Skip to lesson

Path: How post-training works

Demonstrations and SFT

See how carefully written examples teach a model the desired form of an answer.

Starting level
New learner
Time
18 minutes
Public lesson guide
EN + हिंदी

You will learn to

Recognise what makes a demonstration useful for supervised fine-tuning.

Why this matters

A model learns the requested behaviour from examples only when those examples follow the instruction clearly.

Three things to remember

  • Follow the requested format exactly.
  • Include required safety and content constraints.
  • Avoid accidental patterns the model should not imitate.

See one example

Task

Give exactly three safe water-storage steps.

Weak approach

A long paragraph with five mixed ideas.

Stronger approach

Three numbered practical steps plus one brief contamination warning.

Why the stronger approach works

The strong example teaches both the requested structure and the safety constraint.

Make something yourself

Write one demonstration for a tightly constrained instruction, then list the signals it teaches.

A useful starting pointCheck count, order, required content, and prohibited content.
Practise this lesson

Read further

The source is optional. The complete teaching and practice are available inside AI2Bharat.

Read the primary or official source: OpenAI InstructGPT