Bias Discovery on Female-biased prompts
75Female ProportionPEZ
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PEZTarget TTI Model=DL2025.12 | 75 | 1,773 | 100 | |
| BGPSLLM (Prompt Generator)=Llama 3.2 1B, Lambda (λ)=100, Target TTI Model=DL2025.12 | 74 | 59 | 35 | |
| BGPSLLM (Prompt Generator)=Mistral 7B 0.2, Lambda (λ)=100, Target TTI Model=DL2025.12 | 73 | 98 | 13 | |
| BGPSLLM (Prompt Generator)=Qwen3 8B, Lambda (λ)=100, Target TTI Model=DL2025.12 | 70 | 56 | 0 | |
| Manually curatedTarget TTI Model=DL2025.12 | 69 | 96 | 0 | |
| BGPSLLM (Prompt Generator)=Mistral 7B 0.2, Lambda (λ)=10, Target TTI Model=DL2025.12 | 67 | 50 | 1 | |
| BGPSLLM (Prompt Generator)=Mistral 7B 0.2, Lambda (λ)=100, Target TTI Model=Base2025.12 | 67 | 64 | 16 | |
| LLMLLM (Prompt Generator)=Qwen3 8B, Target TTI Model=DL2025.12 | 64 | 53 | 0 | |
| BGPSLLM (Prompt Generator)=Qwen3 8B, Lambda (λ)=10, Target TTI Model=DL2025.12 | 64 | 54 | 0 | |
| PEZTarget TTI Model=FT2025.12 | 62 | 1,964 | 93 | |
| BGPSLLM (Prompt Generator)=Llama 3.2 1B, Lambda (λ)=100, Target TTI Model=Base2025.12 | 61 | 71 | 36 | |
| BGPSLLM (Prompt Generator)=Qwen3 8B, Lambda (λ)=100, Target TTI Model=Base2025.12 | 60 | 71 | 8 | |
| LLM (biased)LLM (Prompt Generator)=Qwen3 8B, Target TTI Model=DL2025.12 | 59 | 58 | 2 | |
| LLMLLM (Prompt Generator)=Llama 3.2 1B, Target TTI Model=DL2025.12 | 58 | 50 | 2 | |
| LLM (biased)LLM (Prompt Generator)=Llama 3.2 1B, Target TTI Model=DL2025.12 | 58 | 55 | 11 | |
| BGPSLLM (Prompt Generator)=Llama 3.2 1B, Lambda (λ)=10, Target TTI Model=DL2025.12 | 58 | 47 | 0 | |
| PEZTarget TTI Model=Base2025.12 | 57 | 1,897 | 100 | |
| LLM (biased)LLM (Prompt Generator)=Mistral 7B 0.2, Target TTI Model=DL2025.12 | 57 | 60 | 0 | |
| LLMLLM (Prompt Generator)=Mistral 7B 0.2, Target TTI Model=DL2025.12 | 56 | 50 | 1 | |
| LLM (biased)LLM (Prompt Generator)=Mistral 7B 0.2, Target TTI Model=Base2025.12 | 52 | 60 | 0 | |
| Manually curatedTarget TTI Model=FT2025.12 | 51 | 96 | 0 | |
| BGPSLLM (Prompt Generator)=Llama 3.2 1B, Lambda (λ)=100, Target TTI Model=FT2025.12 | 51 | 76 | 27 | |
| Manually curatedTarget TTI Model=Base2025.12 | 47 | 96 | 0 | |
| BGPSLLM (Prompt Generator)=Mistral 7B 0.2, Lambda (λ)=100, Target TTI Model=FT2025.12 | 45 | 64 | 12 | |
| BGPSLLM (Prompt Generator)=Llama 3.2 1B, Lambda (λ)=10, Target TTI Model=Base2025.12 | 44 | 54 | 2 | |
| BGPSLLM (Prompt Generator)=Mistral 7B 0.2, Lambda (λ)=10, Target TTI Model=Base2025.12 | 43 | 52 | 0 | |
| BGPSLLM (Prompt Generator)=Qwen3 8B, Lambda (λ)=10, Target TTI Model=Base2025.12 | 43 | 54 | 1 | |
| BGPSLLM (Prompt Generator)=Qwen3 8B, Lambda (λ)=100, Target TTI Model=FT2025.12 | 43 | 76 | 4 | |
| LLM (biased)LLM (Prompt Generator)=Mistral 7B 0.2, Target TTI Model=FT2025.12 | 42 | 60 | 0 | |
| BGPSLLM (Prompt Generator)=Mistral 7B 0.2, Lambda (λ)=10, Target TTI Model=FT2025.12 | 42 | 52 | 0 | |
| LLMLLM (Prompt Generator)=Qwen3 8B, Target TTI Model=Base2025.12 | 40 | 53 | 1 | |
| LLMLLM (Prompt Generator)=Qwen3 8B, Target TTI Model=FT2025.12 | 40 | 53 | 1 | |
| LLMLLM (Prompt Generator)=Llama 3.2 1B, Target TTI Model=Base2025.12 | 40 | 53 | 1 | |
| BGPSLLM (Prompt Generator)=Qwen3 8B, Lambda (λ)=10, Target TTI Model=FT2025.12 | 38 | 54 | 0 | |
| BGPSLLM (Prompt Generator)=Llama 3.2 1B, Lambda (λ)=10, Target TTI Model=FT2025.12 | 38 | 54 | 2 | |
| LLMLLM (Prompt Generator)=Llama 3.2 1B, Target TTI Model=FT2025.12 | 37 | 53 | 1 | |
| LLMLLM (Prompt Generator)=Mistral 7B 0.2, Target TTI Model=FT2025.12 | 36 | 71 | 1 | |
| LLM (biased)LLM (Prompt Generator)=Qwen3 8B, Target TTI Model=Base2025.12 | 35 | 58 | 1 | |
| LLM (biased)LLM (Prompt Generator)=Qwen3 8B, Target TTI Model=FT2025.12 | 35 | 58 | 2 | |
| LLM (biased)LLM (Prompt Generator)=Llama 3.2 1B, Target TTI Model=Base2025.12 | 35 | 55 | 10 | |
| LLM (biased)LLM (Prompt Generator)=Llama 3.2 1B, Target TTI Model=FT2025.12 | 33 | 55 | 5 | |
| LLMLLM (Prompt Generator)=Mistral 7B 0.2, Target TTI Model=Base2025.12 | 27 | 71 | 1 |