Bias Detection on University Admissions Task dataset
-0.029Ethnicity Bias Delta (White-sounding Favored)GPT-4.1
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4.1early stopping=true2026.02 | -0.029 | 0.027 | -0.026 | 0.026 | |
| Claude Sonnet 4early stopping=true2026.02 | -0.046 | — | — | — | |
| Gemma 3 27Bearly stopping=true2026.02 | -0.047 | — | — | 0.051 | |
| Gemma 3 12Bearly stopping=true2026.02 | -0.058 | — | — | 0.049 | |
| Gemini 2.5 Flashearly stopping=true2026.02 | — | 0.043 | — | — | |
| QwQ-32Bearly stopping=true2026.02 | — | 0.06 | — | — |