Comment by einrealist

And there is another problem: LLMs generating too much code, code that is doing more than was asked. And that cannot be fixed by tests. Usually, we create tests for wanted behavior and expected exceptions. But we don't create tests for undesired behavior.