The GPTree decision tree has demonstrated superior performance in identifying "unicorn" startups at their inception stage, achieving a 7.8% precision rate. This surpasses the gpt-4o with few-shot learning and even outperforms the best human decision-makers, who have a precision rate of 3.1% to 5.6%.
Key takeaways:
- The paper presents GPTree, a new framework that combines the explainability of decision trees with the advanced reasoning capabilities of LLMs.
- GPTree eliminates the need for feature engineering and prompt chaining, requiring only a task-specific prompt and leveraging a tree-based structure to dynamically split samples.
- The authors introduce an expert-in-the-loop feedback mechanism to enhance performance by enabling human intervention to refine and rebuild decision paths.
- The decision tree achieved a 7.8% precision rate for identifying 'unicorn' startups at the inception stage, surpassing gpt-4o with few-shot learning as well as the best human decision-makers (3.1% to 5.6%).