Imagine having a crystal ball that could predict which startups will become the next unicorns. While true clairvoyance remains elusive, researchers are getting closer with a fascinating new approach: using AI-powered decision trees. Traditional decision trees are great for explainability—you can easily trace the logic behind their choices—but they struggle with the complexities of real-world data, especially when dealing with nuanced information like text. Large language models (LLMs), on the other hand, excel at understanding text but often lack transparency. Enter GPTree, a clever framework that combines the best of both worlds. It uses the reasoning power of LLMs to dynamically generate questions and build a decision tree, allowing it to analyze complex startup data, including textual descriptions, while retaining the explainability of traditional decision trees. Think of it as an LLM that thinks in a structured, tree-like way. This approach eliminates the need for complicated feature engineering or prompt chaining, making the process more efficient. Interestingly, GPTree also incorporates human expertise through a feedback loop, allowing experts to refine the AI’s decision paths. In a test case focused on predicting "unicorn" startups—those reaching a valuation of $500 million or more—GPTree achieved a remarkable 7.8% precision rate, outperforming not only other AI models but even seasoned venture capitalists, whose success rates range from 3.1% to 5.6%. While GPTree shows incredible promise, challenges remain. The reliability of the code generated by the LLM and the inherent non-deterministic nature of LLMs can impact results. Further research is focused on addressing these limitations and expanding GPTree's capabilities to other areas, like healthcare diagnosis. This exciting development hints at a future where AI can provide not just predictions, but also clear explanations, enabling humans to make more informed and confident decisions.