Classification on AI Persona
41Power-SeekingPrompt
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PromptBackbone=Llama-2-13B2026.02 | 41 | 44 | 58.5 | |
| RagBackbone=Llama-2-13B2026.02 | 45.5 | 50.5 | 64.5 | |
| PromptBackbone=Llama-3-8B2026.02 | 45.5 | 47.5 | 63.5 | |
| WandaBackbone=Llama-2-13B2026.02 | 51.5 | 54.5 | 89 | |
| SparseBackbone=Llama-2-13B2026.02 | 52 | 58.5 | 84.5 | |
| RagBackbone=Llama-3-8B2026.02 | 52.5 | 59 | 72.5 | |
| Wanda with Contrastive PruningBackbone=Llama-2-13B2026.02 | 54 | 66 | 95 | |
| Sparse with Contrastive PruningBackbone=Llama-2-13B2026.02 | 56.5 | 64.5 | 96 | |
| WandaBackbone=Llama-3-8B2026.02 | 57 | 64 | 86 | |
| Wanda with Contrastive PruningBackbone=Llama-3-8B2026.02 | 58.5 | 69.5 | 94 | |
| SparseBackbone=Llama-3-8B2026.02 | 59.5 | 65.5 | 87.5 | |
| Sparse with Contrastive PruningBackbone=Llama-3-8B2026.02 | 60.5 | 66 | 96 | |
| SFTBackbone=Llama-2-13B2026.02 | 64 | 71 | 97.5 | |
| SFTBackbone=Llama-3-8B2026.02 | 69.5 | 71.5 | 98.5 |