Comment by echelon

LLM reports misinformation --> Bug report --> Ablate.

Next pretrain iteration gets sanitized.