I agree with Weinberger's on the challenges for the Semantic Web.  His statements about it being over his head, notwithstanding, it strikes me that he's nailed much of what I said less effectively in the Semantic Web panel at Stanford Venture Lab.  Yes it does seem like the grand claims of Knowledge Representation and CYC all over.  And yes, you must focus on situations where you can add value with this kind of technology, for example, as he says

"normalization of metadata works real well in confined applications where the payoff is high, control is centralized and discipline can be enforced. In other words: not the Web."

His other example of adding metatags to every page using say Dublin Core also demonstrates one other key requirement of SemWeb application.  The cost of doing the encoding of knowledge (or simply tagging of content) has to be paid by the beneficiary.  I think he has a great theory that Google indexing the tags would create the necessary benefit to justify the pain of tagging.  This is what I felt missing in an earlier blog entry on "Dublin Core metadata: A character in search of an author?"

It's these baby steps toward the big vision, or perhaps really just more carefully selected applications, that are likely to be the payoff of the protocol work around the Semantic Web or metadata.  Just as it was with AI where there have been some tremendously useful applications of the technology even as the grand scientific challenge remains illusive.