Misaligned Task Learning on Security In-domain
2.1MisalignmentPersona Vectors
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Persona VectorsModel=Qwen2.5-32B2025.08 | 2.1 | 3.6 | |
| KLModel=Qwen2.5-32B2025.08 | 6.33 | 0.67 | |
| InterleavingModel=Qwen2.5-32B2025.08 | 21.27 | 40.2 | |
| Interleaving++Model=Qwen2.5-32B2025.08 | 22.03 | 40.07 | |
| Interleaving+Model=Qwen2.5-32B2025.08 | 22.2 | 39.7 | |
| MisalignedModel=Qwen2.5-32B2025.08 | 25.33 | 36 |