AI Progress Driven More by Model Improvements Than Human Data Labeling, Says Researcher

RyanGreenblatt · x · 2026-08-13

Ryan Greenblatt clarifies his perspective on what drives AI progress, arguing that improvements in RL environments have largely come from better AI models rather than scaling up spending on human experts. He notes that compared to the massive costs associated with researchers and experiment compute, the spending on having humans label individual data points is relatively small and not the primary driver of scaling.

Related event: Ex-OpenAI Researcher: Human Data Not Core AI Bottleneck(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →