I meant to say "the actual paper" instead of TFA. You can't tell anything from the article linked.

Reading the TFA, it looks like they went to some trouble to model some specific brain structures and synapse properties, including inter-area connectivity and learning, in the model. So it's not "Just a big neural net." However the accuracy of the simulation is limited--both by what we know about the detailed structure of the cat's brain and by the number and complexity of the structures they decided to model.

Unfortunately, Google's mission "Don't be evil" fundamentally conflicts with Verizon's "Be evil".

A non-standard compiler extension that is guaranteed to be supported by the vendor's compilers is pretty much the definition of vendor lock-in, even if the implementation is open-sourced. If other compiler vendors don't pick it up (and they won't with a standards-based alternative) all code that uses it becomes tied to the vendor.