The massive datasets needed to train large language models (LLMs) are becoming increasingly difficult to acquire. Is there a way for AI to bootstrap itself and generate its own training data? New research suggests a surprising solution lies not in more data, but in smarter reasoning. The problem of dwindling training data has AI researchers scrambling for new sources of information to feed the insatiable hunger of LLMs. One intriguing approach is to have AI generate its own training data, effectively pulling itself up by its bootstraps. This new research proposes a method using "strong relevant logics" to automatically generate "empirical theorems." Essentially, this involves using a type of logical reasoning to deduce new facts from existing knowledge. Imagine a system that understands basic mathematical principles. Using strong relevant logics, it could deduce new mathematical relationships, effectively creating new theorems that expand its knowledge base. These new theorems can then be used as fresh training data for the LLM. The advantage of this approach is that the generated data is guaranteed to be logically sound and free from the “hallucinations” that plague LLMs trained on less structured data. The research introduces a tool called "FreeEnCal," a forward reasoning engine capable of performing this kind of logical deduction. FreeEnCal can be customized to work with different logical systems and datasets, making it a versatile tool for generating new knowledge in various domains, from mathematics to cybersecurity. While promising, this research is still in its early stages. One of the key challenges lies in translating real-world knowledge into the formal language of logic and then back into natural language for LLM training. However, the potential of this approach is enormous. If AI can truly generate its own training data, it could unlock a new era of self-improving AI systems, capable of continuously expanding their knowledge and abilities without relying on ever-larger datasets from the human world. This research offers a tantalizing glimpse into a future where AI not only learns from the data we provide, but also actively creates its own, forging a path towards more robust and capable intelligent systems.