I think there are (at least) two very different kinds of talks: those that are meant to teach you worthwhile things, and those that are meant to inspire you to invent worthwhile things. All the talks I've seen from him are in the latter category.
His talks constantly feature working demos of the ideas he is pushing, subtly demonstrating a lot of well-thought-out interaction design details. If you watch his "Media for Thinking the Unthinkable" (http://vimeo.com/67076984) it's a gold mine of specifics. I've watched it several times and always pick some new ideas for my UI design work.
The difference to a run-of-the-mill talk is that he is showing the details, not telling the details.
I have the exact opposite reaction to the video. He is solving toy problems with toy ideas. I think his page Kill Math (http://worrydream.com/KillMath/) illuminates this point. I don't think he can think symbolically very well (no insult intended, I can't think visually very well). There are certainly times where graphing things make a lot of sense, but to throw out analytical math? Come on. By and large he is getting the "feel" of a system, but he cannot really reason about it, prove things about it, extend it, or design new systems with vision (there are obvious counterexamples).
In another video he shows an IDE where he scrubs constants, and it changes the behavior of the concurrently running program (changing the size of an ellipse or tree branch). It's neat. But, again, toy problem. First of all, we shouldn't be programming with constants. Second, anything complicated will have relationships between the data - scrubbing one value will just end up giving you nonsense. Third, it just doesn't make any sense in many contexts. I work in computer vision currently, and I can't think of anything but the most superficial way I could incorporate scrubbing. He made some comment about how no one could know what a bezier curve is unless they had a nice little picture of it in their IDE to match the function call. That's silly. I actually use splines and other curve fitting in my work, and I have to actually understand the math. Do I use cubic splines, a Hermite interpolation, bezier, or something else? I don't decide that by drawing some pictures - the search space is too big, I'll never cover all the possibilities. I have to do math to figure out the best choice.
In that same video he went on to demonstrate programming binary search using visual techniques. Unfortunately he wrote a buggy implementation, and stood there exclaiming how his visual technique found a different bug. It did, a super trivial one, but it completely failed to reveal the deeper issue. And, there was no real way for his visual method to have found it.
Visualization is an very powerful tool, but it is one tool in the toolchest. There is a scene in the movie Contact with Jodie Foster using headphones to listen to the SETI signal. We all know that is bogus - the search space is far too vast for aural search to work.
His ideas are terribly wrong headed. Make interfaces to help give us intuition? Absolutely! Use graphics where analytics fail. Of course! But don't conclude that math is a "freakish knack", as he does, or that math is some sort of temple (he calls mathematicians "clergy", and then goes on to throw in an insult that many are just pretending to understand).
I posted in another comment how crazy it would be to have a calculator that scrubs. Well, he shows one on that page. Really? The day bridge designers start using scrubbing apps to design our bridges is the day I'm never crossing a bridge again.
Edit to add: his website is another example of this. I can't find anything on it. There are a bunch of pictures, and my eyes saccade around, but what is here, what is his point? I dunno. I can click, and click, and click, and start to get an idea, but there is always more hidden away behind pictures. It's barely workable as a personal website, and would be a disaster as a way to organize anything larger. I don't mean to pick on it - as an art project or glimpse into how he thinks, it's great. I just point out it illustrates (pun kind of intended) the strengths and limits of visual presentation. You tell me, for example, without grep or google search, whether he has written about coffee.
If you disagree, please reply in pictures only! ;)
