Tag
The paper applies repetition priming to show that base LLMs use automatic processing while instruct models exhibit controlled processing, with humans displaying a hybrid profile.