Can I read Grp-Obliteration: Unaligning Llms With A Single Unlabeled Prompt on EtoBox?

Grp-Obliteration: Unaligning Llms With A Single Unlabeled Prompt by chepurko is a document available to read on EtoBox.

What is Grp-Obliteration: Unaligning Llms With A Single Unlabeled Prompt about?

The document introduces GRP-Obliteration (GRP-Oblit), a novel method for unaligning safety-aligned large language models (LLMs) using a single unlabeled prompt, achieving higher unalignment success rates while preserving model utility compared to existing methods. GRP-Oblit utilizes Group Relative Policy Optimization (GRPO) to effectively remove safety constraints from models, demonstrating its applicability beyond language models to diffusion-based image generation systems. The evaluation shows that GRP-Ob

Author

chepurko

Language

EN