I empathize with LeCunn but he is a prime example of why scientists often make poor leaders. It’s too tempting to always do the cool and interesting thing instead of drilling down on the boring but reliable thing that actually creates value. There is of course a place for this kind of eternal curiosity and I hope he’s found it, but hyper-competitive industry may not be it.

I'm becoming more sympathetic to LeCun's views, but being able to train something that acts the tiniest bit like those who made the training data (authors, professors, researchers, etc) is an absolute juggernaut and will be for some time.