What's your metric for the "percentage the code is wrong"? Is it how many lines of code were wrong, or how many test cases the code fails? Presumably if AI-gen

What's your metric for the "percentage the code is wrong"? Is it how many lines of code were wrong, or how many test cases the code fails? Presumably if AI-generated code passes every test case, but would fail on edge cases that some human programmer(s) did not anticipate in their suite of tests, the humans potentially might have made similar coding mistakes as the AI if they had had to personally write the code.