I haven't been especially clear. I said "could be valuable" - I might also have said "could be hard to replicate." Nets are hard to create because they need mas

I haven't been especially clear. I said "could be valuable" - I might also have said "could be hard to replicate." Nets are hard to create because they need masses of data to train as the article says. Paradoxically, at the same time they're just about impossible to effectively protect as IP; functional duplication of a net is (nearly) always possible if you're willing to put up the money. So why not pay less to rent the software now and wait 'till recreating the nets is far cheaper or already largely available as open source? I suspect the old-line car companies have tossed away a ton of money because they had zero understanding of neural net commerce or economics.