"None of that really matters though because AFAIK you can't prove a neural net has been trained "correctly" and doesn't contain errors." Progress is being made

"None of that really matters though because AFAIK you can't prove a neural net has been trained "correctly" and doesn't contain errors." Progress is being made in this area. Here is an "old" paper: https://arxiv.org/abs/1702.01135 https://arxiv.org/abs/1702.01135