Unified Framework for Evaluating LLM Inquiry and Causal Discovery

A new framework evaluates LLM inquiry capabilities by extending the Countdown task into a mathematical environment. The setup measures how models integrate hypothesis generation, reasoning, and variable manipulation to uncover relationships within complex systems.