Modifying LLM Beliefs with Synthetic Document Finetuning — AI Alignment Forum