations, 18 figures, 1 table)

This paper contains 36 sections, 41 equations, 18 figures, 1 table.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Methodology of Exposure Scoring and Data Collection
  4. Data on Occupations in China
  5. Data on Wages and Vacancies
  6. Methodology of Exposure Scoring
  7. Results and Assessment of Impacts of LLMs on China's Labor Market
  8. Summary Statistics
  9. Occupational Exposure
  10. Composite Occupational Exposure
  11. Industrial and Demographic Exposure
  12. Vacancies
  13. The Model
  14. The Benchmark Model
  15. The Environment
  16. ...and 21 more sections

Figures (18)

  • Figure 1: Exposure Score of LLMs and Expert: corr.=0.65.
  • Figure 2: Exposure Score on Medium-category Level: LLMs.
  • Figure 3: Salary and Exposure Score (Medium-category): LLMs.
  • Figure 4: Annual Wage Growth Rate and Exposure Score (Medium-category): LLMs.
  • Figure 5: Composite Occupation and LLMs exposure.
  • ...and 13 more figures