I wasn't talking about poisoning LLM for generally degrading performance/improvement. But for manipulating answers for propaganda/ideological reasons.