Biased Prompt Discovery on Black-biased prompts
4Black Bias ProportionPEZ
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PEZEvaluation Model=Base2025.12 | 4 | 2,415 | 3,000 | |
| LLMEvaluation Model=Base2025.12 | 6 | 50 | 0 | |
| BGPSEvaluation Model=Base, lambda (λ)=102025.12 | 7 | 50 | 0 | |
| LLM (biased)Evaluation Model=Base2025.12 | 9 | 58 | 400 | |
| Manually curatedEvaluation Model=Base2025.12 | 10 | 100 | 0 | |
| LLM (biased)Evaluation Model=FT2025.12 | 14 | 58 | 0 | |
| LLMEvaluation Model=FT2025.12 | 16 | 50 | 0 | |
| BGPSEvaluation Model=FT, lambda (λ)=102025.12 | 17 | 49 | 0 | |
| BGPSEvaluation Model=FT, lambda (λ)=1002025.12 | 19 | 96 | 100 | |
| LLM (biased)Evaluation Model=DL2025.12 | 20 | 172 | 0 | |
| Manually curatedEvaluation Model=FT2025.12 | 23 | 100 | 0 | |
| LLMEvaluation Model=DL2025.12 | 24 | 71 | 100 | |
| BGPSEvaluation Model=DL, lambda (λ)=102025.12 | 24 | 121 | 100 | |
| BGPSEvaluation Model=DL, lambda (λ)=1002025.12 | 24 | 213 | 200 | |
| PEZEvaluation Model=FT2025.12 | 27 | 2,427 | 1,000 | |
| BGPSEvaluation Model=Base, lambda (λ)=1002025.12 | 32 | 108 | 1,000 | |
| Manually curatedEvaluation Model=DL2025.12 | 40 | 100 | 0 | |
| PEZEvaluation Model=DL2025.12 | 57 | 1,754 | 4,500 |