Ethical Violation Location on ETHICSCAFF
66.7PrecisionQwen3-8B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen3-8BModel Family=Qwen3-8B, Variant=w/ ETHICSCAFF (Full)2025.08 | 66.7 | 91.45 | 74.24 | 63.03 | 18.4 | |
| Qwen3-14BModel Family=Qwen3-14B, Variant=w/ ETHICSCAFF (Full)2025.08 | 64.62 | 93.32 | 73.25 | 61.65 | 17.41 | |
| Qwen3-14BModel Family=Qwen3-14B, Variant=w/ ETHICSCAFF (SFT)2025.08 | 63.94 | 93.37 | 72.99 | 61.31 | 17.11 | |
| Llama3.1-8BModel Family=Llama3.1-8B, Variant=w/ ETHICSCAFF2025.08 | 63.64 | 93.55 | 72.7 | 61.19 | 17.13 | |
| Qwen3-8BModel Family=Qwen3-8B, Variant=Few-shot + CoT2025.08 | 63.18 | 94.06 | 72.7 | 61.16 | 16.7 | |
| Qwen3-14BModel Family=Qwen3-14B, Variant=Few-shot + CoT2025.08 | 62.48 | 94.4 | 71.79 | 59.87 | 15.88 | |
| GPT-4oModel Family=Closed-Source Models, Variant=Base2025.08 | 62.19 | 89.93 | 69.48 | 57.63 | 13.68 | |
| Qwen3-14BModel Family=Qwen3-14B, Variant=CoT2025.08 | 61.96 | 91.66 | 69.88 | 57.57 | 12.37 | |
| GPT-4oModel Family=Closed-Source Models, Variant=RAG2025.08 | 61.9 | 95.17 | 71.76 | 59.59 | 14.1 | |
| Qwen3-14BModel Family=Qwen3-14B, Variant=RAG2025.08 | 61.23 | 97.15 | 72.18 | 59.89 | 13.78 | |
| Qwen3-8BModel Family=Qwen3-8B, Variant=CoT2025.08 | 60.98 | 91.04 | 69.52 | 57.38 | 12.77 | |
| DeepSeek-V3-685BModel Family=Open-Source Models, Variant=Base2025.08 | 60.71 | 97.05 | 71.85 | 59.66 | 13.57 | |
| Qwen3-14BModel Family=Qwen3-14B, Variant=Base2025.08 | 60.66 | 93.62 | 70.26 | 57.86 | 12.5 | |
| Qwen3-14BModel Family=Qwen3-14B, Variant=Few-shot2025.08 | 60.54 | 96.41 | 71.34 | 59.01 | 13.46 | |
| Llama3.1-8BModel Family=Llama3.1-8B, Variant=RAG2025.08 | 59.95 | 96.49 | 71.25 | 58.81 | 12.3 | |
| Llama3.1-8BModel Family=Llama3.1-8B, Variant=Few-shot2025.08 | 59.75 | 94.92 | 70.49 | 57.86 | 10.57 | |
| Claude-Sonnet-4Model Family=Closed-Source Models, Variant=Base2025.08 | 59.57 | 94.67 | 70.04 | 58.04 | 13.14 | |
| Llama3.1-8BModel Family=Llama3.1-8B, Variant=Base2025.08 | 59.4 | 94.31 | 69.91 | 57.28 | 10.01 | |
| Qwen3-8BModel Family=Qwen3-8B, Variant=RAG2025.08 | 58.01 | 94.85 | 68.84 | 56.23 | 11.86 | |
| Qwen3-8BModel Family=Qwen3-8B, Variant=Few-shot2025.08 | 56.25 | 95.64 | 68 | 54.98 | 9.94 | |
| Qwen3-8BModel Family=Qwen3-8B, Variant=w/ ETHICSCAFF (SFT)2025.08 | 55.99 | 92.26 | 66.85 | 55.26 | 12.29 | |
| Qwen3-8BModel Family=Qwen3-8B, Variant=Base2025.08 | 54.78 | 60.76 | 53.56 | 40.52 | 3.87 |