Inference Attack on MMLU
0.8411AUCCIA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CIAMAS Framework=AGP2026.04 | 0.8411 | 79.42 | 75.67 | |
| CIAMAS Framework=G-Designer2026.04 | 0.8324 | 81.47 | 77.61 | |
| CIAMAS Framework=ARG-Designer2026.04 | 0.8227 | 79.31 | 74.58 | |
| Gemini-2.5-ProMAS Framework=ARG-Designer2026.04 | 0.6888 | 77.95 | 44.82 | |
| Gemini-2.5-ProMAS Framework=G-Designer2026.04 | 0.6869 | 79 | 36.97 | |
| GPT-5MAS Framework=AGP2026.04 | 0.6577 | 69.74 | 54.72 | |
| Gemini-2.5-ProMAS Framework=AGP2026.04 | 0.6324 | 59.35 | 52.17 | |
| Llama-3.1-8B-InstructMAS Framework=G-Designer2026.04 | 0.5995 | 71.57 | 41.56 | |
| GPT-5MAS Framework=ARG-Designer2026.04 | 0.5879 | 75.45 | 40.94 | |
| GPT-5MAS Framework=G-Designer2026.04 | 0.5833 | 80.33 | 30.84 | |
| Mistral-7B-Instruct-v0.2MAS Framework=G-Designer2026.04 | 0.5192 | 75.55 | 28.89 | |
| Mistral-7B-Instruct-v0.2MAS Framework=AGP2026.04 | 0.5076 | 65.79 | 42.46 | |
| Llama-3.1-8B-InstructMAS Framework=ARG-Designer2026.04 | 0.4912 | 65.11 | 38.99 | |
| Llama-3.1-8B-InstructMAS Framework=AGP2026.04 | 0.4812 | 65.81 | 47.86 | |
| Mistral-7B-Instruct-v0.2MAS Framework=ARG-Designer2026.04 | 0.4539 | 64.65 | 26.49 |