Knowledge Graph
Silytics Knowledge Graph
See how models, datasets, papers and organizations connect. Every link comes from public metadata: papers named on a model card, datasets and base models named in its tags, and the organization that published it. Click any dot to re-center; the smaller outer ring is two steps away.
Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration
Open entry pageModelDatasetPaperRepository
Connections
- sbussiso/Qwen2.5-0.5B-abliterated Modelcited by
- The Geometry of Refusal: Why Post-Hoc Safety Is Fragile and Pretraining-Time Safety Persists Papertwo steps away
- Qwen/Qwen2.5-0.5B-Instruct Modeltwo steps away
- Abliteration Is Not a Scalpel: Off-Target Effects of Refusal Removal on Decision Disposition Across Model Families Papertwo steps away
Relationships are extracted automatically from public metadata and may be incomplete or wrong. Absence of a link does not mean absence of a relationship. Intelligence · Compare models