Zero-Shot vs. Few-Shot
In AI, a "Shot" is an example. Learn when to trust the model's instinct (Zero-Shot) and when to show it the way (Few-Shot).
The "New Employee" Analogy
Zero-Shot
Like telling a seasoned expert: "Just do it."
They rely on their years of experience (pre-training) to figure it out immediately.
Few-Shot
Like onboarding a new intern: "Here are 3 examples."
You show them previous reports so they follow your specific style and format.
When to Use Which?
Zero-Shot
Fastest
No Examples
Simple tasks: "Summarize this," "Translate to Hindi," or "Brainstorm ideas".
One-Shot
Balanced
1 Example
Setting a specific tone or adhering to a simple email format.
Few-Shot
Most Accurate
2-5 Examples
Complex tasks: Technical reports, medical coding, or specific legal summaries.
Zero-Shot (Simple)
User: "Translate this into Marathi: 'We will meet at the office at 10 AM tomorrow.'"
AI: "आम्ही उद्या सकाळी १० वाजता ऑफिसमध्ये भेटू."
AI: "आम्ही उद्या सकाळी १० वाजता ऑफिसमध्ये भेटू."
Few-Shot (Saree Shop)
Classify these customer reviews:
1. 'Color fading after one wash.' -> Quality Issue
2. 'Delivery took 10 days.' -> Logistics Issue
User: "3. 'The fabric is very rough.'"1. 'Color fading after one wash.' -> Quality Issue
2. 'Delivery took 10 days.' -> Logistics Issue
AI: " -> Quality Issue"
Interactive Lab: The "Pattern" Challenge