TLDR: A research paper found that occupations highly exposed to Large Language Models (LLMs) like ChatGPT experienced increased earnings (an average of $89 weekly) but no significant change in unemployment rates in the short term following ChatGPT’s release. This suggests LLMs are currently complementing human work, leading to higher wages due to productivity gains rather than job displacement.
The rapid rise of Large Language Models (LLMs) like ChatGPT since late 2022 has sparked widespread debate about their potential impact on the workforce. While many herald these AI tools as drivers of significant productivity gains, others express concerns about widespread job displacement. A recent research paper delves into these questions, offering a detailed analysis of the short-term effects of LLM adoption on unemployment and earnings across various occupations.
Titled “The (Short-Term) Effects of Large Language Models on Unemployment and Earnings,” this study was conducted by Danqing Chen, Carina Kane, Austin Kozlowski, Nadav Kuniesvky, and James A. Evans. Their work provides crucial insights into how the labor market is initially adjusting to this transformative technology.
Understanding the Study’s Approach
To assess the impact of LLMs, the researchers employed a sophisticated statistical method called Synthetic Difference-in-Differences (SDiD). This approach allowed them to compare outcomes for occupations highly exposed to LLMs against a carefully constructed “synthetic” control group of less exposed occupations. The introduction of ChatGPT in November 2022 served as the key event marking the widespread availability of these tools.
Occupational exposure to LLMs was measured by identifying tasks within each job that could potentially be affected or altered by these AI technologies. This involved analyzing millions of user prompts to LLMs and mapping them to specific tasks outlined in the Occupational Information Network (O*NET). The study utilized monthly data on unemployment and earnings from the Current Population Survey (CPS), spanning from January 2010 to August 2025, to track changes over time.
Key Findings: Wages Up, Jobs Stable
The most striking finding from the research is that workers in occupations with high exposure to LLMs experienced a notable increase in earnings following the introduction of ChatGPT. Specifically, these highly exposed occupations saw an average increase of approximately $89 in weekly earnings (adjusted to 2010 prices). This suggests that LLMs are acting as complements to human skills, enhancing worker productivity and thereby increasing the demand for labor, which translates into higher wages.
Crucially, the study found no significant change in unemployment rates within these same highly exposed occupations. While there were minor fluctuations, the overall mean effect on unemployment was negligible, indicating that in the short run, LLMs have not led to widespread job displacement. This pattern suggests that initial labor market adjustments to LLMs are primarily occurring through wage increases rather than through workers being reallocated or losing their jobs.
The researchers interpret these results as consistent with a “complementarity story,” where LLMs boost workers’ productivity. In the short term, the labor market adjusts by offering higher wages because the supply of labor across occupations is relatively inelastic, meaning it doesn’t quickly change in response to new technological demands.
Also Read:
- AI-Driven Job Growth Outpaces Workforce Skills, New Reports Highlight Tech Market Shifts
- Decoding Slang: How AI’s Informal Language Differs from Human Expression
Heterogeneity and Future Implications
While the average effects were positive for earnings and neutral for unemployment, the study also highlighted that the impact of LLMs is not uniform across all occupations. Some occupations experienced even greater earnings gains, while a smaller subset saw declines. This heterogeneity underscores that LLMs interact differently with the specific tasks and skill requirements of various jobs.
The paper acknowledges that its exposure metric reflects the potential applicability of LLMs rather than direct evidence of actual adoption. Therefore, the estimates can be seen as an “intent-to-treat” effect, capturing the average short-run response to the availability of LLMs. This approach helps avoid biases that might arise from measuring exposure based on actual adoption, which could be correlated with other economic factors.
In conclusion, this research offers early, compelling evidence that the immediate impact of LLMs on the labor market is characterized by rising wages in exposed occupations, without a corresponding increase in unemployment. This suggests a short-term scenario where AI augments human capabilities, leading to productivity gains that benefit workers through higher pay. For a deeper dive into the methodology and detailed results, you can access the full research paper here.
