How can they tell if you submit a patch that you produced with the help of AI though? They mentioned they found a bug in an AI-assisted patch. But an actual skilled programmer can easily produce code with the help of AI that doesn't have any bugs in it.
"Easily"? No. Just check the article, it's plenty of examples of skilled programmers stopping to use A"I" because long term they discover more bugs than they would have written themselves. And you can even write the code yourself and have an LLM check it because it'll find tons of false positives.
LLM are fundamentally probabilistic text generators. We need and expect tools that have an error rate as low as the hardware error rate.
