Amazing that when ai "solves" an erdos problem by finding the solution in an existing paper it gets hundreds of points and comments, but when ai fails a more rigorous test designed by practitioners (i.e. a much better test of the claim that ai will soon do research level mathematics) it gets zip.