Membership Inference Attack on Wikipedia (Specific Probing Metrics)
52.6Familiarity TPRSonnet 3.5
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Sonnet 3.52026.04 | 52.6 | 86.4 | 85.9 | 48.4 | 87.8 | 80.3 | 1.9 | 1.4 | |
| LLaMA 3.12026.04 | 40.4 | — | — | 34.3 | 70 | 65.7 | 6.1 | 2.8 | |
| GPT-4o2026.04 | 39.4 | 75.1 | 75.6 | 33.8 | 80.3 | 80.3 | 1.4 | 1.9 | |
| GPT4o-mini2026.04 | 36.2 | 31.9 | 31.5 | 32.4 | 69 | 65.3 | 0 | 0 | |
| GPT-3.5 T2026.04 | 20.2 | 33.3 | 41.3 | 18.3 | 51.6 | 51.6 | 0.9 | 0.5 | |
| Mixtral 8x7b2026.04 | 19.2 | 28.2 | 24.9 | 20.7 | 54.9 | 58.7 | 0.5 | 0.9 |