A new benchmark reveals most AI models confidently answer nonsensical questions, highlighting significant flaws in their reasoning and a need for improved AI safety.
