That reminds me of a question I have since I saw my first LLM hallucination: How much do people think hallucination/confabulation can be attributed to trolling

That reminds me of a question I have since I saw my first LLM hallucination: How much do people think hallucination/confabulation can be attributed to trolling and sarcasm having slipped into the training data? Is it possible we could get the rate of hallucinations down by better filtering of cynicism from the traing data?