Testing AI Output, Not Building It, Is the Real Bottleneck

Original: Why Evaluation Is the Bottleneck: A Structural Account of Human Judgment in Production AI

Source: x.com ↗