Decoder. plain-English AI glossary

Fine-tune vs Prompt

● Core

The recurring choice: teach the model new behavior by retraining it, or just by asking better in the prompt.

Think of it like

Sending an employee to a training course versus writing them a really clear instruction note.

Example

A team wants consistent JSON output; they try a detailed prompt first, and only fine-tune once prompting hits its ceiling.

How it actually works

Prompting is fast, free to change, and reversible, but every request pays the token cost and behavior can drift. Fine-tuning bakes the behavior in — cheaper at inference, more consistent — but it’s slow to iterate and risks forgetting other skills. The rule of thumb: exhaust prompting first, fine-tune when you need consistency, lower latency, or behavior prompting can’t reach.

For product teams

Prompting to validate an idea quickly; fine-tuning once the behavior is proven and volume justifies the cost.

For engineers

A trade between updating weights (durable, costly to iterate) and steering via context (flexible, per-request cost).

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome