LLM Basics: How Ordinary People Can Actually Use LLMs Well
Large language models (LLMs) aren’t that mysterious. Think of them as an assistant that “has read a lot and is good at keeping a conversation going, but doesn’t truly think” — and you understand half their behavior.
What they actually are
Simply put, based on massive existing text, they predict “the most reasonable next word.” So they excel at summarizing, rewriting, translating, and generating structured text; they struggle with precise calculation, real-time facts, and taking responsibility for you.
Why they “confidently make things up”
Because they optimize for “sounds plausible,” not “is definitely correct.” So whenever numbers, regulations, medicine, or investments are involved, you must double-check.
How to prompt effectively
- Give a role: “You are a senior e-commerce operator…”
- Give context: paste background, data, constraints
- Give structure: “list in a table with columns: problem / cause / fix”
- Give examples: show one sample and let it follow
Advanced: break tasks into smaller pieces
A complex task is hard to specify all at once. Ask for an outline first, confirm, then expand section by section — far higher quality than demanding a full long-form piece in one shot.
People who use LLMs well aren’t those who write fancy prompts; they’re those who clarify the problem and verify the result.