It is neither accurate to say that LLMs “think like humans” nor that they merely “repeat patterns.” Instead, they occupy a new category: systems that derive generalizable internal models from vast amounts of data and use those models to plan and generate text.