"No" when he's talking about the Linux kernel specifically. For example, the headers distributed by Android, which are basically verbatim copies of some of the Linux headers but with a different license attached. RMS says this is okay.
"Yes" when he's talking about other GPLd code. If you write an application which uses a GPLd library, and distributed only the object code of your own application, expecting the user to link in the GPLd library (either implicitly or explicitly), then somehow the mere use of the API plus distribution of only your own object code causes infringement. RMS says this is not okay. Specifically that it is direct infringement as a derived work, as opposed to constituting vicarious or contributory infringement.
Go figure. FWIW, I'm not a GPL hater so please don't flame me for supposedly spreading FUD on Free Software. I'm just pointing out an inconsistency with RMS'--and many others'--analysis without taking a stance either way.