That’s at least the way they put it. As I unfortunately have not yet been able to make it among the lucky ones, who get the first bunches of early-adopter tickets for Intel’s MashMaker, I have to stick with the documentation when it comes to figuring out the details:
In order to bring mash-up creation to "the rest of us", the service provides you a toolbar for Firefox 2+ (with other versions still to come) which brings pre-programmed access to service APIs like Google Maps or Yahoo! Search with it. Whenever you then visit a webpage and choose a mash-up kind from the toolbar’s menu or simply push one of the buttons for the more popular services, the program trys to relate and mash-up the (optionally selected) webpage content with popular web services of choice.
And while the toolbar software seems to make use of semantic web content extraction more in the sense search engines do it, obviously users are enabled to share with others if they are being happy with the automated processing results and especially if they could successful use the mashup they created. So you can annotate and refine the results lateron and in return the MashMaker server will lern about web pages’ content and start soon proposing suitable mash-ups by itself.
While it currently still looks a bit like messing up DabbleDB and de.icio.us, it may indeed have a real semanti RDF based backend, making sure, it won’t mess with its databases either (which has not yet been confirmed) . Nevertheless MashMaker finally seems to be a really powerful new tool for collaborative webpage annotation.