AI solved an Erdős problem, but the math community shouldn't celebrate too soon

OpenAI announced that their internal AI solved Erdős Problem 1196, which had puzzled the math community for 80 years. A Stanford mathematician received an email about it while he himself had been grinding away at this problem for seven years. This blew up on Hacker News, and the math section on Reddit was discussing it too—AI can really propose arguments humans never thought of and discover hidden structures. But as someone who works with computer vision every day, my first reaction was: Is this direction good for publishing papers? Just kidding. The more serious question is: Does the 'solution' found by AI count as mathematical understanding?