#pre-training-errors

#pre-training-errors

LLM hallucinations largely result from pretraining exposure and evaluation metrics that reward guessing; penalizing confident errors and rewarding uncertainty can reduce hallucinations.