The second batch of “First Proof” problems is meant to evaluate AI’s usefulness for research-level math. The best model got ...