evhub comments on unRLHF—Efficiently undoing LLM safeguards

[evhub]

(Moderation note: added to the Alignment Forum from LessWrong.)