Study Finds AI Capabilities Growing as 'Rising Tides' Rather Than Sudden 'Crashing Waves'

Summary

A comprehensive arXiv research study presents preliminary findings from over 17,000 worker evaluations of AI capabilities across more than 3,000 labor market tasks derived from the U.S. Department of Labor's O*NET classification system. The research challenges recent narratives of sudden AI breakthroughs, instead documenting continuous, broad-based capability improvements across task categories—what researchers term 'rising tides' rather than 'crashing waves.'

The study evaluated text-based labor tasks and found AI model performance on work typically requiring 3-4 hours of human effort improved from approximately 50% success rate in Q2 2024 to 65% by Q3 2025. Projecting forward, researchers estimate large language models could achieve 80-95% success rates on most text-related tasks by 2029, assuming current improvement trajectories continue. Critically, the research emphasizes that AI capability gains will likely precede labor market disruption, as actual economic impact depends on organizational adoption decisions that operate on much longer timelines.

• Research-backed methodology evaluating 3,000+ actual job tasks provides empirical grounding for AI automation projections

Editorial Opinion

This research offers crucial empirical grounding for understanding AI's labor market trajectory by systematizing evaluation across thousands of actual job tasks. The 'rising tides' framework presents a more nuanced picture than recent dramatic claims, suggesting broad but gradual workforce disruption driven by organizational decisions rather than technical capability alone. Yet the projection to 80-95% success rates by 2029 still represents transformative impact for text-based work, making preparation for labor market transition essential regardless of adoption pace.