Few-Shot Prompting — Animate | AI Wisdom

← Back to topic
🎞 ANIMATE · Mode 3 of 5 · +100 XP— Few-Shot Prompting
📚Examples
🔍Pattern
❓Query
✅Classify
🎯 Next: Test →
Step 1 / 4
Speed
Provide labeled examples
3 input→output examples teach the model the task💡 Analogy
Like showing a new employee three completed forms before asking them to fill the next one.
We show the model three examples of sentiment classification before asking it to classify a new input.
Console Output
> Examples loaded: 3
> Task: sentiment classification
> Format: Input → LABEL
●few-shot.ts
Provide labeled examplesL4–6
1const fewShotPrompt = `Classify sentiment as POSITIVE, NEGATIVE, or NEUTRAL.
2 
3Examples:
4Input: "I love this product!" → POSITIVE
5Input: "Terrible service, never coming back." → NEGATIVE
6Input: "The package arrived on Tuesday." → NEUTRAL
7 
8Now classify:
9Input: "The battery life exceeded my expectations." → `;
10 
11const response = await llm.complete(fewShotPrompt);
12// Output: "POSITIVE"
13 
14// Few-shot with structured output:
15const examples = [
16  { input: "fast shipping", label: "POSITIVE" },
17  { input: "broken on arrival", label: "NEGATIVE" },
18];
18 lines
Timeline — 4 Steps
Provide labeled examples
3 input→output examples teach the model the task💡 Analogy
Like showing a new employee three completed forms before asking them to fill the next one.
We show the model three examples of sentiment classification before asking it to classify a new input.
Console Output
> Examples loaded: 3
> Task: sentiment classification
> Format: Input → LABEL

Was this page helpful?

Discussion

3 input→output examples teach the model the task

💡 Analogy

Like showing a new employee three completed forms before asking them to fill the next one.

We show the model three examples of sentiment classification before asking it to classify a new input.

Console Output

> Examples loaded: 3
> Task: sentiment classification
> Format: Input → LABEL

3 input→output examples teach the model the task

💡 Analogy

Like showing a new employee three completed forms before asking them to fill the next one.

We show the model three examples of sentiment classification before asking it to classify a new input.

Console Output

> Examples loaded: 3
> Task: sentiment classification
> Format: Input → LABEL