Fine-Tuning vs Prompting

Fine-Tuning vs Prompting

Step 1 of 13

prompt880 tokens

Prompting steers a frozen model with instructions supplied at request time.

The Idea

Prompting steers a frozen model with instructions and examples supplied at request time. Fine-tuning continues training on your own examples so the behaviour is baked into the weights. One is instant and reversible; the other is neither.

Real-World Example

Airline cabin crew. The safety card is read out every flight and can be reprinted overnight. Recurrent training reshapes reflexes instead — weeks of work, and no memo can undo it by tomorrow.

The Tradeoff

Fine-tuning buys consistent formatting, shorter prompts and lower cost per call. It charges you dataset curation, a training run, and a model version to maintain and redo on every base-model upgrade. Prompting and retrieval are almost always the cheaper first attempt.

Your turn

Put the steps in the right order.

1. 1Fine-tune on curated examples, but only if the gap survives everything above↑↓
2. 2Write a clear prompt with a few worked examples↑↓
3. 3Add retrieval if the failures turn out to be missing facts↑↓
4. 4Measure where the output actually falls short↑↓

Check order

Mini quiz

1 / 3

Fine-tuning is best suited to:

• AAdding this week's company facts
• BLocking in a consistent style or output shape
• CEnlarging the context window
• DShrinking the vocabulary