Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning | Read Paper on Bytez