Sorry but these are just not accurate as blanket statements anymore, given how good the models have gotten.

As other similar projects have pointed out, if you have a good test suite and a way for the model to validate its correctness, you can get very good results. And you can continue to iterate, optimize, code review, etc.