Of Jobs and Men: An AI-Assisted Approach for Studying Skills Mismatch in Hong Kong

Recent advancements in generative AI and natural language processing have significantly enhanced the ability of social scientists to analyze large-scale observational data. In this study, I employ ChatGPT to process millions of online job postings in Hong Kong, for the purpose of assessing real-time labor market demands and constructing skills mismatch measures. This AI-assisted approach surpasses the limitations of traditional machine learning methods that require manual data labeling, a process that is time-consuming, labor-intensive, and cognitively demanding. The study demonstrates that general foundation models like GPT-4 can handle job posting data impressively without explicit pre-training. Notably, GPT-3.5, when fine-tuned, achieves close to GPT-4-level performance on automated data labeling and skill identification benchmarks, at just one-tenth of the cost. The study suggests that large language models can attain state-of-the-art performance on information extraction and classification tasks, outperforming smaller supervised models and even well-trained human experts in terms of efficiency and accuracy.