“The large model falls into a narrative trap during multiple rounds of dialogue”

Research has found that large language models are prone to falling into self-generated narrative loops during multiple rounds of dialogue, resulting in logical inconsistencies or repetitive content, which affects the accuracy and reliability of the output.