การป้อนพรอมป์ต์แบบ Zero-Shot และ Few-Shot

With LLMs, your main control is the words you send. The art of shaping those words is called prompting. ✍️

In zero-shot prompting you just describe the task in plain words, with no examples, and trust the model to figure it out.

Zero-shot is great for common tasks the model has seen often, like summarizing a paragraph or answering a simple question.

When the task is unusual, examples help. Giving the model sample inputs and outputs is called few-shot prompting.

You list a couple of worked examples, then leave the final input blank so the model continues the same pattern.

Few-shot does more than teach the task; it also fixes the output format, so the model answers in exactly the shape you showed.

Choose examples that are correct, varied, and unambiguous. Sloppy or biased examples quietly teach the model the wrong pattern.

Each example eats into the context window and adds cost, so use the fewest examples that reliably get the job done.

Vague prompts give vague answers. Spell out the role, the task, the format, and any constraints you care about.

For tricky reasoning, ask the model to work it out step by step. This chain-of-thought hint often improves the final answer.

Prompting is empirical. Try a version, check the outputs, then refine the wording until results are consistent.

What is the key difference between zero-shot and few-shot prompting?

Start zero-shot for common tasks. Add a few clear examples when you need a specific pattern or format, then iterate. ✅