Utility assessment on MMLU-Pro
16.5Personalization Bias (PB)Identity-Robust Generation
Evaluation Results
| Method | Links | |
|---|---|---|
| Identity-Robust GenerationModel=Llama3.3-70B2026.01 | 16.5 | |
| Identity-Robust GenerationModel=Qwen32026.01 | 19.6 | |
| Identity-Robust GenerationModel=gpt-oss-20B2026.01 | 25.2 | |
| Vanilla GenerationModel=Llama3.3-70B2026.01 | 35 | |
| Prompt SteeringModel=Llama3.3-70B2026.01 | 36.6 | |
| Prompt SteeringModel=Qwen32026.01 | 49 | |
| Vanilla GenerationModel=Qwen32026.01 | 52.3 | |
| Prompt SteeringModel=gpt-oss-20B2026.01 | 60.8 | |
| Vanilla GenerationModel=gpt-oss-20B2026.01 | 448.3 |