Compute‑Productivity Scaling Law

Measure and model how increases in LLM training compute map to real‑world professional productivity (e.g., percent task‑time reduction) using preregistered, role‑specific experiments. Early evidence suggests roughly an 8% annual task‑time reduction per year of model progress, with compute accounting for a majority of measurable gains and agentic/tooled workflows lagging behind. — If robust, a compute→productivity scaling law anchors macro forecasts, labor policy, and industrial strategy—turning abstract model progress into quantifiable economic expectations and regulatory triggers.