Ask ChatGPT that, It will stay thinking for hours, not being able to answer.

Try it yourself... If that is a llm issue in general why this small model gets it right first shot? (check link).

Zero—“knowledge” contains no w. But knowledge-building relies on the five Ws (who, what, when, where, why) plus how. In a multi-agent, agentic AI/LLM workflow you can mirror that by assigning distributed agents to each W and running them in parallel, then reconciling their outputs, which often improves coverage and reduces errors.