Dictionary entry Trending
abliteration
also abliterated model or refusal ablation or refusal-direction ablation
Definition of abliteration
-
: a white-box model-editing technique that identifies an internal activation direction associated with refusal and removes or suppresses that direction in a language model's weights, producing an abliterated model that is less likely to reject harmful requests
- The red team compared the original checkpoint with an abliterated model to measure how much safety came from refusal behavior.
- Abliteration removed many refusals without requiring a new fine-tuning dataset, but it also changed behavior on some ordinary decisions.
Origin & history
The open-weight model community coined abliteration in 2024 as a blend of ablation and obliteration. The technique spread after the 2024 paper Refusal in Language Models Is Mediated by a Single Direction showed that erasing a refusal-related direction in the residual stream could disable refusals across several chat models.
Editorial references
Sources & further reading
- Refusal in Language Models Is Mediated by a Single Direction↗ Primary Research · arXiv · 2024-06-17
- Uncensor any LLM with abliteration↗ Technical Explanation · Hugging Face · 2024-06-13
- There Is More to Refusal in Large Language Models than a Single Direction↗ Follow Up Research · arXiv · 2026-02-02
Test yourself
Which of these is the meaning of abliteration?
Cite this entry
"Abliteration." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/abliteration/. Accessed 28 Oct. 2026.