This article discusses the distinction between two types of AI in science: language models (scientists) and domain-specific models (simulators). It argues that while language models excel in processing and generating knowledge, real scientific progress, especially in complex fields like biology, requires simulators that can learn directly from data and predict physical phenomena.