@alice The few times I’ve played with LLM’s, I always tell it that the code doesn’t work, the specific functions don’t exist, then ask it why it confidently lies. I have no issues poisoning training data that was stolen from the sum total of written human experience.