Active Testing on GPQA Diamond
22Labeling BudgetActive Testing via Approximate Neyman Allocation
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Active Testing via Approximate Neyman AllocationSurrogate Model=GPT-5.4 Nano, Target Model=GPT-5.42026.05 | 22 | 26.7 | |
| Uniform SamplingSurrogate Model=GPT-5.4 Nano, Target Model=GPT-5.42026.05 | 30 | — |