An experiment evaluating the reliability of AI-generated code revealed that language models are fundamentally incapable of doubting their own solutions, even when producing blatant bugs.
