Large language models (LLMs) have impressive knowledge stores, but can they actively learn by asking questions like humans do? A new study explores this by prompting LLMs to generate questions about scientific statements, as if encountering them for the first time. This "curiosity-driven question generation" reveals fascinating insights into how LLMs approach learning and knowledge acquisition. Researchers tested various models, from giants like GPT-4 to smaller, specialized models like Phi-2, across physics, math, chemistry, and general knowledge statements, even including intentionally incorrect statements to gauge critical thinking. The results challenge the assumption that bigger models are always better. While GPT-4 generally excelled, Phi-2 often performed comparably, demonstrating that careful design and training data can significantly impact a model's ability to learn through inquiry. The study also highlighted how LLMs react differently to false information, with some effectively questioning the validity of incorrect statements. This research opens up exciting possibilities for creating AI systems that not only answer questions but also actively seek knowledge through questioning, pushing the boundaries of AI learning and discovery.