The article mentions that it will display adds based on the content of the discussion. If this actually works at all I will be impressed. It wasn't long ago that voice recognition took hours to train to a single voice and then it still worked like ass. I seem to remember a thread on either UPL or old cal's where everyone used some recognition software and you couldn't get a cognizant sentence out of it.
Now, its still pretty shitty to use a skype ripoff that throws adds everywhere and databases all your calls. Obviously there is potential for abuse, but at what cost....