ryan_greenblatt comments on Modifying LLM Beliefs with Synthetic Document Finetuning