Expert marketplaces teach AI

A new class of firms (e.g., Mercor) recruits highly paid domain experts — poets, critics, clinicians, economists — to build rubrics, evaluation datasets, and fine‑grading protocols that train and validate frontier AI models. These marketplaces monetize human expertise by turning one‑time expert judgments into scalable model improvements and diagnostics. — If this model scales, it will reshape labor markets (premium pay for ephemeral evaluative work), concentrate who controls evaluation standards for AI, create new governance risks around provenance and conflict of interest, and change how we regulate training data and model audits.