Is Abliteration Getting Harder? Not Uniformly: Refusal Removal by Direction-Based Ablation and ARA in Heretic on the Qwen3.8, Gemma 4, Olmo 3.1 and Mistral Small 3.2 LLMs Versus Their Predecessors
Summary
This study tests how hard it is to strip refusal behavior from Qwen3.8, Gemma 4, Olmo 3.1 and Mistral Small 3.2, comparing them with their predecessors.
Why this matters
Covers model and LLM · Relevant to this topic
Source: arXiv cs.AI · arxiv.org