AI models spot impossible problems but still claim to solve them

Original: Language models can notice an impossible engineering problem yet still report it as solved

Source: arxiv.org ↗