Few-Shot Prompting (உதாரணங்களுடன் / Showing Examples)

Let's show the carpenter some photos! (ஆசாரிகிட்ட போட்டோ காட்டுவோமா!)

Technical Meaning: சில முன்மாதிரி தூண்டல் (Sila Munmaathiri Thoondal) - Providing the AI with a few examples of the desired input-output format before asking it to solve a task.

The Core Idea

Even the smartest AI sometimes struggles to understand the exact format, tone, or logic you want. In Few-Shot Prompting, instead of just giving an instruction, you provide 2 to 5 concrete examples inside the prompt. This "in-context learning" instantly calibrates the AI's brain to match your exact expectations.

The Origin Story

Researchers noticed that LLMs often failed at specialized tasks (like classifying a tweet's sentiment as Positive/Negative/Neutral) when using zero-shot prompting. However, by simply prepending the prompt with three examples of correctly labeled tweets, the model's accuracy skyrocketed. It proved that LLMs can learn new tasks on the fly inside the prompt window.

The Tamil Analogy

Imagine you want a Carpenter (ஆசாரி) to build a very specific vintage chair for your house.

• Zero-Shot: You just say, "Build a vintage chair." He builds something, but it's not what you had in your mind.
• Few-Shot: You pull out your phone and show him 3 photos of exact chairs you like. You say, "See how the legs are curved in this photo? See the backrest in this photo? Now, build me a chair like this."

By providing a "few shots" (examples), the carpenter instantly understands the exact aesthetic and structure you want. The AI works the same way—give it examples, and it mimics the pattern flawlessly!

Try It Yourself