The Biggest “Lie” in AI? LLM doesn’t think step-by-step