AI scores a ‘C–’ on its hardest math test yet

The second batch of “First Proof” problems is meant to evaluate AI’s usefulness for research-level math. The best model got six or seven of the 10 questions basically right