Doesn't seem to show anything the can't already be done with deep learning. Is it just deep learning packaged to be easier to use for end users or something actually new?
The major differences are that it can move from one domain of expertise to another without changing any code, it's a unique method (not a neural network -- but we'll elaborate more on this in the next release), and perhaps most importantly, it can run on a Raspberry Pi (~12MB memory usage).