Nature as the RL Environment

A new lab model treats real experiments as the feedback loop for AI 'scientists': autonomous labs generate high‑signal, proprietary data—including negative results—and let models act on the world, not just tokens. This closes the frontier data gap as internet text saturates and targets hard problems like high‑temperature superconductors and heat‑dissipation materials. — If AI research shifts from scraped text to real‑world experimentation, ownership of lab capacity and data rights becomes central to scientific progress, IP, and national competitiveness.