Safety Evaluation on Jailbreaking Scenarios (Origin)
5.3Unsafe RatioDeepSeek-V3
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DeepSeek-V3risk control system=true2025.01 | 5.3 | 25.4 | |
| DeepSeek-R1risk control system=true2025.01 | 8.5 | 27.3 | |
| o1release date=2024-12-172025.01 | 9 | 50.4 | |
| Claude-3.7-Sonnet2025.01 | 10.7 | 3.6 | |
| Qwen2.5 Instructparameters=72B2025.01 | 13.8 | 5.4 | |
| DeepSeek-V3risk control system=false2025.01 | 17.6 | 8.1 | |
| GPT-4orelease date=2024-05-132025.01 | 22 | 17.1 | |
| DeepSeek-R1risk control system=false2025.01 | 25.2 | 5.6 |