No disagreement there. I certainly wasn't trying to dismiss the ability to come up with novel solutions to problems never encountered as an indicator for programming ability and general intelligence completely. However, I certainly don't believe it's a good enough indicator to use as the sole basis for simply rejecting otherwise perfectly capable candidates before you've even talked to them.
I'm also of the opinion that interviews should strive to assess how well candidates can perform on the specific position you're hiring for (that is the main purpose of interviews, is it not?). In my own case (I was applying for a front-end engineer), and in the case of most other engineering positions, I don't believe puzzle solving is a very useful indicator for that purpose, because an overwhelming majority of engineering tasks involve applying the scientific method and engineering knowledge to solve variations of well-known engineering problems in their specific fields.
EDIT: Another point that I forgot to mention: If you are in fact interested in a candidate's ability to solve novel problems they've never encountered, at least do a better job of ensuring the problem you give them is actually novel. I found the StackOverflow thread for the problem they gave me in about 30 seconds of Googling. By not ensuring the problem novel enough that searching for it won't do you any good, you're going to be disproportionately passing dishonest candidates who would simply cheat to get through this part of your process. Certainly that can't be the result you're looking for.