Ontologies, structured representations of knowledge, are crucial for various AI applications, but building them is a laborious process. Could large language models (LLMs) automate this? New research explores using LLMs to learn ontologies from the ground up. Researchers have developed a method called OLLM that goes beyond simply extracting individual facts. Instead, OLLM trains an LLM to model entire interconnected chunks of an ontology, capturing the relationships between concepts more effectively. This approach uses a clever trick: a custom regularizer that prevents the model from overfitting on common concepts, ensuring it learns a broader, more generalized understanding of the knowledge domain. To evaluate the quality of the generated ontologies, the researchers also introduced innovative metrics that go beyond simple text matching. These metrics leverage deep learning techniques to compare the semantic and structural similarity between the generated ontology and the ground truth, providing a more nuanced assessment. Experiments with Wikipedia data show that OLLM outperforms traditional methods, creating ontologies that are more semantically accurate and structurally sound. Impressively, the model can also be adapted to new domains, like the arXiv scientific paper repository, with minimal additional training. This suggests that OLLM could be a powerful tool for automatically building ontologies across various fields, potentially revolutionizing how we organize and access knowledge. While the research primarily focuses on simpler ontologies with taxonomic (is-a) relationships, it lays the groundwork for future explorations into more complex ontological structures. The ability to learn and adapt to different knowledge domains also opens doors to exciting possibilities for applications in knowledge management, information retrieval, and other AI-driven fields.