Mid-20th century problems posed by Paul Erdős are now being solved by AI. These specific challenges provide a benchmark for machine reasoning. Mathematicians are analyzing these successes to determine if AI can tackle broader, unstructured proofs. This progress suggests a shift from simple calculation to complex logical discovery for LLMs.