AI Models Confidently Produce Wrong Answers to Simple Questions

A test of 17 open-source LLMs reveals a critical vulnerability: models confidently produce incorrect answers to elementary questions, with no structural difference between right and wrong responses. The danger lies not in occasional hallucinations but in deterministic failures masked by the same authoritative tone as correct answers. This fundamentally undermines the reliability of any single-model AI system handling factual information.