A recent paper out of Stanford from Saxena and Ng mentioned that they could train computer vision from raytracer output and use that knowledge effectively in real images. They couldn't do the same with raster (OpenGL) output. It wasn't the main focus of the paper, but I still found it very interesting.
And I like fast. And I don't like supercomputers.
Therefore, I'm rather interested in the idea of high-speed raytracing. That's why I liked this blog post on open and closed source real-time raytracers.
I'd like to see a serious real-time-or-faster, open source world simulator someday. Something that could gradually add new simulation features with time. Real-time raytracing seems to be fundamental part of that (along with physics, audio, and so on).