A recent position paper by Google Deepmind's Tom Zahavy suggests that language models are not capable of sparking scientific revolutions. According to Zahavy, these models lack the cognitive mechanisms necessary to create novel ideas. This limitation may be due to their focus on generating text within existing frameworks, rather than introducing entirely new concepts. In contrast, world models, which are designed to simulate the world and make predictions, may have more potential for driving scientific breakthroughs. The distinction highlights the different capabilities and limitations of various AI models.