The definitions of "weakly connected" were inconsistent, and some versions permitted empty graphs (which they really shouldn't, though it's hard to say that a definition is a mistake exactly -- at any rate he agreed that they shouldn't).