This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.
Back to datasetsKey Variables Task exposure scores (0-1 scale); human and GPT-4 ratings; 50% time reduction threshold; complementary tool exposure AI/Tech Tracking LLM/GPT capabilities assessment; both direct model and LLM-powered software Access Details Available from authors; data appendix in paper Notes Uses both human annotators and GPT-4 classification; high correlation between human and AI ratings; distinguishes model-only vs tool-enhanced exposure
Eloundou et al. (2023)
GPTs are GPTs: An early look at the labor market impact potential of large language models
AI-focusedPublicWorker-side
Visit Dataset- Specific Type
- AI exposure measure
- Dataset Type
- Cross-sectional
- Institution
- OpenAI; University of Pennsylvania
- Institution Type
- Academia; AI Lab
- Level of Focus
- Task
- Most Granular Level
- O*NET task level (DWAs)
- Perspective
- Worker-side
- Time Coverage
- 2023
- Frequency
- One-time static snapshot
- Sample Size
- 1016 occupations; 19265 tasks
- Geographic Detail
- National (US)
- Occupational Classification
- O*NET-SOC 2019
- Industrial Classification
- Not specified
- Other Classification
- Task-level DWAs (Detailed Work Activities)
Key Papers
The Rapid Adoption of Generative AI
Bick, Blandin, Deming (2024)
The Unequal Adoption of ChatGPT Exacerbates Existing Inequalities Among Workers
Humlum, Vestergaard (2024)
Which Economic Tasks Are Performed with AI? Evidence from Millions of Claude Conversations
Handa, Tamkin, McCain, Huang (2025)
Bick et al. (2024); Humlum & Vestergaard (2024); Handa et al. (2025); Eloundou et al. (2023) AEA Papers and Proceedings