The “Woodpecker” hallucination correction system can, ostensibly, be applied to any multi-modal large language model, according to the research.