Mid-20th century problems posed by Paul Erdős are now falling to AI. These specific mathematical challenges provide a unique testing ground for machine reasoning. Researchers are analyzing these successes to determine if AI can tackle broader, non-structured proofs. This progress suggests a shift in how LLMs handle formal logic and complex discovery.