I agree. Very light. All the description I've read say is that they have found a way of finding or generation collisions. So what? We always knew there were collisions.

The only "evidence", shows two pre-calculated datasets that produce the same md5. What I haven't seen (though maybe the conference gave more detail than I can find), is any technique for generating the plaintext from the md5?

If the mechanism starts with the original plaintext, to generate another plaintext with the same md5--why are they bothering? They already have the original!

Hmmm. Maybe there is some real information out there somewhere.