LLMs respond differently to harmful prompts when AI watermarking is usedarstechnica.com2 points·joozio··0 commentsOpen articleSaveView on HN