Before you spend a week fine-tuning a model, it is worth asking whether a better prompt would get you there faster and cheaper. Usually it would.
Start with prompting
Clear instructions, examples, and structured output handle the large majority of tasks with zero training cost and instant iteration.
Fine-tune when
- You need a consistent style or format at scale.
- Prompts are getting long and expensive.
- You have quality labeled data and a stable task.
Think of prompting as the default and fine-tuning as an optimization you reach for once you have proven the task is worth it.