Science journalism is so frustrating! Here's the version for smart people:
http://web.media.mit.edu/~bates/Summary_files/BatesTalk.pdf
In summary: Low-precision high-dynamic range arithmetic (floating point with small mantissa, ~1% error) uses ~100x less power and die area than IEEE floating point. The errors are acceptable for a huge class of applications (basically anything you'd consider running on a GPU today).