Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AI

Faster experiments do not automatically make an entire research program progress equally fast. In their discussion of AI self-improvement, Nathan Lambert and JS Denain, who leads Epoch AI's Insights Team, distinguish speeding up individual researchers from accelerating a lab's total output.

Lambert argues that testing ideas can become much faster while building scientific understanding remains slower. Denain adds that computing resources and other constraints can limit the gain even if individual researchers improve greatly. The distinction develops Lambert's earlier argument about research bottlenecks, also set out in his essay. Their discussion offers a way to assess self-improvement forecasts by asking which stage accelerates and what still constrains the whole process.