Combining human rankings with AI scores improves evaluation when true answers are unknown

Original: Aggregate-then-Calibrate for Human-centered Assessment with Theoretical Guarantees

Source: arxiv.org ↗