MiguelDev comments on unRLHF—Efficiently undoing LLM safeguards