Speculative Advice on MLLM Secondary Risk Multimodal (test)
48.61ASR (Cosine Similarity >= 80%)SecLens
Evaluation Results
| Method | Links | |
|---|---|---|
| SecLensModel=GPT-4o2025.06 | 48.61 | |
| SecLensModel=GPT-4-turbo2025.06 | 45.91 | |
| MCTSModel=GPT-4o2025.06 | 45.89 | |
| SecLensModel=Claude-3.72025.06 | 42.14 | |
| MCTSModel=GPT-4-turbo2025.06 | 42.03 | |
| MCTSModel=Claude-3.72025.06 | 37.65 | |
| TuningModel=GPT-4o2025.06 | 36.23 | |
| TuningModel=GPT-4-turbo2025.06 | 31.68 | |
| TuningModel=Claude-3.72025.06 | 30.18 | |
| RandomModel=GPT-4-turbo2025.06 | 11.75 | |
| RandomModel=GPT-4o2025.06 | 11.57 | |
| RandomModel=Claude-3.72025.06 | 5.39 |