I expect it would require significantly more human labor to train (ie no longer fully unsupervised). I imagine that this constraint would lead to significant ad

I expect it would require significantly more human labor to train (ie no longer fully unsupervised). I imagine that this constraint would lead to significant additional research to improve the efficiency of the training process, and that novel approaches would be developed. In other words I think it would suck up a lot of money over a few years and then we would arrive back pretty much where we are now.